JP4345129B2 - Document processing method and apparatus, and recording medium - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a document processing method and apparatus for processing an electronic document, and a recording medium on which a document processing program for processing the electronic document is recorded.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, on the Internet, WWW (World Wide Web) is provided as an application service that provides hypertext type information in a window format.
[0003]
The WWW is a system that performs document processing for creating, publishing, or sharing a document, and shows a new style of document. However, from the viewpoint of practical use of documents, advanced document processing exceeding WWW, such as document classification and summarization based on document contents, is required. For such advanced document processing, mechanical processing of document contents is indispensable.
[0004]
However, mechanical processing of document contents is still difficult for the following reasons. First, HTML (Hyper Text Markup Language), which is a language for describing hypertext, defines the expression of a document but hardly specifies the contents of the document. Second, a hypertext network formed between documents is not always easy to use for the reader of the document to understand the content of the document. Third, in general, the author of the text writes without regard to the convenience of the reader, but the convenience of the reader of the document is not coordinated with the convenience of the author.
[0005]
As described above, the WWW is a system that indicates a new document. However, since the document is not mechanically processed, advanced document processing cannot be performed. In other words, in order to perform advanced document processing, it is necessary to mechanically process the document.
[0006]
Therefore, with the goal of mechanical processing of documents, a system that supports mechanical processing of documents has been developed based on the results of natural language research. As document processing by natural language research, mechanical document processing that uses the tag attached to the document on the premise of the attribute information on the internal structure of the document by the author of the document, so-called tag assignment, has been proposed. Yes.
[0007]
[Problems to be solved by the invention]
By the way, with the spread of computers in recent years and the progress of networking, it is necessary to increase the functionality of document processing that creates, labels, and changes text documents with text processing and indexing depending on the contents of the document. It has been. For example, document summarization or document classification according to the user's request is desired.
[0008]
The present invention is proposed in view of the above-described circumstances, and is a document processing method and apparatus for calculating a user's interest in a document, and a document processing program for calculating a user's interest in a document. Relates to a recording medium on which is recorded.
[0009]
[Means for Solving the Problems]
  In order to solve the above problems, a document processing method according to the present invention processes a plurality of electronic documents.Document processing equipmentIn the document processing method,A receiving unit receives a plurality of electronic documents, a recording unit records a plurality of electronic documents received in the receiving step, and a display unit is recorded in the recording step. A display step of displaying an electronic document selected by the user from among the plurality of electronic documents and a summary of the electronic document, and an input step of inputting user operation information for the electronic document displayed in the display step And the actual interest level detection means based on the user operation information input in the input step with respect to the electronic document displayed in the display step.For electronic documentsUser'sActual interestcalculateActual interest level detection process;The priority order setting means has a relevance level based on the internal structure of the electronic document among the electronic documents for which the actual interest level is calculated in the actual interest level detection step with respect to the electronic document for which the actual interest level is not calculated. Based on the predicted interest level, the actual interest level of the highest electronic document is defined as the predicted interest level.A priority setting step for setting priorities;A rearrangement step in which the rearrangement unit rearranges electronic documents not selected by the user among the plurality of electronic documents recorded in the recording step according to the priority set in the priority order setting step; In the actual interest level detection step, for each display area for displaying the electronic document and the summary of the electronic document, the appearance position in the electronic document among the elements of the electronic document selected by the user Is the first actual interest level element that is the ratio of the appearance position of the element farthest from the top of the electronic document and the size of the electronic document, the number of keywords specified by the user, and the element selected by the user A second actual interest level element of the actual interest level consisting of a number, a size of a display area of the summary of the electronic document displayed in the display step, and the electronic document It calculates the actual interest with a third of the actual interest element consisting of the ratio of the size of the display region.
[0010]
  A document processing apparatus according to the present invention processes a plurality of electronic documents.Document processing deviceInReceiving means for receiving a plurality of electronic documents; recording means for recording a plurality of electronic documents received by the receiving means; an electronic document selected by a user among the plurality of electronic documents recorded by the recording means; and Display means for displaying a summary of the electronic document, input means for inputting user operation information for the electronic document displayed by the display means, and input by the input means for the electronic document displayed by the display means Based on the user's operation informationFor electronic documentsUser'sActual interestcalculateReal interest level detection means;The actual interest of the electronic document having the highest relevance based on the internal structure of the electronic document among the electronic documents of which the actual interest level is calculated by the actual interest level detection unit with respect to the electronic document for which the actual interest level is not calculated Is the predicted interest level, based on the predicted interest levelPriority setting means for setting the priority;Rearrangement means for rearranging electronic documents not selected by the user among the plurality of electronic documents recorded by the recording means according to the priority order set by the priority order setting means, and the actual interest level For each display area for displaying the electronic document and the summary of the electronic document, the detection means has the most appearance position in the electronic document from the top of the electronic document among the elements of the electronic document selected by the user. A first actual interest level element composed of a ratio between an appearance position of an element at a distant position and the size of the electronic document, and the actual interest level composed of the number of keywords specified by the user and the number of elements selected by the user. The second actual interest level element, the size of the summary display area of the electronic document displayed by the display means, and the display area of the electronic document It calculates the actual interest with a third of the actual interest element consisting of the ratio of the size.
[0011]
  The recording medium according to the present invention processes a plurality of electronic documents.Let your computer perform document processingDocument processing program was recordedComputer readableIn the recording medium, the document processing program isA receiving step for receiving a plurality of electronic documents, a recording step for recording the plurality of electronic documents received in the receiving step, and a plurality of electronic documents in which the display means is recorded in the recording step. A display step of displaying an electronic document selected by the user and a summary of the electronic document, an input step of inputting user operation information for the electronic document displayed in the display step, and an actual interest The degree detection means is based on the user operation information input in the input step with respect to the electronic document displayed in the display step.For electronic documentsUser'sActual interestcalculateActual interest level detection process;The priority order setting means has a relevance level based on the internal structure of the electronic document among the electronic documents for which the actual interest level is calculated in the actual interest level detection step with respect to the electronic document for which the actual interest level is not calculated. The priority level setting step of setting the actual interest level of the highest electronic document as the predicted interest level and setting the priority order based on the predicted interest level, and the rearranging means include a plurality of electronic documents recorded in the recording step. The computer executes a rearrangement step of rearranging electronic documents not selected by the user according to the priority order set in the priority order setting step. In the actual interest level detection step, the electronic document and the electronic document For each display area for displaying the summary of the electronic document, the appearance position in the electronic document among the elements of the electronic document selected by the user is from the top of the electronic document. A first actual interest level element composed of a ratio of an appearance position of an element at a far position to the size of the electronic document, and the actual interest level composed of the number of keywords specified by the user and the number of elements selected by the user And the third actual interest level element comprising a ratio of the size of the display area of the summary of the electronic document displayed in the display step and the size of the display area of the electronic document. Calculate the actual interest level usingIs.
[0012]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of a document processing method and apparatus and a recording medium according to the present invention will be described with reference to the drawings.
[0013]
As shown in FIG. 1, a document processing apparatus according to an embodiment of the present invention includes a main body 10 including a control unit 11 and an interface 12, an input unit 20 that receives input from a user and sends the input to the main body 10, and an external device. A receiving unit 21 that receives and transmits the signal to the main body 10, a display unit 30 that displays the output from the main body 10, and a recording / reproducing unit 31 that records / reproduces information on the recording medium 32. Yes.
[0014]
The main body 10 has a control unit 11 and an interface 12 and constitutes a main part of the document processing apparatus. The control unit 11 includes a CPU 13 that executes processing in the document processing apparatus, a RAM 14 that is a volatile memory, and a ROM 15 that is a nonvolatile memory. For example, according to the procedure recorded in the ROM 15, the CPU 13 temporarily stores data in the RAM 14 when necessary, and performs control for executing the program. The interface 12 is connected to the control unit 11, the input unit 20, the reception unit 21, the display unit 30, and the recording / playback unit 31. Under the control of the control unit 11, the interface 12 transmits data for data input from the input unit 20 and the reception unit 21, data transmission to the display unit 30, and data transmission / reception to the recording / playback unit 31. Adjust timing and convert data format.
[0015]
The input unit 20 is a part that receives user input to the document processing apparatus. The input unit 20 is configured by a keyboard or a mouse, for example. The user can use the input unit 20 to input a keyword using a keyboard, or to select and input an element of an electronic document displayed on the display unit 30 using a mouse. Hereinafter, an electronic document is simply referred to as a document. Here, an element is an element that constitutes a document, and includes, for example, a document, a sentence, and a word.
[0016]
The receiving unit 21 is a part that receives a signal transmitted to the document processing apparatus from the outside via, for example, a communication line. The receiving unit 21 receives a plurality of documents transmitted from the outside. The receiving unit 21 sends the received data to the main body 10.
[0017]
The display unit 30 displays the output of characters and image information from the document processing apparatus. The display unit 30 is composed of, for example, a cathode ray tube (CRT) or a liquid crystal display (LCD), and displays, for example, one or more windows, and displays characters, figures, etc. on the windows. To do.
[0018]
The recording / reproducing unit 31 records / reproduces data on / from a recording medium 32 such as a so-called floppy disk. The recording medium 32 records a document processing program for processing a document. The recording medium 32 will be further described later.
[0019]
Next, the document in the present embodiment will be described. In the present embodiment, document processing is performed with reference to a tag that is attribute information assigned to a document. The tags used in the present embodiment include a syntactic tag indicating the structure of the document, and a semantic word that enables understanding of the mechanical contents of the document between multiple languages. There is a logical tag.
[0020]
Some syntactic tags describe the internal structure of a document. As shown in FIG. 2, the internal structure by tagging is configured such that each element such as a document, a sentence, and a vocabulary element is linked by a normal link and a reference / referenced link. In the figure, a white circle “◯” indicates an element, and the lowest white circle is a vocabulary element corresponding to the word at the lowest level in the document. A solid line is a normal link indicating a connection between elements such as a document, a sentence, and a vocabulary element. A broken line is a reference link indicating a dependency relationship by reference / reference. The internal structure of the document is from top to bottom, from document, subdivision, paragraph, sentence, subsentential segment, ..., vocabulary elements Composed. Of these, subdivision and paragraph are optional.
[0021]
On the other hand, as semantic / pragmatic tagging, there is one in which information such as meaning is described like the meaning of a multiple meaning word. Tagging in the present embodiment is based on an XML (Extended Markup Language) format similar to HTML (Hyper Text Markup Language).
[0022]
An example of tagging is shown below, but tagging of documents is not limited to this method. Moreover, although the example of an English and Japanese document is shown below, the description of the internal structure by tagging can be applied to other languages similarly.
[0023]
For example, the sentence “Time flies like an arrow.” Can be tagged as follows.
[0024]
<Sentence> <noun phrase meaning = “time0”> time </ noun phrase>
<Verb phrase> <Verb meaning = “fly1”> flies </ verb>
<Adjective verb phrase> <adjective verb meaning = like0> like </ adject verb> <noun phrase> an <noun meaning = “arrow0”> arrow </ noun> </ noun phrase>
</ Adjective verb phrase> </ Verb phrase>. </ Sentence>
Where <sentence>, <noun>, <noun phrase>, <verb>, <verb phrase>, <adjective verb>, and <adjective verb phrase> are sentence, noun, noun phrase, verb, verb phrase, and adjective, respectively. It represents a syntactic structure of a sentence such as a prepositional phrase or a postpositional phrase / adjective phrase, an adjective phrase / adjective verb phrase. The tags are arranged correspondingly immediately before the end of the element and immediately after the end. A tag placed immediately after the end of the element indicates the end of the element by the symbol “/”. Elements represent syntactic constructs, ie phrases, clauses and sentences. Note that word sense = “time 0” indicates a plurality of meanings of the word “time”, that is, the 0th meaning among the plurality of meanings. Specifically, the word “time” has at least a noun, an adjective, and a verb meaning, but here the word “time” indicates a noun. Similarly, the word “orange” has at least a plant name, color, and fruit meaning, which can also be distinguished by meaning.
[0025]
The document according to the present embodiment can display the syntactic structure in the window 101 of the display unit 30 as shown in FIG. In this window 101, vocabulary elements are displayed on the right half surface 103, and the internal structure of the sentence is displayed on the left half surface 102.
[0026]
In this window 101, the following document with the internal structure described by tagging "C city where Mr. A's B meeting ended, some popular newspapers and general newspapers self-regulate their photo coverage Part of "Clarified policy on paper" is displayed. Here is an example of tagging this document:
[0027]
<Document> <Sentence> <Adjective Verb Phrase Relationship = “Position”> <Noun Phrase> <Adjective Verb Phrase Location = “C City”>
<Adjective verb phrase relation = “subject”> <noun phrase identifier = “B society”> <adjective verb phrase relation = “affiliation”> <person name identifier = “Mr. A”> </ adject verb of Mr. A </ person name> Phrase> <organization name identifier = “group B”> group B </ organization name> </ noun phrase> is </ adjective verb phrase>
</ Adjective verb phrase> <place name identifier = “C city”> C city </ place name> </ noun phrase> </ adject verb phrase> <adjective verb phrase relation = “subject”> <noun phrase identifier = "Press" Syntax = "Parallel"> <noun phrase> <adjective verb phrase> Some </ adject verb phrases> popular paper </ noun phrase> and <noun> general paper </ noun> </ noun phrase> </ Adjective verb phrase>
<Adjective verb phrase relation = “object”> <adjective verb phrase relation = “content” subject = “press”> <adjective verb phrase relation = “object”> <noun phrase> <adjective verb phrase> <noun co-reference = "B society"> So </ noun> </ adjective verb phrase> Photo report </ noun phrase> </ adjective verb phrase>
Self-regulating </ adjective verb phrase> policy </ adjective verb phrase>
<Adjective Verb Phrase Relation = “Position”> On the page </ Adjective Verb Phrase>
Revealed. </ Sentence> </ Document>
[0028]
In this document, “some popular papers and general papers” are represented as parallel by the tag “Syntax =“ Parallel ””. The definition of parallel is to share a dependency relationship. If nothing is specified, for example, <noun phrase relationship = x> <noun> A </ noun> <noun> B </ noun> </ noun phrase> indicates that A is dependent on B To express. Relation = x represents a relation attribute.
[0029]
Relational attributes describe the interrelationships between syntax, meaning, and rhetoric. Grammatical functions such as subject, object, indirect object, subject roles such as actors, activists, beneficiaries, and rhetorical relationships such as reasons, results, etc. are described by this relation attribute. In the present embodiment, relational attributes are described for relatively easy grammatical functions such as a subject, object, and indirect object.
[0030]
In this document, the attributes of proper nouns such as “Mr. A”, “Group B”, and “C City” are described by tags such as place names, person names, and organization names. Words to which tags such as place names, person names, and organization names are given are proper nouns.
[0031]
The operation of the document processing apparatus as an embodiment according to the present invention will be described below. The document processing device detects the actual interest level with respect to the document, and sets a priority order for other documents based on the detected actual interest level. The document processing apparatus displays a document and detects an actual interest level based on the displayed document. The actual interest level is detected according to the user's operation on the document. Based on the degree of association with the actual interest level, a predicted interest level is defined for a document for which no actual interest level is given. When the predicted interest level is used, a priority order can be given to a document that is not operated by the user.
[0032]
Prior to describing the actual interest level, manual document classification and automatic document classification will be described. That is, the operation of the document processing apparatus will be described in the order of (1) manual document classification, (2) automatic document classification, (3) actual interest level and predicted interest level.
[0033]
The contents of the explanation will be briefly described. (1) In the manual classification of a document, an operation will be described in which the document processing apparatus receives a document sent from the outside and the user manually classifies the document. With this manual classification, a classification model for classifying documents is created. (2) In the automatic document classification, an operation of classifying a document using the relevance between document classifications based on a classification model created by manual document classification will be described. (3) In the actual interest level and the predicted interest level, processing performed based on the actual interest level detected based on the user's operation and the predicted interest level obtained based on the actual interest level and the inter-document relevance level. Will be described.
[0034]
(1) Manual classification of documents
In this embodiment, there is no classification model in the initial state. In the initial state, it is necessary to manually classify documents sent from the outside in order to create a classification model. The manual classification operation of such a document processing apparatus will be described with reference to FIG.
[0035]
In step S11 of FIG. 4, the receiving unit 21 of the document processing apparatus receives a plurality of documents transmitted via a communication line, for example. The receiving unit 21 sends the received document to the main body 10 of the document processing apparatus.
[0036]
In step S12, the control unit 11 of the document processing apparatus extracts features of a plurality of documents sent from the receiving unit 21, and creates feature information, that is, an index of each document. The control unit 11 stores the received documents and the created index in, for example, the RAM 14. The index includes proper nouns, meanings other than proper nouns, and the like that are characteristic of the document.
[0037]
Here, a specific example of an index is shown.
[0038]
<Index date = "AAAA / BB / CC" Time = "DD: EE: FF" Document address = "1234">
<User operation history Maximum summary size = "100">
<Number of selected elements = “10”> Picturetel </ Selected>
...
</ User operation history>
<Summary> Tax reduction scale, untouched-Prime Minister X's meeting </ summary>
<Word meaning = “0003” Central activity value = “140.6”> Do not touch </ word>
<Word meaning = “0105” Identifier = “X” Central activity value = “67.2”> Prime Minister </ Word>
<Person name identifier = “X” Word meaning = “6103” Central activity value = “150.2”> Prime Minister X </ word / person name>
<Word meaning = “5301” central activity value = “120.6”> determined </ word>
<Word meaning = “2350” Identifier = “X” Central activity value = “31.4”> Prime Minister </ word>
<Word meaning = “9582” Central activity value = “182.3”> Emphasized </ Word>
<Word meaning = “2595” central activity value = “93.6”> touch </ word>
<Word meaning = "9472" Central activity value = "12.0"> Noticed </ Word>
<Word meaning = "4934" Central activity value = "46.7"> I didn't touch </ Word>
<Word meaning = “0178” central activity value = “175.7”> explained </ word>
<Word meaning = “7248” identifier = “X” central activity value = “130.6”> I </ word>
<Word meaning = “3684” Identifier = “X” Central activity value = “121.9”> Prime Minister </ word>
<Word meaning = “1824” central activity value = “144.4.”> Appealed </ word>
<Word meaning = “7289” central activity value = “176.8”> showed </ word>
</ Index>
[0039]
In this index, <index> and </ index> are the beginning and end of the index, <date> and <time> are the date and time this index was created, and <summary> and </ summary> are It shows the beginning and end of the summary of the contents of this index. <Word> and </ Word> indicate the beginning and end of a word, respectively. The meaning = “0003” indicates the third meaning. The same applies to other cases. That is, since the same word may have a plurality of meanings, a number is predetermined for each meaning in order to distinguish them. Therefore, one or more meanings exist for the same word.
[0040]
<User operation history> and </ User operation history> indicate the start and end of the user operation history, and <Select> and </ Select> indicate the start and end of the selected element, respectively. ing. The maximum summary size = “100” indicates that the maximum size of the summary is 100 characters, and the number of elements = “10” indicates that the number of selected elements is 10.
[0041]
In step S13 in FIG. 4, the user browses the document displayed on the display unit 30 of the document processing apparatus as shown in the specific example of the display in FIG. In FIG. 5, the document before classification by the user is classified as “other topics”, and the icon and title of the document are displayed in “other topics” of the first display unit 303 of the window 301. The control unit 11 of the document processing apparatus controls the display unit 30 to display a document desired by the user from among the plurality of documents displayed in this way. The control unit 11 selects a document to be displayed on the display unit 30 in accordance with a user input to the input unit 20. The display unit 30 displays the document selected by the user in a window that can change the size of the area. When the entire document cannot be displayed in this window, a part of the document is displayed.
[0042]
Note that step S13 in which the user browses the document is provided according to the user's needs. In the figure, the step S13 represented by a parallelogram indicates that the user operates. The same applies to the following.
[0043]
Here, a specific example of the display shown in FIG. 5 will be described in detail. In this specific example, a user can freely set or change a category for classifying documents. Such a category setting or change is manually performed by the user.
[0044]
A specific example of a graphic user interface (GUI) used for displaying the document classification in the display unit 30 is as shown in FIG. The document classification window 301 includes a position reset button for returning the window state to the initial position, a browser button for calling a browser for browsing the contents of the document, and an escape from this window. And an operation button 302 including an (exit) button.
[0045]
The document classification window 301 also includes a first classification display section 303 that displays the above-mentioned “other topics”, a second classification display section 304 that displays “business news”, and a third classification that displays “political news”. The category display section 305 and the like are displayed. In these classification sections, icons of documents and titles of documents classified into the categories are displayed corresponding to the respective categories. If there is no title, a one sentence summary is displayed. The size of each classification display unit is not fixed, and can be changed to a desired size by operating the mouse of the input unit 20, for example. Also, the title or label of the classification display section can be freely changed.
[0046]
In the “other topics” of the first classification display unit 303, for example, the titles of documents before being classified into categories corresponding to the second classification display unit 304 and the following are displayed. That is, in the manual classification process, the document received by the document processing apparatus is once displayed in “other topics” of the first classification display unit 303. The documents displayed on the first classification display unit 303 are classified into categories by the user as follows.
[0047]
In step S14 of FIG. 4, the user creates a classification model composed of a plurality of categories for classifying the plurality of documents viewed on the display unit 30 of the document processing apparatus in step S13. Then, the plurality of documents are classified into each category of the classification model.
[0048]
The classification model is composed of a plurality of classification items, that is, categories, for classifying documents. The category is composed of a category index including proper nouns, meanings other than proper nouns, document addresses included in the categories, and the like characteristic of the category. The category index is composed of a document index including proper nouns and meanings other than proper nouns.
[0049]
For example, the classification model shown in FIG. 7 has columns of proper nouns, meanings other than proper nouns, and document addresses for category indexes corresponding to the respective categories. In this classification model, the proper nouns “Mr. A,...”, “Mr. B,. .. ”,“ C company, G company,... ”,“ D type,... ”,“ Mr. E,... ”And“ Mr. F ”, meaning“ baseball (4546), ground (2343) ), ... "," Labor (3112), Unique (9821), ... "," Mobile (2102), ... "," Cherry 1 (11111), Orange 1 (9911) "," Cherry 2 (11112), orange 2 (9912), and “cherry tree 3 (11113)” as document addresses “SP1, SP2, SP3,...”, “SO1, SO2, SO3,.・ ”,“ CO1, CO2, CO3,... ”,“ PL1, PL2, PL3 ··· "," AR1, AR2, AR3, ··· "and" EV1, EV2, EV3, each have a ... ". “Sakura 1”, “Sakura 2”, and “Sakura 3” indicate the first meaning (11111), the second meaning (11112), and the third meaning (11113) of “Sakura”. “Orange 1” and “Orange 2” indicate the first meaning (9911) and the second meaning (9912) of “orange”. For example, “Orange 1” represents a plant orange and “Orange 2” represents an orange color.
[0050]
When the classification model is updated, the update date and time is recorded in the classification model. In the figure, “Dec. 10, 1998 19:56:10” is recorded as the update date.
[0051]
The category of the classification model is manually created by the user by changing or deleting the classification display section corresponding to each category or setting a new classification display section in the document classification window 301.
[0052]
For example, in the document classification window 301, a classification display unit corresponding to a desired category is displayed by using the mouse corresponding to the title of the document displayed on the classification display unit in the document classification window 301. This is done by dragging to. The titles of documents classified into categories are displayed on the classification display unit corresponding to each category in the document classification window 301.
[0053]
In step S15, the control unit 11 of the document processing apparatus creates a classification model based on the creation of the category performed in step S14 and the index of each document classified by the user's manual classification operation according to this category. Create That is, the control unit 11 of the document processing apparatus collects the indexes of the plurality of documents classified into each category, and generates a classification model.
[0054]
The category index of each category includes a proper noun characteristic for the category, a meaning other than the proper noun, and a document address classified into each category. Here, in the case of other than proper nouns, the meaning is used instead of the word itself because the same word may have a plurality of meanings. Then, the control unit 11 of the document processing apparatus stores the classification model created in this way in, for example, the RAM 14.
[0055]
The classification model can be created in step S15 each time a category is created in step S14 and a user's manual classification operation is performed.
[0056]
In step S16, the control unit 11 of the document processing apparatus registers the classification model created in step S15. The control unit 11 stores the registered classification model in, for example, the RAM 14.
[0057]
(2) Automatic document classification
Next, automatic document classification performed by the document processing apparatus based on the classification model will be described with reference to FIG. This document classification is performed on a document received after the classification model is created by the processing shown in FIG. In this example, it is assumed that the processing shown in FIG. 8 is performed every time one document is received. However, the processing may be performed every time a predetermined number of documents are received, or the user can display the screen shown in FIG. Processing may be performed on all documents received so far when an operation of opening is performed.
[0058]
In step S21, the receiving unit 21 of the document processing apparatus receives a document from the outside. Since the reception of this document has been described in step S11, the description thereof will be omitted here.
[0059]
In step S22, the control unit 11 of the document processing apparatus reads the document stored in the RAM 14 in step S21 and creates an index. The creation of this index will be further described later.
[0060]
In step S23, the control unit 11 of the document processing device automatically classifies each indexed document into any category of the classification model based on the classification model. Then, the control unit 11 stores the classification result in, for example, the RAM 14. Details of the automatic classification will be described later.
[0061]
In step S24, the control unit 11 of the document processing apparatus updates the classification model based on, for example, the result of automatic classification of a new document stored in the RAM 14 in step S23. In step S25, the control unit 11 of the document processing apparatus registers the classification model updated in step S24. The control unit 11 stores the registered classification model in, for example, the RAM 14.
[0062]
Next, index creation in step S12 of FIG. 4 and step S22 of FIG. 8 will be described with reference to FIG.
[0063]
In step S31, the control unit 11 of the document processing apparatus performs active diffusion for diffusing the central activation value of the element based on the internal structure of the document received in step S11 of FIG. 4 and step S21 of FIG. Execute. The center activity value diffusion process will be described later. The control unit 11 stores the central activity value of each element obtained as a result of active diffusion in, for example, the RAM 14.
[0064]
In step S32, the control unit 11 of the document processing device extracts elements whose center activity value exceeds a preset threshold based on the center activity value of each element obtained in step S31. The control unit 11 stores the extracted elements in, for example, the RAM 14.
[0065]
In step S33, the control unit 11 of the document processing apparatus reads the element extracted in step S32 from the RAM 14, for example. Then, the control unit 11 extracts all proper nouns from this element and adds them to the index. Proper nouns have no special meaning and have special properties such as not being listed in the dictionary, so they are handled separately from words other than proper nouns. Here, the word meaning corresponds to each meaning of a plurality of meanings of the word.
[0066]
The control unit 11 of the document processing apparatus determines whether or not the element is a proper noun based on a tag attached to the received document. For example, in the internal structure with tagging shown in FIG. 3, “Mr. A”, “Group B”, and “C City” have the relationship attributes of “person name”, “organization name”, and “place name”, respectively. It can be seen that it is a proper noun. And the control part 11 adds the taken-out proper noun to an index, and memorize | stores the result in RAM14, for example.
[0067]
In step S34, the control unit 11 of the document processing apparatus extracts the meaning other than the proper noun from the element extracted in step S32 from the RAM 14, for example, adds it to the index, and stores the result in the RAM 14.
[0068]
As described above, the procedure for finding the feature of the document and creating the index is to find the feature of the tagged document and create an index in which the feature is arranged. The feature of the document is determined based on the central activity value that has been subjected to the diffusion processing according to the internal structure of the document.
[0069]
Note that the above-described index includes a document address indicating a position where the document is stored in the RAM 14 together with a meaning and proper noun representing the document characteristics.
[0070]
Since the index includes meanings and proper nouns representing features that represent the document, it can be used when referring to a desired document.
[0071]
Next, active diffusion for diffusing the central active value corresponding to the element based on the internal structure of the document will be described with reference to FIG. The active diffusion is performed in step S31 of FIG. The active diffusion is a process for giving a high central activity value to an element related to an element having a high central activity value. Since this central activity value is determined according to the internal structure by tagging, it is used for extracting document features and the like.
[0072]
In step S81, the control unit 11 of the document processing apparatus sets the end point activation value of the end point of the link connecting the elements to 0 for the reference / referenced link and the normal link. The control unit 11 stores the initial value of the end point activation value thus assigned, for example, in the RAM 14.
[0073]
The connection between the elements is as shown in FIG. 11, for example. In this figure, element E is part of the structure of the elements and links that make up the document._iAnd element E_jIt is shown. Element E_iAnd element E_jAnd the central activity value e_iAnd e_jEach with a link L_ijConnected at. Link L_ijElement E_iThe end point connected to is T_ij, Element E_jThe end point connected to is T_jiIt is. Element E_iIs the link L_ijE connected by_jIn addition to link L_ik, L_ilAnd L_imElement E (not shown)_k, E_lAnd E_mIs connected to each. Element E_jIs element E_jLink L based on_ijL_jiE connected by_iIn addition to link L_jp, L_{j q}And L_jrElement E (not shown)_p, E_qAnd E_rIs connected to each.
[0074]
In step S82, the control unit 11 of the document processing apparatus performs element E constituting the document._iInitialize the counter that counts. That is, the count value i of the counter for counting elements is set to 1. This counter is the first element E₁Will be referred to.
[0075]
In step S83, the control unit 11 of the document processing apparatus executes a link process for calculating a new center activation value for the element referred to by the counter. This link process will be further described later.
[0076]
In step S84, the control unit 11 of the document processing apparatus determines whether or not the calculation of new center activation values has been completed for all elements in the document. Then, when the calculation of the central activation value for all elements in the document is completed, the control unit 11 proceeds to step S85 as “YES”, and the calculation of the new central activation value for all elements in the document is completed. If not, "NO" is determined and the process proceeds to step S87.
[0077]
Specifically, the control unit 11 determines whether or not the count value i of the counter has reached the total number of elements included in the document. Then, when the count value i of the counter reaches the total number of elements included in the document, the control unit 11 determines that all elements have been calculated and proceeds to step S85. When the count value i of the counter does not reach the total number of elements included in the document, the control unit 11 determines that calculation has not been completed for all elements and proceeds to step S87.
[0078]
In step S87, the control unit 11 of the document processing apparatus increments the count value i of the counter by 1, and sets the count value of the counter to i + 1. As a result, the counter becomes i + 1th E_{i + 1}The next element is referred to. Then, the process returns to step S83, and the calculation of the end point activation value and the subsequent series of steps are performed to determine the next i + 1th element E._{i + 1}To be executed.
[0079]
In step S85, the control unit 11 of the document processing apparatus calculates the average value of the change in the central activity value of all elements included in the document, that is, the change in the newly calculated central activity value with respect to the original central activity value. Calculate
[0080]
The control unit 11 of the document processing apparatus reads, for example, the original central activity value stored in the RAM 14 and the newly calculated central activity value for all elements included in the document. The control unit 11 divides the sum of the respective changes of the newly calculated center activity value with respect to the original center activity value by the total number of elements included in the document, so that the change amount of the center activity value of all the elements is calculated. Calculate the average value. The control unit 11 stores, for example, the RAM 14 in the average value of the change in the central activity value of all the elements calculated in this way.
[0081]
In step S86, the control unit 11 determines whether or not the average value of the change in the central activity value of all the elements calculated in step S89 is within a preset threshold value. And the control part 11 will complete | finish this series of steps as "YES", when the said change is less than a threshold value. When the change is not within the threshold value, the control unit 11 determines “NO” and sets the counter count value i to 1 in step S82, and again executes a series of steps for calculating the central activation value of the document element. To do. Each time the loop from step S82 to step S84 configured in this series of steps is repeated, the amount of change gradually decreases.
[0082]
Next, the link process executed in step S83 in FIG. 10 will be described with reference to FIG. Here, one element E_iAs an example, in the case of the center active value diffusion process, the link process is performed for all elements.
[0083]
In step S51, the control unit 11 of the document processing apparatus performs element E constituting the document._iAnd a counter that counts the links to which one end is connected is initialized. That is, the count value j of the counter for counting links is set to 1. Counter is element E_iThe first element L connected to_i1Will be referred to.
[0084]
In step S52, the control unit 11 of the document processing apparatus performs element E._iAnd E_jLink L connecting_ijWith reference to the tag of the relation attribute, it is determined whether or not the link is a normal link. The control unit 11 is connected to the link L_ijWhen is a normal link, “YES” is determined, and the process proceeds to step S53._ijIf is a reference link, "NO" is determined and the process proceeds to step S54.
[0085]
In step S53, the control unit 11 of the document processing apparatus performs element E._iNormal link L_ijEnd point T connected to_ijThe process of calculating a new end point activation value of is performed.
[0086]
Here, the link L is determined by the determination in step S52._ijIs usually a link. Element E_iNormal link L_ijEnd point T connected to_ijEnd point activation value t_ijIs element E_jLink L of the end point activation values of_ijAll end points T connected to links other than_jp, T_jq, T_jrEnd point activation value t_jp, T_jq, T_jrAnd element E_iIs link L_ijE connected by_jCenter activity value e_jAnd the value obtained by this addition is divided by the total number of elements included in the document.
[0087]
The control unit 11 of the document processing apparatus reads the end point activation value and the center activation value from the RAM 14, for example. The control unit 11 calculates a new end point activation value of the end point connected to the normal link as described above with respect to the read end point activation value and center activation value. And the control part 11 memorize | stores the endpoint active value calculated in this way in RAM14, for example.
[0088]
In step S54, the control unit 11 of the document processing apparatus performs element E._iEnd point T connected to the reference link of_ijThe process of calculating the end point activation value of is performed.
[0089]
Based on the determination in step S52, the link L_ijIs a reference link. Element E_iReference link L_ijEnd point T connected to_ijNew end point activation value t_ijIs element E_jThis link L of the end point activation values of_ijAll end points T connected to the link excluding_jp, T_jq, T_jrEnd point activation value t_jp, T_jq, T_jrAnd element E_iIs link L_ijE connected by_jCenter activity value e_jIs obtained by adding.
[0090]
The control unit 11 of the document processing apparatus reads out necessary end point activation values and center activation values from, for example, end point activation values and center activation values stored in the RAM 14. The control unit 11 calculates a new end point activation value connected to the reference link as described above, using the read end point activation value and center activation value. And the control part 11 memorize | stores the endpoint active value calculated in this way in RAM14, for example.
[0091]
The normal link process in step S53 and the reference link process in step S54 are in the loop from step S52 to step S55, and the element E referred to by the count value i is used._iAll links L connected to_ijIs executed against.
[0092]
In step S55, the control unit 11 of the document processing apparatus performs element E._iIt is determined whether or not endpoint activation values have been calculated for all links connected to. If end point activation values are calculated for all links, the process proceeds to step S57 as “YES”, and if end point activation values are not calculated for all links, the process proceeds to step S57 as “NO”.
[0093]
In step S56, element E in step S55._iAll links of L_ijEnd point activation value t_ijTherefore, the control unit 11 of the document processing apparatus determines that the element E_iCenter activity value e_iPerform the update.
[0094]
Element E_iCenter activity value e_iThe new or updated value of is the element E_iCurrent central activity value e_iAnd element E_iThe sum of new endpoint activation values for all endpoints of_i‘= E_i+ Σt_jIt is calculated | required by taking '. Here, the prime “′” means a new value.
[0095]
The control unit 11 of the document processing apparatus reads a necessary end point activation value from, for example, the end point activation value and the center activation value stored in the RAM 14. The control unit 11 executes the calculation as described above, and the element E_iCenter activity value e_iIs calculated. Then, the control unit 11 calculates the new center activity value e_iIs stored in the RAM 14, for example.
[0096]
Next, automatic classification in step S23 of FIG. 8 will be described with reference to FIG.
[0097]
In step S71, the control unit 11 of the document processing apparatus performs category C of the classification model._iFor the set of proper nouns and the set of proper nouns of the words extracted from the document received in step S21 and put into the index, the number of these common sets is expressed as P (C_i ). The control unit 11 then calculates the number P (C_i ) Is stored in the RAM 14, for example.
[0098]
In step S72, the control unit 11 of the document processing apparatus determines all meanings included in the index of the document and each category C._iThe relationship between meanings with all meanings included in the word meaning is referred to a table of relationship between meanings shown in FIG._i ) Is calculated. That is, the control unit 11 sums up the degree of relevance between all meanings R (C for words other than proper nouns in the classification model._i ) Is calculated. Then, the control unit 11 calculates the total sum R (C_i ) Is stored in the RAM 14, for example.
[0099]
In step S73, the control unit 11 of the document processing apparatus performs category C._i Relevance between document classifications
Rel (C_i ) = M₁P (C_i ) + N₁R (C_i )
It is defined as Where the coefficient m₁, N₁Is a constant and represents the degree of contribution of each value to the degree of association between document classifications. The control unit 11 determines the number P (C of common sets calculated in step S72._i ) And the sum R of meanings between meanings calculated in step S73 (C_i ) Is read from, for example, the RAM 14, and is applied to the above-described formula to determine the degree of relationship between document classifications Rel (C_i ) Is calculated. These coefficients m₁, N₁For example, m₁= 10, n₁= 1. Then, the control unit 11 calculates the relevance level between document classifications Rel (C_i ) Is stored in the RAM 14, for example.
[0100]
Coefficient m₁And n₁The value of can also be estimated using statistical techniques. That is, the control unit 11 determines the relationship between document classifications Rel (C_i ) Can be obtained by optimization.
[0101]
In step S74, the control unit 11 of the document processing apparatus performs category C._iRelevance between document classifications Rel (C_i ) Is the largest, and the degree of relevance Rel (C_i ) Value exceeds a certain threshold, the category C_iSort documents into That is, the control unit 11 creates a document category relevance level for each of a plurality of categories, and when the maximum document category relevance level exceeds a threshold, the document has the maximum document category relevance level. Category C_iClassify into: When the maximum degree of association between document classifications does not exceed the threshold value, document classification is not performed.
[0102]
Next, the calculation of the degree of association between meanings used in step S72 of FIG. 13 will be described with reference to FIG. The process shown in FIG. 14 need only be performed once before the process shown in FIG.
[0103]
In step S <b> 61, the control unit 11 of the document processing apparatus creates a meaning network using this dictionary using the explanation of the meaning of words in the electronic dictionary. That is, a meaning network is created from the reference relationship between each meaning in the dictionary and the meaning that appears in the description. As a result, a tree-like semantic network with the dictionary as the highest vertex is constructed. The internal structure of the network is described by tagging as described above. The control unit 11 of the document processing apparatus creates a network by sequentially reading the meaning and explanation of the electronic dictionary stored in the RAM 14, for example. The control unit 14 stores the semantic network created in this way in, for example, the RAM 14.
[0104]
The network is created by the control unit 11 of the document processing apparatus using a dictionary, received from the outside by the receiving unit 21, or reproduced from the recording medium 32 by the recording / reproducing unit 31. Can also be obtained. The dictionary is obtained by being received from the outside by the receiving unit 21 or reproduced from the recording medium 32 by the recording / reproducing unit 31.
[0105]
In step S62, the central activation value corresponding to each semantic element is diffused on the semantic network created in step S61. By this active diffusion, the central activity value corresponding to each meaning is given according to the internal structure by tagging given by the dictionary. The center activity value diffusion process will be described later.
[0106]
In step S63, one meaning s constituting the meaning network created in step S61._iIn step S64, this one meaning s_iVocabulary element E corresponding to_iCenter activity value e_iThe initial value of the center activation value at this time is changed._iCalculate
[0107]
In step S65, element E in step S64._iCenter activity value e_iDifference Δe_iOther meanings corresponding to_jElement E corresponding to_jCenter activity value e_jDifference Δe_jAsk for. In step S66, the difference Δe obtained in step S65._jΔe obtained in step S64_iQuotient Δe divided by_j/ Δe_i, Meaning_iMeaning of_jIs the degree of relevance between meanings.
[0108]
In step S67, one meaning s_iAnd other meanings_jIt is determined whether or not the calculation of the relationship between meanings is completed for all pairs. Then, when the calculation of the relationship between meanings is completed for all the meaning pairs, this processing is ended as “YES”. When the calculation of the relationship between meanings is not completed for all semantic pairs, “NO” is returned to step S63, and the calculation of the relationship between meanings is continued for the pair whose calculation of the relationship between meanings is not completed. To do.
[0109]
In the loop from step S63 to step S67, the control unit 11 of the document processing apparatus reads necessary values sequentially from the RAM 14, for example, and calculates the degree of association between meanings as described above. The control unit 11 stores the calculated degree-of-sense relevance in order, for example, in the RAM 14.
[0110]
The degree of association between meanings calculated in this way is defined between each meaning and meaning as shown in FIG. In this table, the degree of association between meanings is normalized so as to take a value from 0 to 1. In this table, the degree of relevance between meanings among “computer”, “television”, and “VTR” is shown. The degree of association between the meanings of “computer” and “TV” is 0.55, the degree of association between the meanings of “computer” and “VTR” is 0.25, and the degree of association between the meanings of “TV” and “VTR” is 0.60. It is.
[0111]
(3) Actual interest level and predicted interest level
Next, details of step S13 in FIG. 4 will be described with reference to FIG. By performing this processing, the actual interest level is detected.
[0112]
In step S101, the user selects a desired document from the document classification window 301 shown in FIG. For example, the user selects an icon corresponding to the title of the document displayed in the classification display section of the document classification window 301 with the mouse of the input section 20. Then, by selecting the “browser” button of the operation button 302, the process proceeds to the display step of the next step S102.
[0113]
In step S102, the control unit 11 of the document processing apparatus reads the document selected by the user in step S101 from the RAM 14, for example. In the display unit 30, the control unit 11 displays the read document on the document display unit 53 of the window 51. As described above, when the entire document cannot be displayed on the document display unit 53 of the window 51, a part of the document is displayed.
[0114]
In step S <b> 103, the user reads the document displayed in the document display unit 53 of the window 51 in step S <b> 102 and creates a summary. That is, the user reads the document on the document display unit 53 of the window 51 displayed in step S102. In addition, the user selects the “summerize” button of the operation button 56 of the window 51 to display the summary of the document displayed on the document display unit 53 on the summary display unit 54.
[0115]
Here, when a summary is created and displayed on the summary display unit 54, a procedure for increasing the importance of the element selected by the user in the document with respect to the document displayed on the document processing unit 53 by the user's operation is shown in FIG. This will be described with reference to the flowchart shown in FIG.
[0116]
In the first step S91, the control unit 11 determines whether or not an element in the document has been selected by the user. This determination is made by selection using a graphic user interface (GUI) shown in FIG. 18 that accepts user input.
[0117]
The window 51 includes a file name display unit 52 for displaying the file name of the document, a document display unit 53 for displaying the document with the file name displayed on the file name display unit 52, and the document displayed on the document display unit 53. A summary display unit 54 for displaying the summary is provided. The document display unit 53 displays all or part of the document in which the file name or the first part of the document is displayed in the file name display unit 52. When only a part of the document is displayed on the document display unit 53, for example, by scrolling the document displayed on the document display unit 53, the entire document can be browsed sequentially. The summary display unit 53 displays a summary of the document displayed on the document display unit 53 by processing to be described later, corresponding to the size of the summary display unit 54. The summary display section 53 is blank because no summary has been created. Note that the sizes of the document processing unit 53 and the summary display unit 54 can be changed. The document handled in the window 51 is, for example, received by the receiving unit 21 of the document processing apparatus and recorded in the recording / reproducing unit 31 or the RAM 14.
[0118]
The window 51 includes a keyword input unit 55 for inputting a keyword and a button unit 56 having a plurality of buttons. By inputting a keyword into the keyword input unit 55, the importance of a word having a high degree of association with the keyword among the words displayed on the document display unit 54 is increased. The button unit 56 includes an “Undo” button for returning the execution result, and a process of summarizing the text displayed on the document display unit 53 and displaying it on the summary display unit 54. summarize) ”button. Among these, by selecting the “summary” button, for example, when the size of the summary display portion 54 is changed, the summary is displayed on the document processing portion 53 so as to correspond to the new size of the new summary display portion 54. A summary of the current document is generated, and the generated summary is displayed on the summary display unit 54.
[0119]
In step S91 in FIG. 17, the control unit 11 determines whether or not an element in the text displayed on the document display unit 53 has been selected by the user in the window 51 displayed on the display unit 30 of the document processing apparatus. . The input unit 20 of the document processing apparatus that selects and inputs an element in the document display unit 53 is performed by operating a cursor displayed on the display unit 30 linked to the pointing device using a pointing device. Can do. For example, when a mouse is used as the pointing device, the mouse is operated to move the cursor to a desired element of the document processing unit 53, and the element is selected by clicking with the mouse. When an element is selected in the document display section 53, the selected element is highlighted, for example, in order to clearly show the selected element. In FIG. 19, in the document display section 53 of the window 51, the vocabulary element “mainframe” 57 which is the smallest selected element is highlighted. The summary display section 53 is blank because no summary has been created yet. When the element is selected in this way, the control unit 11 determines “YES” and proceeds to the next step S92. When no element is selected, for example, when there is no input within a predetermined time, or when a portion other than the portion where the text is displayed in the document display unit 53 is clicked with the mouse, the control unit 11 again sets “NO” to this step S91. The processing is returned to and the input of the element is waited. In the following description, for convenience of explanation, the description will be made assuming that a mouse is used as the pointing device of the input unit 20.
[0120]
In step S92, the control unit 11 of the document processing apparatus determines whether the word has been selected in step S91 but has been selected by clicking with the mouse in the past. When the element is an element selected by clicking with the mouse in the past, the control unit 11 determines “YES” and advances the process to step S93. If the element is not the element selected by clicking with the mouse in the past, the control unit 11 advances the process to step S94 as “NO”.
[0121]
In step S93, the control unit 11 of the document processing apparatus determines whether the selected element is a text element. When the level is a text element, the control unit 11 returns “YES” to step S91. When the level is not a text element, the control unit 11 determines “NO” and proceeds to the next step S95.
[0122]
In step S94, the control unit 11 of the document processing apparatus sets the level to the vocabulary element that is the minimum element of the document and the lowest element of the internal structure by tagging the document. And the control part 11 returns a process to step S91.
[0123]
In step S95, the control unit 11 of the document processing apparatus increases the level by one. For example, when the level is increased by 1, the vocabulary element “mainframe” 57 selected in step S91 is, as shown in FIG. 20, the next higher order element “Big mainframe computers” including this vocabulary element. “59” is selected, and this portion “Big mainframe computers” 59 is highlighted. At the same time, the control unit 11 increases the weight of the selected upper element, that is, the central activity value higher than that of the element not selected. And the control part 11 returns a process to step S11.
[0124]
When the “summary” button displayed on the button section 56 of the window 51 is selected by clicking the mouse, the summary of the text displayed on the document display section 53 is displayed on the summary display section 54. When the “summary” button is selected, the control unit 11 controls the process to exit from the series of steps shown in FIG. 17 by interruption, and starts a process for creating a summary. The summary is generated from the document displayed on the document display unit 53 so as to fill the area of the summary display unit 54 in accordance with the size of the summary display unit 54. As shown in FIG. 21, the summary displayed on the summary display unit 54 displays an element “Big mainframe computers” 60 corresponding to the element “Big mainframe computers” 59 highlighted on the document display unit 59. Yes. Thus, by selecting a desired element in the document display unit 53 of the window 51 and increasing the importance, the possibility that the element is included in the summary can be increased. Details of the summary generation will be described later.
[0125]
In the window 51 shown in FIG. 18, selection of elements in the document displayed on the document display unit 53 can be performed by inputting a keyword to the keyword input unit 55 in addition to clicking with the mouse. The control unit 11 performs a process of increasing the importance of the element related to the keyword input to the keyword input unit 55 in this way. The degree of association between the keyword and the element is obtained by referring to a table recorded in the ROM 15, for example. This reference is performed by referring to the element including the keyword by tagging.
[0126]
In step S104 of FIG. 16, the operation unit 11 of the document processing apparatus calculates the actual interest level of the user's document. The actual interest level is calculated based on an operation on the document displayed in the user window 51 in step S103.
[0127]
Here, the actual interest level and the predicted interest level used in the present embodiment will be described. The actual interest level is calculated in step S104 and is an actual interest level for the document operated by the user, which is detected by the user operation. On the other hand, the predicted interest level is a prediction of the interest level of the user's document. This predicted interest level is predicted based on the actual interest level, for example.
[0128]
In step S105, the control unit 11 records the user operation history in the index. In the specific example of the index described above, as the user operation history,
<User operation history Maximum summary size = "100">
<Number of selected elements = “10”> Picturetel </ Selected>
...
</ User operation history>
Was exemplified. In step S105, the control unit 11 updates the operation history such as the maximum size of the summary, the selected element, and the number of selected elements. The control unit 11 stores the updated index in, for example, the RAM 14.
[0129]
The index can include the actual interest level of the document. For example, the actual interest level for each document may be included in the index for each category. In such a case, in step S105, the actual interest level itself included in the index related to the document is also updated.
[0130]
Next, the user operation in step S103 of FIG. 16 will be described with reference to FIG. 22, FIG. 23, FIG. 23, and FIG.
[0131]
The document whose title is displayed in the document classification window 301 can be displayed on the display unit 30 by selecting on the display unit 30 using the mouse of the input unit 20, for example. A specific example of the document display window for displaying the document in this way is shown in FIG. 18, and the description thereof is omitted here.
[0132]
Next, an example including a more detailed control than that shown in FIG. 4 of the process of creating a summary will be described in detail with reference to the flowchart shown in FIG. This series of steps is started by turning on the “Summary” button 103.
[0133]
The process of creating a summary from a document is performed based on the internal structure by tagging the document. As described above, the size of the display area 130 for displaying the summary in the window 100 can be changed. When the window 101 is newly drawn in the window 100 of the display unit 30 or the size of the display area 130 is changed and the execution button 103 is operated, the control unit 11 of the document processing apparatus adapts to the display area 130. A process of creating a summary from the document displayed in the display area 120 of the window 100 is executed.
[0134]
In the first step S120 in FIG. 22, the control unit 11 of the document processing apparatus performs active diffusion. In the present embodiment, documents are summarized by adopting the central activity value obtained by active diffusion as the importance. That is, in a document given an internal structure by tagging, a central activity value corresponding to the internal structure by tagging can be given to each element by performing a process called active diffusion. The active diffusion is a process for giving a high central activity value to an element related to an element having a high central activity value. That is, active diffusion is an operation on a central activity value in which the central activity value is equal between the anaphoric (co-reference) expression and its antecedent, and otherwise the central activity value is attenuated. Since this central activity value is determined according to the internal structure by tagging, it can be used for the analysis of documents in consideration of the internal structure by tagging.
[0135]
In step S121, the control unit 11 of the document processing apparatus sets the size of the document processing unit 53 of the window 51 displayed on the display unit 30, specifically, the maximum number of characters that can be displayed on the document processing unit 53 as w._sAnd set. Further, the control unit 11 of the document processing apparatus initializes s for storing the summary character string and initializes the initial value s.₀Set “=”. The control unit 11 sets the maximum number of characters w that can be displayed on the document display unit 53 as set above._sAnd an initial value s for storing the summary string₀Is recorded in the RAM 14, for example.
[0136]
In step S122, the control unit 11 of the document processing apparatus sets the count value i of the counter that counts the sequential creation of the summary skeleton to zero. That is, the control unit 11 sets i = 0 for the count value. The control unit 11 records the count value i set in this way in, for example, the RAM 14.
[0137]
In step S123, the control unit 11 of the document processing apparatus extracts the skeleton of the sentence having the i-th average central activity value from the sentence with respect to the count value i of the counter. The average central activity value is an average of the central activity values of the elements constituting one sentence. The control unit 11 stores the summary recorded in the RAM 14, for example._i-1Read this_i-1Add the extracted sentence skeleton to s_iAnd Then, the control unit 11 obtains s thus obtained._iIs recorded in the RAM 14, for example. At the same time, the control unit 11 makes a list l in the order of the central activation values of elements not included in the skeleton of the sentence._iCreate this list_iIs recorded in the RAM 14, for example.
[0138]
That is, in this step S123, the summary algorithm selects sentences in descending order of the average central activity value using the result of active diffusion, and extracts the skeleton of the selected sentence. The skeleton of the sentence is composed of essential elements extracted from the sentence. The required elements can be the element head, subject, object, indirect object, possessor, cause, condition Alternatively, an element having a relation attribute of comparison and an element directly included in the case where the coordinate structure is an essential element constitute an essential element. Then, the skeleton of the sentence is generated by connecting the essential elements of the sentence and added to the summary.
[0139]
In step S124, the control unit 11 of the document processing apparatus performs s_iIs the maximum number of characters w in the summary display section 54 of the window 51_sJudge whether it is larger. And the control part 11 is s_iIs the maximum number of characters w_sWhen it is larger, “YES” is set and this series of processing is ended. The control unit is s_iIs the maximum number of characters w_sIf not, the process proceeds to step S125 as “NO”. In other words, in step S124, the process ends when the summary reaches the designated amount. If there is still room, the sentence with the highest central activity value is compared with the central activity value of the omitted element, and the higher one is added to the summary.
[0140]
In step S129, the control unit 11 of the document processing apparatus performs s in step S124._iIs the maximum number of characters w_sSince it was determined to be greater than_i-1Set to. In this case, the summary does not fit in the window, so_i= S₀= "" Is output. Therefore, the summary is not displayed at this time. And the control part 11 complete | finishes this series of processes.
[0141]
In step S125, the control unit 11 of the document processing apparatus determines that the i + 1th average central activity value is the central activity value and the list l created in step S23._iThe central activity values of the elements having the highest central activity value are compared. Then, the control unit 11 determines that the central activity value of the sentence having the i + 1th average central activity value is the list l_iIf the central activation value of the element is higher than the central activation value of the element having the highest value, the process proceeds to the next step S27 as “YES”. The control unit 11 determines that the central activity value of the sentence with the i + 1th average central activity value is the list l_iIf the center activity value of the element is not higher than the center activity value of the element, the process proceeds to step S126 as “NO”.
[0142]
In step S126, the control unit 11 of the document processing apparatus increases the count value i of the counter by 1. And the control part 11 returns a process to step S123.
[0143]
In step S127, the control unit 11 of the document processing apparatus performs the list l._iThe element e with the highest central activity value of_iIn addition to ss_iIs generated. Element e_iDelete from. Then, the control unit 11 generates the ss generated in this way._iIs recorded in the RAM 14, for example.
[0144]
In step S128, the control unit 11 of the document processing apparatus performs ss._iIs the maximum number of characters w in the summary display section 54 of the window 51_sIt is determined whether it is larger. The control unit 11 is ss_iLength of w_sIf it is larger, “YES” is set and this series of steps is ended. The control unit 11 is ss_iLength of w_sIf it is not greater, “NO” is returned to step S125.
[0145]
In step S130, the control unit 11 of the document processing apparatus performs SS in step S128._iIs the maximum number of characters w_sSince it was determined to be greater than_iSet to. As a result, the maximum number of characters w_sA summary sentence is generated so as not to become larger. And the control part 11 complete | finishes this series of processes.
[0146]
The window 51 includes a keyword input unit 55 for inputting a keyword and a button unit 56 having a plurality of buttons. By inputting a keyword into the keyword input unit 55, among the words displayed on the document display unit 53, the actual interest level of the keyword and a word having a high degree of association between meanings to be described later is increased. The button unit 56 includes an “Undo” button for returning the execution result, and a process for summarizing the text displayed on the document display unit 53 and displaying it on the summary display unit 54. summarize) ”button. Among these, by selecting the “summary” button, for example, when the size of the summary display unit 54 is changed, the summary is displayed in the document processing unit 53 so as to correspond to the new size of the summary display unit 54. A summary of the document is generated, and the generated summary is displayed on the summary display unit 54.
[0147]
The actual interest level of the user with respect to the document is calculated based on a plurality of factors as follows. It should be noted that the actual interest level element is different from the elements constituting the document.
[0148]
In the calculation of the actual interest level, among the elements designated by the user, the position of the element whose appearance position in the document is farthest from the beginning of the document is the first element A (D_i). According to the first element, among the elements specified by the user, the position where the appearance position in the document is farthest from the beginning of the document is larger, and the user thinks that the document has been read more. It is assumed that the actual interest in the document is large. Specifically, the ratio between the maximum appearance position of the selected element and the size of the entire document is set as the first element A (D_i). Where D_iRepresents the i-th document.
[0149]
In the document display section 53 of the window 51 shown in FIG. 23, the first element 57, the second element 58, and the third element 59 are designated by the user and highlighted. Of these, the third element 59 farthest from the beginning of the document is used for the calculation of the actual interest level.
[0150]
Further, in the calculation of the actual interest level, the second element E indicates the number of items selected by the user from the document elements displayed on the document display unit 53 of the window 51 and the number of keywords input by the user in the keyword input unit 55. (D_i).
[0151]
In the document display section 53 of the window 51 shown in FIG. 23, designation of the first element 57, the second element 58, and the third element 59 is input by the user. The keyword “AAA” is input to the keyword input unit 55. The number of inputs of these elements and keywords is the second element E (D_i).
[0152]
Further, in the calculation of the actual interest level, the ratio of the size of the area of the summary display section 54 in the window 51 to the size of the area of the text display section 53 is set as the third element W (D_i). This is because the summary is displayed according to the size of the area of the summary display unit 54. However, the higher the user's actual interest level, the easier the user will find a detailed summary, that is, a long summary. Therefore, as the ratio of the size of the region of the summary display unit 54 to the size of the region of the text display unit 53 increases, the actual interest level can be increased.
[0153]
In the window 51 shown in FIG. 24, the ratio of the maximum size of the summary display unit 54 that displays the summary to the size of the document display unit 53 that displays the entire document is set as the third factor W (D_i).
[0154]
First element A (D of actual interest level_i), Second factor E (D_i) And the third element W (D_i) Based on the user's document D_iThe actual interest level IR (Di) for
IR (Di) = l₂W (Di) + m₂A (Di) + n₂E (Di)
It is defined as Where the coefficient l₂, M₂, N₂Is a constant and represents the contribution of each value to the actual interest level. These coefficients l₂, M₂, N₂The value of is l₂= M₂= 10, n₂= 1. The coefficient l₂, M₂, N₂The value of can also be estimated using statistical techniques. That is, the control unit 11 has a plurality of coefficients l₂, M₂, N₂Given the actual interest level IR (Di) for the set, the coefficient can be obtained by optimization.
[0155]
Next, document rearrangement performed based on the predicted interest level obtained using the actual interest level will be described with reference to FIG. Such document rearrangement is performed with the browser of FIG. 6 open.
[0156]
In step S111, the control unit 11 of the document processing apparatus sets a count value C of a counter that counts categories for classifying documents to 0. In step S112, the control unit 11 of the document processing apparatus calculates the degree of association between documents. That is, the control unit 11 is given an actual interest level for each unread document in the category indicated by the count value C among the uncategorized documents classified in step S23 in FIG. The degree of relevance between documents for each document is calculated. As described above, the actual interest level is given by the user's operation. The calculation of the degree of association between documents is performed based on the above-described index. Details of the calculation of the degree of association between documents will be described later.
[0157]
In step S113, the control unit 11 of the document processing apparatus calculates a predicted interest level. The predicted interest level is calculated based on the inter-document relevance level between the document and the document that has already been given the actual interest level. Therefore, the predicted interest level is calculated for a document for which no actual interest level is given.
[0158]
For one unread document in the category, the control unit 11 selects another document in the category having the maximum value of the relationship between documents calculated in step S112. The control unit 11 sets the actual interest level of the selected other document as the predicted interest level of one unread document. The control unit 11 stores the predicted interest level thus obtained in, for example, the RAM 14.
[0159]
In S118, the control unit 11 of the document processing apparatus branches the process depending on whether or not the calculation of the predicted interest level has been completed for all the documents in the category. The control unit 11 proceeds to step S114 as “YES” when the calculation is completed for all the documents in the category, and returns to step S112 as “NO” otherwise.
[0160]
In step S114, the control unit 11 of the document processing apparatus sorts unread documents for each category based on the predicted interest level calculated in step S113. As a document rearrangement method, a high priority order is given to unread documents with a high predicted interest level, and an unread document with a higher priority order can be arranged at the head of the unread document title array. . If there is no significant difference in the priority order, the received date and time is newer. The document titles are arranged in such an order for each category in the classification display sections 303, 304, and 305 of the document classification window 301, for example.
[0161]
In step S115, the control unit 11 of the document processing apparatus determines whether all categories have been completed. The control unit 11 advances the process to step S117 as “YES” when all categories are completed. The control unit 11 advances the process to step S116 as “NO” when all categories are not completed.
[0162]
In step S116, the control unit 11 of the document processing apparatus increases the counter value C for counting the category by 1. That is, the control unit 11 sets C = C + 1. And the control part 11 returns a process to step S112. In step S117, since it is determined that the processing for all categories has been completed in step S115, the control unit 11 displays the rearranged documents. Specifically, as shown in FIG. 6, a document icon and a document title are displayed. If there is no document title, a one sentence summary is displayed. Then, this series of steps is completed.
[0163]
Next, the calculation for calculating the degree of association between documents in step S112 in FIG. 25 will be described in detail with reference to FIG. The degree of relevance between documents is one document D_iAnd other documents D_jThe degree of relevance.
[0164]
In step S41, the control unit 11 of the document processing apparatus uses one document D._i A set of proper nouns included in the index of, and another document D already classified in the category specified in step S111 or S116 of FIG._jFor the set of proper nouns included in the index of P (D_i, D_j ). The control unit 11 then calculates the number P (D_i, D_j ) Is stored in the RAM 14, for example.
[0165]
In step S42, the control unit 11 of the document processing device refers to the meaning-of-sense relationship table shown in FIG._iMeaning and other documents included in the index of_jThe sum R of relations between meanings with meanings included in the index of R (D_i, D_j) Is calculated.
[0166]
In step S42, the control unit 11 of the document processing apparatus determines one unread document D._iFor words other than proper nouns, refer to the table of relevance between meanings and other documents D_jR (D_i , D_j) Is calculated. Then, the control unit 11 calculates the total sum R (D_i , D_j) Is stored in the RAM 14, for example.
[0167]
In step S43, the control unit 11 of the document processing apparatus uses one document D._i Other documents for_jRelevance between documents
Rel (D_i , D_j) = M_ThreeP (D_i, D_j ) + N_ThreeR (D_i, D_j )
It is defined as Where the coefficient m_Three, N_ThreeIs a constant and represents the degree of contribution of each value to the inter-document relevance. The control unit 11 determines the number P (D of common sets calculated in step S41._i , D_j ) And the total sum R (D_i , D_j) Is read from the RAM 14, for example, and is applied to the above-described formula to determine the inter-document relevance Rel (D_i , D_j ) Is calculated. These coefficients m_Three, N_ThreeFor example, m_Three= 10, n_Three= 1.
[0168]
Coefficient m_ThreeAnd n_ThreeThe value of can also be estimated using statistical techniques. That is, the control unit 11 includes a plurality of coefficients m_ThreeAnd n_ThreeInter-document relevance Rel (D_i , D_j) Can be obtained by optimization.
[0169]
Next, the recording medium 32 recorded / reproduced by the recording / reproducing unit 31 of the document processing apparatus will be described. A document processing program for processing a document having an internal structure by tagging from a plurality of elements is recorded on the recording medium. As the recording medium 32, for example, a floppy disk capable of recording / reproducing information is used.
[0170]
The recording medium 32 includes an actual interest level detection process for detecting the actual interest level for the document, and a priority order setting process for setting a priority order for the document based on the actual interest level detected by the actual interest level detection process. Furthermore, the recording medium 32 further includes a display process for displaying the document and an input process for receiving manual input for the document displayed in the display process. The actual interest level detection process is performed by the input unit. The actual interest level is detected based on.
[0171]
In the present embodiment, an example of a tagging method for a document has been shown, but it is needless to say that the present invention is not limited to this tagging method. In the present embodiment, the document is transmitted from the outside to the receiving unit 21 of the document processing apparatus, but the present invention is not limited to this. For example, the document may be written in the ROM 13 of the document processing apparatus or read from the recording medium 32 by the recording / reproducing unit 31.
[0172]
In the above-described embodiment, the mouse is exemplified as a device for selecting a desired element from the document displayed on the display unit 30 of the document processing apparatus. However, it goes without saying that the present invention is not limited to this. . Other devices such as a tablet and a light pen can be used to input elements in the document processing apparatus.
[0173]
Furthermore, in the above-described embodiment, Japanese and English sentences are exemplified, but it goes without saying that the present invention is not limited to these languages.
[0174]
【The invention's effect】
As described above, the present invention processes an electronic document, detects an actual interest level for the electronic document, and sets a priority order for the electronic document based on the detected actual interest level. The present invention also displays an electronic document, accepts manual input for the displayed electronic document, and detects the actual interest level based on this input. Therefore, according to the present invention, the priority of the electronic document is set reflecting the actual interest level of the user, so that the convenience of the user is achieved.
[0175]
Furthermore, according to the present invention, the priority level is set based on the predicted interest level, with the actual interest level of the document having the highest relevance among the electronic documents for which the actual interest level has already been obtained as the predicted interest level. Therefore, according to the present embodiment, a priority order can be given to a document to which no actual interest level is given.
[0176]
In the present invention, the electronic document is classified into a plurality of classification items, and the priority order is set for the electronic document for each classification item. Therefore, the present invention provides convenience to the user by setting a priority for each classification item.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration of a document processing apparatus to which an exemplary embodiment is applied.
FIG. 2 is a diagram showing an internal structure by tagging a document.
FIG. 3 is a diagram showing a window displaying an internal structure by tagging a document.
FIG. 4 is a flowchart showing an operation of a document processing apparatus to which the embodiment is applied.
FIG. 5 is a diagram illustrating a GUI for performing document classification before document classification;
FIG. 6 is a diagram showing a GUI for classifying documents.
FIG. 7 shows a table of classification models.
FIG. 8 is a flowchart for automatically classifying documents.
FIG. 9 is a flowchart for creating an index by finding document features;
FIG. 10 is a flowchart showing active diffusion.
FIG. 11 is a diagram for explaining active diffusion processing;
FIG. 12 is a flowchart of link processing for active diffusion.
FIG. 13 is a flowchart for calculating the degree of association between document classifications.
FIG. 14 is a flowchart of calculation of the degree of association between meanings.
FIG. 15 is a diagram showing a table of relevance levels between meanings.
FIG. 16 is a flowchart for browsing and sorting documents.
FIG. 17 is a flowchart showing a series of steps for increasing the importance of an arbitrary part of a sentence.
FIG. 18 shows a summary window.
FIG. 19 is a diagram showing a state in which a word is selected in the summary window.
FIG. 20 is a diagram illustrating a state where a region selected in the summary window is further clicked.
FIG. 21 is a diagram showing a state in which a summary is displayed in a summary window.
FIG. 22 is a diagram showing details of the summary creation process.
FIG. 23 is a diagram for explaining the calculation of the actual interest level from the maximum appearance position of the selected element.
FIG. 24 is a diagram for explaining the calculation of the actual interest level from the maximum size of the summary element and the ratio of the entire document.
FIG. 25 is a flowchart for automatically classifying a document based on a predicted interest level.
FIG. 26 is a flowchart for calculating the degree of association between documents.
[Explanation of symbols]
10 main body, 11 control unit, 12 interface, 13 CPU, 20 input unit, 21 receiving unit, 30 display unit, 31 recording / reproducing unit

Claims (6)

In a document processing method of a document processing apparatus that processes a plurality of electronic documents,
A receiving step in which the receiving means receives a plurality of electronic documents;
A recording step for recording a plurality of electronic documents received in the receiving step;
A display step for displaying an electronic document selected by the user among the plurality of electronic documents recorded in the recording step and a summary of the electronic document;
An input step in which the input means inputs user operation information for the electronic document displayed in the display step;
The actual interest level that the actual interest level detection means calculates the actual interest level of the user for the electronic document based on the operation information of the user input in the input step for the electronic document displayed in the display step A detection process;
The priority order setting means has a relevance level based on the internal structure of the electronic document among the electronic documents for which the actual interest level is calculated in the actual interest level detection step with respect to the electronic document for which the actual interest level is not calculated. A priority setting step in which the actual interest level of the highest electronic document is set as a predicted interest level, and a priority order is set based on the predicted interest level ;
A sorting step in which sorting means sorts electronic documents not selected by the user among the plurality of electronic documents recorded in the recording step according to the priority set in the priority setting step; Have
In the actual interest level detection step, for each display area for displaying the electronic document and the summary of the electronic document, the appearance position in the electronic document among the elements of the electronic document selected by the user is the electronic document. The first actual interest level element composed of the ratio between the appearance position of the element farthest from the head of the document and the size of the electronic document, the number of keywords specified by the user, and the number of elements selected by the user A third actual interest level comprising a second actual interest level element of the actual interest level and a ratio of the size of the display area of the summary of the electronic document displayed in the display step and the size of the display area of the electronic document. A document processing method for calculating the actual interest level using an interest level element . An index creating means for extracting features based on the internal structure of the plurality of electronic documents received in the receiving step, and creating an index of the extracted features;
An index recording step in which the index recording means records the index created in the index creation step;
The index update means performs an update process for recording the user's operation history based on the user's operation information input in the input step and / or the actual interest level calculated in the actual interest level detection step in the index. The document processing method according to claim 1, further comprising an index update step. The electronic document, the document processing method according to claim 1, wherein the internal structure that is described by the attribute information. The classification means further comprises a classification step of classifying the electronic document into a plurality of classification items;
Above the priority setting step, the priority setting unit, the document processing method according to claim 1, wherein to set the priority of the electronic document for each classified category items by the classifying means in the classification step. In a document processing apparatus that processes a plurality of electronic documents,
Receiving means for receiving a plurality of electronic documents;
Recording means for recording a plurality of electronic documents received by the receiving means;
Display means for displaying an electronic document selected by the user among the plurality of electronic documents recorded by the recording means and a summary of the electronic document;
Input means for inputting user operation information for the electronic document displayed by the display means;
Real interest level detection means for calculating the actual interest level of the user for the electronic document based on the user operation information input by the input means for the electronic document displayed by the display means;
The actual interest of the electronic document having the highest relevance based on the internal structure of the electronic document among the electronic documents of which the actual interest level is calculated by the actual interest level detection unit with respect to the electronic document for which the actual interest level is not calculated A priority setting means for setting a priority based on the predicted interest level,
Reordering means for reordering electronic documents not selected by the user among the plurality of electronic documents recorded by the recording means according to the priority set by the priority setting means;
The actual interest level detection means is configured such that, for each display area for displaying the electronic document and the summary of the electronic document, the appearance position in the electronic document among the elements of the electronic document selected by the user is the electronic document. The first actual interest level element composed of the ratio between the appearance position of the element farthest from the head of the document and the size of the electronic document, the number of keywords specified by the user, and the number of elements selected by the user A third actual interest level comprising a second actual interest level element of the actual interest level, and a ratio of the size of the display area of the summary of the electronic document displayed by the display means and the size of the display area of the electronic document. A document processing apparatus that calculates the actual interest level using a degree element . In a computer-readable recording medium recorded with a document processing program for causing a computer to execute document processing for processing a plurality of electronic documents,
The above document processing program
A receiving step for receiving a plurality of electronic documents;
A recording step for recording a plurality of electronic documents received in the receiving step;
A display step for displaying an electronic document selected by the user among the plurality of electronic documents recorded in the recording step and a summary of the electronic document;
An input step in which the input means inputs user operation information for the electronic document displayed in the display step;
The actual interest level that the actual interest level detection means calculates the actual interest level of the user for the electronic document based on the operation information of the user input in the input step for the electronic document displayed in the display step A detection process;
The priority order setting means has a relevance level based on the internal structure of the electronic document among the electronic documents for which the actual interest level is calculated in the actual interest level detection step with respect to the electronic document for which the actual interest level is not calculated. A priority setting step in which the actual interest level of the highest electronic document is set as a predicted interest level, and a priority order is set based on the predicted interest level;
A sorting step in which sorting means sorts electronic documents not selected by the user among the plurality of electronic documents recorded in the recording step according to the priority set in the priority setting step; Let the computer run,
In the actual interest level detection step, for each display area for displaying the electronic document and the summary of the electronic document, the appearance position in the electronic document among the elements of the electronic document selected by the user is the electronic document. The first actual interest level element composed of the ratio between the appearance position of the element farthest from the head of the document and the size of the electronic document, the number of keywords specified by the user, and the number of elements selected by the user A third actual interest level comprising a second actual interest level element of the actual interest level and a ratio of the size of the display area of the summary of the electronic document displayed in the display step and the size of the display area of the electronic document. A recording medium on which a document processing program for calculating the actual interest level using an interest level element is recorded .

Images