JPH0748181B2 - How to edit text knowledge in the knowledge organization editor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION <Industrial field of application> The present invention relates to a text knowledge organization support function, and more specifically, knowledge acquisition work from experts, which is said to be the bottleneck in the construction of expert systems. At
In general problem solving and organizing work represented by the KJ method, etc., even experts who are not familiar with computers and persons involved in the work can easily give their knowledge to the computer based on the sentences that they use everyday. The present invention relates to a text knowledge editing method in the knowledge organizing editor.

Areas of use include background knowledge and domain knowledge when constructing an expert system related to AI (artificial intelligence), information organization, problem solving, idea support for general individuals (any of these materials are acceptable), and general text efficiency. Areas that require specific work such as creation and management.

<Prior art> When structuring and organizing unstructured sentence knowledge and managing it, it is easy to operate while visually appealing the conceptual items represented by the sentence and related conceptual items and sentences. An information organization support function that can be done is required.

In response to such needs, the number of text processing software that can structurally manage and edit texts on a computer is increasing recently.

<Problems to be solved by the invention> However, when unstructured knowledge obtained by interviews with experts is made into a knowledge base, or when a method of organization such as KJ method is implemented with consideration for problem solving and decision making. In addition to structuring and editing sentences, consider links between sentences and intermediate knowledge bases such as referring to domain knowledge elements pointed to by sentences, or conversely referring to sentence knowledge from the domain knowledge base. The required operation is required.

In the existing knowledge organization tools, the part corresponding to the intermediate knowledge base is weak, and many focus only on the management and operation of sentences.

For this reason, it is difficult to perform structured work while verifying the causal relationship between sentences, or to support classification of new sentences.

Therefore, an object of the present invention is to provide a text knowledge editing method in a knowledge organizing editor capable of performing a structured work while verifying a causal relationship between texts and assisting classification of new texts. .

<Means for Solving the Problem> In order to achieve the above object, the text knowledge editing method according to the present invention is configured so that each text and each graphic input from the input means (1) are
Under the control of the control means (13), the structured document and the graphic information are edited by the editing means (5), the edited result is stored in the storage means (8), and the classified specific information is displayed in the display means (2). A set of sentences belonging to a concept is displayed together with its classification item name (see Fig. 2), and the meaning content represented by each sentence is arranged by nesting cards of various superordinate or subordinate concepts, and connecting these cards with arcs. In a knowledge organization editor that displays as a graphical net (see Fig. 4) and edits and organizes text groups according to their meaning,
When a sentence is input by the text editor of the editing means (5), the sentence number, the sentence, the classification item name which is the card name of the higher-rank or lower-level concept to which the sentence belongs, and the word in the sentence are displayed on the display means (2). A structured document format (see FIG. 3 (b)) for inputting the card name of the concept to which the word belongs and the card name related to the sentence is displayed, and the data entered according to this format (see From FIG. 3 (c)), the above-mentioned graphical net is created for the card in the data (see FIG. 7), while the graphic editor of the above-mentioned editing means (5) allows a card specific to the above-mentioned display means at the time of graphic input. Display the above graphical net (Fig. 4) and change the mutual relationship of cards, arc replacement, car The graphical net is updated based on the operation of adding or deleting (see FIG. 6), and by the control means (13), based on the structured document and the graphical net, cards, nests, arcs, sentences and sentence sets. A basic data structure (see FIG. 5) with a bidirectional reference pointer is generated and stored in the storage means (8), and for a specific word in the sentence displayed on the display means (2), The text editor (5) calls a graphical net including a card having the word as an item name based on the basic data structure and displays it on the display means (2) (see FIG. 9), while the display means ( The specific classification items in the graphical net displayed in 2) are related by the graphic editor (5) based on the basic data structure. The sentence set is called and displayed on the display means (2), and the basic data structure is changed in accordance with the change regarding the classification item made on the displayed sentence set through the input means (1) (see FIG. 8). At the same time, for the specific word in the sentence newly input from the input unit (1) or the already stored sentence, the dependency relation of the word is obtained by the input sentence analysis unit (11), and then the text editor ( According to 5), the display means (2) for displaying a particle, a verb, and a noun, which assists in creating a sentence expressing the word itself or upper, lower, and related items by referring to the basic data structure.
The above sentence is displayed to support the sentence creation (see FIGS. 11 and 12), and the sentence newly input from the input means (1) is cut out by the input sentence analysis means (11).
Alternatively, after obtaining the dependency relation between words, the knowledge base search means (12) performs matching of the cut-out words with reference to the above basic data structure or frame matching according to the obtained dependency relation, and the classification is performed. The present invention is characterized in that existing classification items that are candidates are displayed on the display means to support sentence classification (see FIGS. 13 and 14).

<Operation> When the user inputs a text, the text editor of the editing means (5) displays the text number, the text, and the classification item name, which is the card name of the superordinate or subordinate concept to which the text belongs, on the display means (2).
The word in the sentence and the card name of the concept to which the word belongs,
A structured document format (see FIG. 3 (b)) for inputting the card name for each sentence is displayed, and from the data entered by the user according to this format (see FIG. 3 (c)), Graphical net for each card in the data, that is, a diagram in which cards of various upper and lower concepts are arranged in a nested manner for the meaning content represented by the sentence, and these cards are connected by arcs (see Fig. 4)
Is created (see FIG. 7).

Further, when the user inputs a figure, the graphic editor of the editing means (5) causes the display means (2) to display a graphical net relating to a specific card, and the graphical net relating to the specific card is displayed on the graphical net. The graphical net is updated based on operations such as system change, arc replacement, and card addition / deletion (see FIG. 6).

Then, the control means (13) generates a basic data structure (see FIG. 5) in which a bidirectional reference pointer is put on a card, a nest, a sentence and a sentence group based on the structured document and the graphical net, This is stored in the storage means (8).

After the above preparation process is completed, the user inputs the input means (1).
With respect to the word in the sentence specified by using the display means (2), the text editor (5) creates a graphical net including a card having the word as an item name based on the basic data structure (see FIG. 5). It is called from the storage means (8) and displayed on the display means (2). The graphic editor (5) relates the classification items in the graphical net specified by the user using the input means (1) and the display means (2) based on the basic data structure (see FIG. 5). The basic data structure described above is displayed according to the change in the classification item made by the user via the input means (1) in the displayed text set by displaying the text set to be displayed on the display means (2) by calling from the storage means (8). (See FIG. 8).

On the other hand, when the sentence creation support is required, the input sentence analysis means (11) is used for the sentence newly input by the user through the input means (1) or the word specified in the already stored sentence. First, the dependency relation of the word is obtained, and then the text editor (5) refers to the basic data structure (see FIG. 5) to express the word itself or each item of upper, lower, and relation. A particle, a verb, and a noun, which are useful in creating a sentence, are displayed on the display means (2) to assist the user in creating a sentence (see FIGS. 11 and 12).

Further, when it is necessary to support the sentence classification, the input sentence analysis means (11) first cuts out a word or a mutual relation between words for a sentence newly input by the user from the input means (1). Then, the knowledge base search means (12) performs matching of the words cut out by referring to the basic data structure (see FIG. 5) or frame matching according to the obtained dependency relationship, and the classification is performed. The existing classification items that are candidates are displayed on the display means (2) to support the user's sentence classification (see FIGS. 13 and 14).

The above series of actions can be summarized and itemized in order as follows.

It acts so that the classification item of the text set is changed in accordance with the knowledge structure edited on the graphical net editing screen. Also, if the text set is classified and related from the beginning, a graphical net corresponding to it is created.

At the time of editing the graphical net, a sentence set belonging to each item that is an expression element of the graphical net is called, and the attribute to each item set can be corrected / changed.

It works so that you can make a structured definition of the text itself. It also acts to call the related item and the related sentence from the sentence.

By using the relation between items of the graphical net, the sentence and the sentence are combined according to the relation so as to support the creation of a new sentence.

It acts to assist classification of newly input sentences by referring to the knowledge of the already created graphical net.

<Example> An example of the text knowledge editing method of the present invention will be described below with reference to the drawings.

FIG. 1 is an overall configuration diagram showing an example of a system required to carry out this method.

The input means 1 is a peripheral device of a computer for inputting characters or figures, that is, a keyboard, a mouse, or the like.

The display means 2 is a computer display device capable of graphic display by a bitmap method, and an OS capable of displaying a plurality of screens in the same screen such as a window system.
The software includes a screen switching means 3 and a graphic display means 4.

The editing means 5 is a graphic editor provided by the software environment of the computer (one capable of editing rectangles and line segments).
Also, it is software based on a text editor, and in addition to the basic function of editing, it has an object search means 6 for fixing the editing target and an object changing means 7 for actually operating the fixed target.

The storage means 8 is an object storage means 9 for storing a graphical net (see FIGS. 4 and 5) and a structured document (see FIG. 3 (c)) converted into an internal structure in an on-memory form, and a disk. And a file storage means 10 for storing in the form of a file, and the storage device of the computer corresponds to this.

The input sentence analysis means 11 is software that is composed of a morpheme analysis means and a syntactic analysis means and that is used to obtain a dependency relationship of a sentence at the word level. The knowledge base search means 12 is software that searches the contents of the knowledge structure represented by the graphical net and matches the input items.

FIG. 2 shows an example of the screen specifications of this method.

This screen is a graphical net (an example is shown in Fig. 4)
Screen for editing (graphic editor) (here is the screen showing "individual constraints") and screen for editing the document corresponding to this graphic editor (text editor) (here is "constraints regarding visit") At least a screen showing the corresponding sentence set) and a screen for displaying related items of the graphical net and the sentence set (here, a screen showing an “item list”) are required. Here, cards, nesting, arrows (arcs) for associating them, and text are taken up as objects to be edited by the graphic editor. FIG. 2 shows that a set of sentences belonging to the card is called from the card of “Restrictions on visit” of the graphic editor.

FIG. 3A shows an example of a text set, FIG. 3B shows an example of an input format of a structured text, and FIG. 3C shows an example of a structured text input according to the format. A structured sentence is a set of structured sentences managed by sentence numbers (Fig. 3 (b)-). Structured text is the text itself (3rd
Figure (b)-), classification item list (Figure 3 (b)-
), A componentized character string list (Fig. 3 (b)-), and a related item list (Fig. 3 (b)-). The item in the graphical net to which it belongs can be specified by the classification item, the item in the graphical net represented by the term that appears in the sentence can be specified by the componentized character string, and the arrow in the graphical net can be specified by the related item. The item at the destination of the arrow can be specified as an item related to the sentence. On the contrary, the structured text shown in FIG. 3 (c) is the text shown in FIG. 4 in the card item "restriction on visit" in the graphical net.

-Means for realizing action As a result of editing the graphical net, the classification item list and the related item list of the structured sentence set are automatically rewritten by internal processing. On the contrary, it is possible to synthesize a graphical net consisting of cards and nests from structured sentences in which classification items are designated by text editing. The related item list can be changed for each sentence, and the change is not reflected on the graphical side on the sentence side. The componentized character string (see FIG. 3 (c)) is defined by editing on the text side.

In order to enable the above-mentioned operations, it is necessary to provide a data structure in which a bidirectional reference pointer is set for the object and the text set to be edited on the graphical net. An example of the data structure is shown in FIG. If you take such a data structure, for example,
A specific nested parent can be found by the "super" link and a child can be found by the "children" link. You can follow links many times to find offspring and ancestors. Further, it is possible to obtain an arrow (arc) to and from itself by the "arcs" link, and obtain an item as a destination of the arrow (arc) by the "source" or "target" link of the arrow (arc). The "sentences" link links to structured sentences, and the "clas"
You can find out which card or nest you belong to by the "s" link. The "arcs" and "termdef" links in the structured text refer to the related arrow (arc), card or nest item, respectively.

After preparing the data structure as described above, it is possible to realize the action by constructing a control program that traces necessary data links and responds to changes at the time of editing.
Figures 6, 7 and 8 summarize the implementation procedure.

-Means for realizing the action The structured document may be input using an ordinary text editor, for example, according to the format shown in Fig. 3 (b). If it is formatted, it is easy to internally convert it to the basic data structure shown in FIG.

FIG. 9 is an example of an item call from a term in a sentence. In a sentence in which a partized character string is defined, it is highlighted when the cursor comes to each defined word, and when there is input from the keyboard or mouse, a card (or nest) with that word as the item name is displayed in a separate window. The containing graphical net is displayed. If you want to know more detailed information about the term in text, you can call the text from the item of Graphical Net.

The means for realizing these operations are basically the means for tracing the links between the above-mentioned data, except for the display-related programs. FIG. 10 shows an example of the procedure.

・ Means for realizing action FIG. 11 shows an example of text creation support from a graphical net. This is an example in which the partial knowledge represented by the item in the time zone related to the item immediately before and after in the graphical net in FIG. 9 is written. Here, the item name, the relation name, the expression peculiar to the relation (for example, in the case of a parent-child relation, "the content of" is "belongs to", etc.), etc.), the case particles and auxiliary verbs that are often used are the menu screens as sentence parts It is prepared in advance, and the user is presented with it so that the user can select it if necessary so as to assist in the sentence creation. An example of this function is shown in FIG. To realize this, an object related to a specific item may be obtained by a pointer search, and a list of corresponding parts text expressions may be created and presented as a selection menu.

・ Means for realizing action FIG. 13 shows an example of a procedure for the classification support function.

When mechanically supporting classification, the basic is word-level matching. This requires a morphological processing operation to divide the input sentence into words. The item of the existing graphical net that is matched by simple matching in word units becomes one candidate for classification. It is expected that there will be no match at this level, or that multiple classification candidates will appear. As a measure against the former problem, there is a method of increasing the number of matching words by providing a synonym pointer in the item of the graphical net. As a measure against the latter problem, there are two possible methods, namely, a method of presenting candidates and having the user select one, and a method of matching at the level of converting a sentence into an internal frame format.

With this frame matching method (in this case, we will proceed assuming that the internal format of a sentence is a frame, but any form can be used as long as the definition of the internal format is clear), the existing classified sentences are converted into a frame format. Or, it is necessary to prepare a frame structure that serves as a template for each item.

The morphologically processed input sentence is subjected to a syntax analysis process and a dependency extraction process, and then mapped to a frame format. The mapping replaces each element of the sentence with a concept term that is specific to the internal form, and provides a wider range of matching than in the case of word units. As a result, if there is existing frame data that is completely included, the input knowledge is absorbed in the existing graphical net. If some of the slot values are undefined, contact the user and check if it is possible to absorb by defining a synonym.If it is possible, newly define a synonym definition. Done and add synonym knowledge of graphical net. If it cannot be absorbed or if only a part of the frame matches, the classification candidate is presented and the addition of new knowledge is entrusted to the user. Example 14 of actual input sentence classification according to the above procedure
Shown in the figure.

<Effects of the Invention> The following effects can be obtained by the text knowledge editing method of the present invention.

1) You can edit sentences while viewing the entire structure of knowledge.

2) Knowledge extraction and organization from book documents can be performed easily.

3) Since the knowledge structure represented by the term in the sentence can be called and another related sentence or term can be called, the knowledge represented by the sentence can be easily grasped.

4) Since it is possible to support the creation of a new sentence from the graphical relational expression between items, the sentence creation becomes easy.

[Brief description of drawings]

FIG. 1 is a diagram showing an example of an overall configuration of a system for realizing an embodiment of the present invention, FIG. 2 is a diagram showing an example of specifications of a computer display screen in this embodiment, and FIG. FIG. 4 is a diagram showing a definition example of a structured sentence in the example and an actual example according to the definition, FIG. 4 is a diagram showing an example of a graphical net in the present embodiment, and FIG. 5 is an edit necessary for realizing the present embodiment. FIG. 6 is a diagram showing an example of a basic data structure of an object, FIG. 6 is a diagram showing an example of a procedure for associating a graphical net with a structured document in this embodiment, and FIG. 7 is a graphical representation from a structured document in this embodiment. FIG. 8 is a diagram showing an example of a procedure for creating a net, FIG. 8 is a diagram showing an example of a procedure for calling a sentence set belonging to an item of a graphical net in the present embodiment, and FIG. 9 is a diagram showing a structured sentence in the present embodiment. Term? Graphical net and a diagram showing an example of calling a related document, FIG. 10 is a diagram showing an example of a procedure for calling a structured definition of text and related items / documents in this embodiment, and FIG. 11 is a graphical net in this embodiment. FIG. 12 is a diagram showing an example of the document creation support function from FIG. 12, FIG. 12 is a diagram showing an example of the document creation support function in this embodiment, FIG. 13 is a diagram showing an example of the procedure of the classification support in this embodiment, and FIG. It is a figure which shows an example of the classification assistance of the input sentence in this implementation line.

Claims (1)

[Claims] 1. A text and graphic input from an input means are edited into a structured document and graphic information by an editing means under the control of a control means, and the edited result is stored in a storage means and displayed on a display means. , Display a set of sentences that belong to a specific classified concept, along with the name of the classification item, and place the cards of various higher-level or lower-level concepts in a nested manner, and connect these cards with arcs. In a knowledge organization editor that displays as a graphical net and edits and organizes a group of sentences according to their meanings, when a sentence is input by the text editor of the above-mentioned editing means, the sentence number, the sentence, and the sentence are displayed on the above-mentioned display means. Classification item name that is a card name of the above-mentioned or lower-level concept to which it belongs, a word in the sentence and a card name of the concept to which the word belongs, and a car related to the sentence The format for the structured document for inputting each command name is displayed, and the above-mentioned graphical net is created for each card in the data from the data input according to this format, while the graphic editor of the above-mentioned editing means At the time of input, the above-mentioned display means displays the above-mentioned graphical net for a specific card, and the graphical net is updated on this graphical net based on the operations of changing the mutual relationship of cards, replacing arcs, and adding and deleting cards. The control means generates a basic data structure in which a bidirectional reference pointer is attached to a card, a nest, an arc, a text, and a text set based on the structured document and the graphical net, and stores the basic data structure in the storage means. About a specific word in the text displayed on the display , The text editor calls a graphical net including a card having the word as an item name based on the basic data structure and displays it on the display means, while a specific classification item in the graphical net displayed on the display means Regarding the above, the graphic editor calls the related text set based on the basic data structure and displays the text set on the display means, and the basic data according to the change regarding the classification item made on the displayed text set via the input means. While changing the structure, regarding the specific word in the sentence newly input from the input means or the already stored sentence, after the dependency relation of the word is obtained by the input sentence analysis means, by the text editor, Refer to the basic data structure and refer to the above words themselves or higher or lower , The particles, verbs, and nouns that assist in creating sentences expressing each item of the relationship are displayed on the display means to assist the sentence creation, and the above-mentioned input of the text newly input from the input means is performed. After the words are cut out by the sentence analysis means or the dependency relationships between the words are obtained, the knowledge base search means refers to the above basic data structure to match the extracted words or follow the obtained dependency relationships. A text knowledge editing method characterized by performing frame matching and displaying existing classification items as classification candidates on the display means to support text classification.