CN110321534A - A kind of method for editing text, device, equipment and readable storage medium storing program for executing - Google Patents
A kind of method for editing text, device, equipment and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN110321534A CN110321534A CN201810262255.3A CN201810262255A CN110321534A CN 110321534 A CN110321534 A CN 110321534A CN 201810262255 A CN201810262255 A CN 201810262255A CN 110321534 A CN110321534 A CN 110321534A
- Authority
- CN
- China
- Prior art keywords
- word
- user
- edit commands
- edit
- participle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract 14
- 239000013598 vector Substances 0.000 claims 21
- 230000009466 transformation Effects 0.000 claims 9
- 238000004590 computer program Methods 0.000 claims 2
- 238000013518 transcription Methods 0.000 claims 2
- 230000035897 transcription Effects 0.000 claims 2
- 238000013501 data transformation Methods 0.000 claims 1
- 235000013399 edible fruits Nutrition 0.000 claims 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/166—Editing, e.g. inserting or deleting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Document Processing Apparatus (AREA)
- Machine Translation (AREA)
Abstract
This application discloses a kind of method for editing text, device, equipment and readable storage medium storing program for executing, the application gets urtext data and user's edit commands to be edited, determine the corresponding edit operation of the edit commands, and according to urtext data, the semantic dependency for each word for being included with user's edit commands, determine that command object word edits the command object word in urtext data finally according to edit operation from user's edit commands.Application scheme user only needs input editing order, and the process edited automatically to urtext data can be realized, greatly reduce manual operation, improve editorial efficiency.And, the application has comprehensively considered urtext data and user's edit commands when determining command object word include the semantic dependency of each word, the accuracy that command object word determines is substantially increased, more accurately can complete entire text editing process according to the wish of user.
Description
Technical field
This application involves natural language processing technique field, more specifically to a kind of method for editing text, device,
Equipment and readable storage medium storing program for executing.
Background technique
What text editing referred to, typesetting, the editor of text are carried out to urtext.Traditional method for editing text is all borrowed
Keyboard and mouse is helped to realize text editing.This mode can undoubtedly occupy the both hands of user, and editing process takes time and effort.
With the arrival in artificial intelligence epoch, more and more people wish to realize by automatic method to text data
Editor, with improve editorial efficiency and reduce manual operation.
Summary of the invention
In view of this, this application provides a kind of method for editing text, device, equipment and readable storage medium storing program for executing, to realize
Editing process manual operation is reduced, editorial efficiency is improved.
To achieve the goals above, it is proposed that scheme it is as follows:
A kind of method for editing text, comprising:
Obtain urtext data and user's edit commands to be edited；
Determine the corresponding edit operation of user's edit commands；
According to the urtext data, the semantic dependency for each word for being included with user's edit commands, from institute
It states and determines command object word in user's edit commands；
According to the edit operation, the command object word in the urtext data is edited.
Preferably, the semanteme of each word for according to the urtext data, being included with user's edit commands
Correlation determines command object word from user's edit commands, comprising:
According to the semantic dependency of each word and the urtext data in user's edit commands, equivalent is determined
Correlative character；
The correlative character of each word in model and user's edit commands is determined using order word trained in advance,
Command object word is determined from each word that user's edit commands includes；
The order word determines that training sample when model pre-training includes: that user corresponding with training text data edits
Order includes the term vector of training word, and is determined according to the trained word and the semantic dependency of the training text data
Training word correlative character；Sample label include: the trained word whether be command object word annotation results.
Preferably, described related to the semanteme of the urtext data according to each word in user's edit commands
Property, determine the correlative character of equivalent, comprising:
To user's edit commands and the urtext data segment and term vector, are respectively contained
The term vector of participle；
Determine that the i member entry for the participle that user's edit commands and the urtext data respectively contain, i take respectively
It is worth [1, N], N is setting constant；
According to the i member entry for the participle that user's edit commands includes, with each participle in the urtext data
I member entry match condition, determine the correlative character for the participle that user's edit commands includes.
Preferably, the i member entry of the participle for including according to user's edit commands, with the urtext data
In each participle i member entry match condition, determine the correlative character for the participle that user's edit commands includes, wrap
It includes:
The i member entry for calculating the participle that user's edit commands includes, with each participle in the urtext data
I member entry matching score；
According to the size relation of the matching score and setting matching score threshold value, determine that user's edit commands includes
Participle i member entry, the coverage in the urtext data；
According to the matching score and/or the coverage, the correlation for the participle that user's edit commands includes is determined
Property feature.
Preferably, described to determine each word in model and user's edit commands using order word trained in advance
Correlative character determines command object word from each word that user's edit commands includes, comprising:
The first input layer that model is determined by the order word inputs the term vector of each word in user's edit commands
And correlative character；
The first hidden layer set that model is determined by the order word, converts the correlative character of each word, obtains
Correlative character after the transformation of each word；
The output layer that model is determined by the order word, it is determining and defeated according to correlative character after the transformation of each word
Command object word in user's edit commands out.
Preferably, further includes:
The second input layer that model is determined by the order word, input the word of each word in the urtext data to
Amount；
The second hidden layer set that model is determined by the order word, to the term vector of each word in the urtext data
It is converted, obtains term vector after the transformation of each word；
The coding layer that model is determined by the order word, by k-th of word in pre-set user edit commands to urtext
The attention rate weight of each word in data obtains k-th of word in urtext data multiplied by term vector after the transformation of equivalent
The attention rate weight vectors of each word, k value [1-m], m are the word sum that user's edit commands includes；
The coding layer that model is determined by the order word, by k-th of word to the pass of each word in urtext data
Note degree weight vectors are added with correlative character after the transformation of k-th of word of the first hidden layer set output, and addition result is made
For the input of the output layer, so that the output layer is determining according to the addition result and exports in user's edit commands
Command object word.
Preferably, the process of user's edit commands is obtained, comprising:
User's edit commands of speech form is obtained, and transcription is user's edit commands of textual form；
Or,
Obtain user's edit commands of textual form.
A kind of text editing apparatus, comprising:
User's edit commands acquiring unit, for obtaining urtext data and user's edit commands to be edited；
Edit operation determination unit, for determining the corresponding edit operation of user's edit commands；
Command object word determination unit, for being included with user's edit commands according to the urtext data
Each word semantic dependency, from user's edit commands determine command object word；
Editing and processing unit is used for according to the edit operation, to the command object in the urtext data
Word is edited.
Preferably, the command object word determination unit includes:
Correlative character determination unit, for according to each word in user's edit commands and the urtext data
Semantic dependency, determine the correlative character of equivalent；
Model prediction unit, for being determined in model and user's edit commands using order word trained in advance
The correlative character of each word determines command object word from each word that user's edit commands includes；
The order word determines that training sample when model pre-training includes: that user corresponding with training text data edits
Order includes the term vector of training word, and is determined according to the trained word and the semantic dependency of the training text data
Training word correlative character；Sample label include: the trained word whether be command object word annotation results.
Preferably, the correlative character determination unit includes:
Term vector determination unit, for user's edit commands and the urtext data are segmented and word to
Quantization, the term vector of the participle respectively contained；
I member entry determination unit, for determining that user's edit commands and the urtext data are respectively wrapped respectively
The i member entry of the participle contained, i value [1, N], N are setting constant；
Matching score processing unit, the i member entry of the participle for including according to user's edit commands, with the original
The match condition of the i member entry of each participle, determines the correlation for the participle that user's edit commands includes in beginning text data
Property feature.
Preferably, the matching score processing unit includes:
Matching score computing unit, for calculating the i member entry for the participle that user's edit commands includes, with the original
The match condition of the i member entry of each participle in beginning text data；
Coverage determination unit is determined for the size relation according to the matching score and setting matching score threshold value
The i member entry for the participle that user's edit commands includes, the coverage in the urtext data；
Integrated treatment unit, for determining user's edit commands according to the matching score and/or the coverage
The correlative character for the participle for including.
Preferably, the model prediction unit includes:
First input unit inputs the user and edits for determining the first input layer of model by the order word
The term vector and correlative character of each word in order；
Fisrt feature converter unit, for determining the first hidden layer set of model by the order word, to the phase of each word
Closing property feature is converted, and correlative character after the transformation of each word is obtained；
Output unit, for determining the output layer of model by the order word, according to related after the transformation of each word
Property feature it is determining and export the command object word in user's edit commands.
Preferably, the model prediction unit further include:
Second input unit inputs the urtext for determining the second input layer of model by the order word
The term vector of each word in data；
Second feature converter unit, for determining the second hidden layer set of model by the order word, to described original
The term vector of each word is converted in text data, obtains term vector after the transformation of each word；
Coding unit will be k-th in pre-set user edit commands for determining the coding layer of model by the order word
Word obtains k-th of word to original multiplied by term vector after the transformation of equivalent to the attention rate weight of each word in urtext data
The attention rate weight vectors of each word in beginning text data, k value [1-m], m are the word sum that user's edit commands includes；By institute
State k-th word of k-th of word to the attention rate weight vectors of each word in urtext data, with the first hidden layer set output
Transformation after correlative character be added, input of the addition result as the output layer, so that the output layer is according to the phase
Add result determining and exports the command object word in user's edit commands.
Preferably, user's edit commands acquiring unit includes:
First user's edit commands obtains subelement, and for obtaining user's edit commands of speech form, and transcription is text
User's edit commands of this form；
Or,
Second user edit commands obtains subelement, for obtaining user's edit commands of textual form.
A kind of text editing apparatus, comprising: memory and processor；
The memory, for storing program；
The processor realizes each of method for editing text disclosed above for executing described program
Step.
A kind of readable storage medium storing program for executing is stored thereon with computer program, real when the computer program is executed by processor
Now each step of method for editing text disclosed above.
It can be seen from the above technical scheme that method for editing text provided by the embodiments of the present application, gets to be edited
Urtext data and user's edit commands, determine the corresponding edit operation of the edit commands, and according to urtext number
According to the semantic dependency for each word for being included with user's edit commands determines command object word, finally from user's edit commands
According to edit operation, the command object word in urtext data is edited.Application scheme user only needs to input volume
Order is collected, the process edited automatically to urtext data can be realized, greatly reduce manual operation, improve editor
Efficiency.
Also, the application has comprehensively considered urtext data with user's edit commands when determining command object word
The semantic dependency of each word substantially increases the accuracy that command object word determines, can be more accurately according to the wish of user
Complete entire text editing process.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is a kind of method for editing text flow chart disclosed in the embodiment of the present application；
Fig. 2 is a kind of determining command object word method flow diagram disclosed in the embodiment of the present application；
Fig. 3 is a kind of correlative character method flow diagram of determining word disclosed in the embodiment of the present application；
Fig. 4 is that a kind of determined according to matching score disclosed in the embodiment of the present application segments correlative character method flow diagram；
Fig. 5 is that a kind of exemplary order word of the embodiment of the present application determines model structure schematic diagram；
Fig. 6 determines that model determines command object word method flow based on order word for one kind disclosed in the embodiment of the present application
Figure；
Fig. 7 is that the exemplary another order word of the embodiment of the present application determines model structure schematic diagram；
Fig. 8 determines that model determines command object word method flow based on order word for another kind disclosed in the embodiment of the present application
Figure；
Fig. 9 is a kind of text editing apparatus structural schematic diagram disclosed in the embodiment of the present application；
Figure 10 is a kind of hardware block diagram of server disclosed in the embodiment of the present application.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on
Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall in the protection scope of this application.
In order to solve it is traditional carried out present in the mode of text editing by keyboard, mouse, need a large amount of artificial behaviour
Make, and the problem that editorial efficiency is low, inventor wishes to provide a kind of new text editing by artificial intelligence technology
Scheme.It is expected that the target reached is, it is only necessary to which user's input editing order, machine can be voluntarily to urtext numbers to be edited
According to progress edit operation.To greatly reduce manual operation, editorial efficiency is improved.
Further, user can may further liberate user's both hands by speech form input editing order in this way.
During inventor is the study found that realize text editing by artificial intelligence, a vital link
Command object word is exactly recognized accurately, command object word is the object of the wanted edit operation of user.
The solution that inventor expects first is:
A large amount of user's edit commands is collected, and manually marks command object word therein.Using being labeled with command object
Neural network model is trained in user's edit commands of word, the neural network model after being trained.Subsequent use process, can be with
The command object word in user's edit commands is predicted using the neural network model.
But by practical application, inventor has found that this mode is asked there are command object word prediction accuracy is not high
Topic.Discovery of tracing it to its cause is chased after, urtext to be edited is miscellaneous, and user is directed to different urtext data,
The edit commands of sending be also it is multifarious, this result in the training data of neural network model can not cover comprehensively institute it is in love
Condition, therefore the prediction accuracy of obtained neural network model is trained also just to be unable to reach satisfied degree.
On this basis, inventor continues deeper into research, finally obtains following text editing schemes, overcomes above-mentioned use
The scheme of Neural Network model predictive command object word is limited to the scale of construction of training sample, the lower problem of prediction accuracy.
Next, the method for editing text of the application is introduced in conjunction with attached drawing 1, as shown in Figure 1, this method comprises:
Step S100, urtext data and user's edit commands to be edited are obtained；
Wherein, urtext data to be edited refer to the text data that user needs to edit.Obtained in this step to
The urtext data of editor can be speech form, textual form, photo form etc..That is, user can by voice, beat
Word, the modes such as take pictures input urtext data.For various forms of urtext data, the application can use corresponding
Processing mode is converted into textual form.Show for example, user inputs urtext data by voice, it can in the present embodiment
To be textual form by speech recognition by speech transcription technology.When user inputs the urtext of photo form by taking pictures
Data can identify urtext data in the present embodiment by image recognition technology from photo.
For user's edit commands, refers to the order used when user modifies urtext data, be such as directed to urtext
" today, the weather of blue whale was fine, I plans to go that and play several days for data.", it needs " blue whale " being changed to " Nanjing ", then needs to provide
User's edit commands are as follows: " blue whale is revised as Nanjing ".
Ibid, the user's edit commands obtained in this step can be speech form, textual form, other multimedia forms
Deng.As user speech says edit commands, or direct input editing order in the form of text or other touch screen hand-writings input volume
Collect order etc..It is directed to various forms of user's edit commands in the present embodiment, text can be converted to according to corresponding processing mode
User's edit commands of form.
Step S110, the corresponding edit operation of user's edit commands is determined；
Specifically, user is denumerable to the type of the various edit operations of urtext data, and the application can be preparatory
Collect various edit operations.And then after getting user's edit commands, user can be determined based on the edit operation of collection
The corresponding edit operation of edit commands.
A kind of optional mode, can compare the edit operation collected in advance, in user's edit commands present in lookup
Edit operation, as corresponding edit operation.
Another optional mode, the application can advance with training user's edit commands comprising various edit operations
As training data, training neural network model, and user's edit commands pair is identified using trained neural network model
The edit operation answered.
Step S120, according to the urtext data, the semantic phase for each word for being included with user's edit commands
Guan Xing determines command object word from user's edit commands；
Specifically, user's edit commands is therefore wrapped in user's edit commands for urtext data to be edited
Also there are semantic dependencies with urtext data for the command object word contained.In the present embodiment, urtext is not being cut off out
Command object word is predicted in the case where data individually for user's edit commands, but comprehensively considers urtext data and uses
Family edit commands includes the semantic dependency of each word, obtains the command object word in user's edit commands on this basis, by
In the semantic dependency for considering the two, therefore obtained command object word is more accurate.
Step S130, according to the edit operation, the command object word in the urtext data is compiled
Volume.
Specifically, aforementioned to have determined that the corresponding edit operation of user's edit commands and the edit operation are targeted
Command object word therefore command object word executive editor described in urtext data can be operated according to edit operation.
Show for example, edit operation is " underlining ", command object word is that " weather " then can be by urtext data " today blue whale
Weather is fine, I plans to go that and play several days " in " weather " addition underscore, effect is as follows after clean up editing: " today blue whaleWeatherVery well, I plans to go several days of that object for appreciation ".Further, user further issues edit commands " blue whale on the basis of the above
Be revised as Nanjing ", then edit operation be " ... be revised as ... ", command object word be modify before it is " blue whale " and modified
" Nanjing ", then edited effect is as follows: " today NanjingWeatherVery well, I plans to go several days of that object for appreciation ".
Method for editing text provided by the embodiments of the present application gets urtext data and user to be edited and edits life
Enable, determine the corresponding edit operation of the edit commands, and according to urtext data, with user's edit commands included it is each
The semantic dependency of word determines command object word, finally according to edit operation, to urtext data from user's edit commands
In command object word edited.Application scheme user only needs input editing order, can be realized automatically to original text
The process that notebook data is edited, greatly reduces manual operation, improves editorial efficiency.
Also, the application has comprehensively considered urtext data with user's edit commands when determining command object word
The semantic dependency of each word substantially increases the accuracy that command object word determines, can be more accurately according to the wish of user
Complete entire text editing process.
In next embodiment, to above-mentioned steps S120, according to the urtext data, life is edited with the user
The semantic dependency for enabling each word for being included determines that the process of command object word is situated between in detail from user's edit commands
It continues.
Referring to fig. 2, Fig. 2 is a kind of determining command object word method flow diagram disclosed in the embodiment of the present application.Such as Fig. 2 institute
Show, this method comprises:
Step S200, according to the semantic dependency of each word and the urtext data in user's edit commands,
Determine the correlative character of equivalent；
Wherein, the correlative character of word is used to describe the semantic relation between user's edit commands and urtext data.
Step S210, the phase of each word in model and user's edit commands is determined using order word trained in advance
Closing property feature, determines command object word from each word that user's edit commands includes.
Wherein, the order word determines that model is obtained by training in advance.
When determining that model carries out pre-training to order word, training sample may include:
User's edit commands corresponding with training text data includes the term vector of training word, and according to the trained word
The correlative character of training word determined by semantic dependency with the training text data.
Sample label may include: the trained word whether be command object word annotation results.
Specifically, the application can collect training text data and user to the training text data distributing in advance
User's edit commands.Term vector is carried out by the training word for being included by user's edit commands, obtains term vector.According to training
Word and the semantic dependency of training text data determine the correlative character of training word.Meanwhile whether manually marking each trained word
For command object word.Final utilize trains the corresponding term vector of word, the correlative character of training word and artificial annotation results, right
Order word determines that model is trained.
Order word in the present embodiment determines that model can be using deep neural network model, by using depth nerve net
Network model, the corresponding term vector of combined training word, the correlative character of training word and artificial annotation results, can be accurately pre-
Survey the command object word in user's edit commands.
Further, determine that the process of the correlative character of word is introduced for above-mentioned steps S200.Referring in detail to Fig. 3
Shown, which may include:
Step S300, to user's edit commands and the urtext data segment and term vector, obtain
The term vector of the participle respectively contained；
Specifically, participle tool can be used respectively to segment user's edit commands and urtext data, obtain
The participle that the participle and urtext data that user's edit commands includes include.
If urtext data are " I will order a today from Nanjing to Pekinese's train ticket ", result is as follows after participle
" I will order a today from Nanjing to Pekinese's train ticket ", wherein different participles are spaced apart by space.User's edit commands
For " buying two instead a today is ordered ", result is following " buying two instead a today is ordered " after participle.
After participle, term vector further is carried out to each participle, obtains segmenting corresponding term vector.
Optionally, if urtext data are longer, part urtext data is can choose and carry out word segmentation processing.Such as,
One or more snippets that can choose in urtext data carries out word segmentation processing, or the urtext data of selection regular length
Word segmentation processing is carried out, which can set according to application demand.
Step S310, the i for the participle that user's edit commands and the urtext data respectively contain is determined respectively
First entry；
Wherein, i value [1, N].N is preset constant, and the value of N is determined according to application demand.When N value is different
When, determining i member number of entries is also different.As N value be 1 when, i member entry is only 1 yuan of entry, when N value be 2 when, i member word
Item includes 1 yuan of entry and 2 yuan of entries.
Wherein, the definition of the i member entry for the participle that text includes is: by the continuous i-1 after the participle in participle and text
The phrase of a participle composition, the i member entry of the referred to as described participle.
Still illustrate by taking above-mentioned urtext data and user's edit commands as an example:
1 yuan of entry that urtext data included respectively segment, which is respectively as follows: me, will order a today from Nanjing to Pekinese
Train ticket.
2 yuan of entries that urtext data included respectively segment be respectively as follows: I to order one by one today today from
From Nanjing Nanjing to the train ticket to Beijing Pekinese.
1 yuan of entry that user's edit commands included respectively segment is respectively as follows: buys two a today is ordered instead.
2 yuan of entries that user's edit commands included respectively segment are respectively as follows: that today one by one, today changed instead ordering
At buying two.
The i member entry of above-mentioned different participles is spaced apart by space.
Step S320, the i member entry for the participle for including according to user's edit commands, in the urtext data
The match condition of the i member entry of each participle, determines the correlative character for the participle that user's edit commands includes.
Specifically, the i member entry for each participle for including for user's edit commands, determines itself and urtext data
In the match condition of i member entry that respectively segments.And the correlation for the participle that user's edit commands includes is determined based on match condition
Property feature.Wherein, the correlative character of participle can be embodied by match condition.
In a kind of optional embodiment, above-mentioned steps S320 determines the process of participle correlative character according to matching score
It is referred to Fig. 4 realization, as shown in figure 4, the process may include:
Step S400, the i member entry for calculating the participle that user's edit commands includes, in the urtext data
The matching score of the i member entry of each participle；
The Euclidean distance or COS distance of the specific term vector that two i member entries can be calculated when calculating, as matching
Point.
Wherein, in user's edit commands each participle i member entry, with the i member entry respectively segmented in urtext data
Matching score composition i member entry matching score vector, dimension is equal to the total quantity that segments in urtext data.
Step S410, according to the size relation of the matching score and setting matching score threshold value, determine that the user compiles
Collect the i member entry for the participle that order includes, the coverage in the urtext data；
Specifically, matching score threshold value can be preset in the present embodiment, when matching score is greater than setting matching score
When threshold value, it is believed that the i member entry of participle occurred in urtext data namely coverage is 1.When matching score not
When greater than setting matching score threshold value, it is believed that the i member entry of participle did not occur in urtext data, coverage
It is 0.
Step S420, according to the matching score and/or the coverage, point that user's edit commands includes is determined
The correlative character of word.
In the present embodiment, it can determine that the correlation for the participle that user's edit commands includes is special only with reference to matching score
Sign, the matching score vector of the i member entry for the participle for such as including using user's edit commands of foregoing description is as the phase of the participle
Closing property feature.
In addition, the present embodiment can also determine that the correlation for the participle that user's edit commands includes is special only with reference to coverage
Sign, the i member entry for the participle for such as including by user's edit commands, phase of the coverage as the participle in urtext data
Closing property feature.
It certainly, can be with comprehensive reference matching score and coverage, to determine user's edit commands packet in the present embodiment
The correlative character of the participle contained.
Application scheme in order to facilitate understanding is illustrated by following examples.
It will again be assumed that urtext data are " I will order a today from Nanjing to Pekinese's train ticket ".User's edit commands
For " buying two instead a today is ordered ".
Vacation lets n equal 2, then urtext data and user's edit commands include 1,2 yuan of for example preceding introduction of entry of participle, this
Place repeats no more.
For segmenting " ordering " in user's edit commands, respectively segmented in 2 yuan of entries " ordering one " and urtext data
The matching score vectors of 2 yuan of entries be (0.11,0.21,0.9 ..., 0.11).Wherein, because " ordering one " appear in it is original
In text data, when it carries out matching primitives with the 2 yuan of entries " ordering one " for segmenting " ordering " in urtext data, match
Point higher, such as 0.9 in aforementioned matching score vector, and the 2 yuan of entries segmented with other in urtext data are matched
When, matching score is lower.
The i member entry respectively segmented in the i member entry and urtext data segmented in final calculating user's edit commands
Matching score vector, and the coverage in urtext data, as a result as shown in the table:
Table 1
Further, model is determined using order word for above-mentioned steps S210 to determine that the process of command object word carries out
It introduces.
Word of issuing orders of introducing simple first determines model.
A kind of optional structure, the order word in the present embodiment determine that model may include the first input layer, the first hidden layer
Set and output layer.Such as Fig. 5, illustrates a kind of optional order word and determine model.Including the first input layer V1-Vm,
One hidden layer set and output layer T1-Tm.M is the word sum that user's edit commands includes.
Model is determined based on the order word of the present embodiment, and in conjunction with Fig. 6, the process for determining command object word is illustrated,
The process may include:
Step S500, by the first input layer, it is special that the term vector of each word and correlation in user's edit commands are inputted
Sign；
Specifically, the term vector and correlation of each word in determining user's edit commands are had been described above in previous embodiment
The term vector of aforementioned determination and correlative character can be inputted order word by the first input layer in the present embodiment and determined by feature
Model.
Step S510, by the first hidden layer set, the correlative character of each word is converted, after obtaining the transformation of each word
Correlative character；
Specifically, the correlative character for each word that the first hidden layer set can input the first input layer converts, and obtains
Correlative character after to the transformation of each word.
Step S520, by output layer, according to correlative character determination after the transformation of each word and the user is exported
Command object word in edit commands.
Specifically, correlative character is converted by the first hidden layer set, order word determines that the output layer of model can
With determining according to correlative character after the transformation of each word and export the command object word in user's edit commands.Exporting can in result
To mark command object word, 1 label target order word is such as used, uses the 0 non-targeted order word of mark.
Such as user's edit commands " buying two instead a today is ordered " of aforementioned exemplary, corresponding output result can be with
It is " order/1 one/1 today/1/0 be changed to/0 buy/1 two/1 ".Accordingly it is found that wherein command object word includes " ordering ", " one
", " today ", " buying ", " two ".
In another embodiment of the application, describes another order word and determine model, in the exemplary order of Fig. 5
On the basis of word determines model, as shown in connection with fig. 7, order word determines that model further comprises another branch in the present embodiment, should
Branched structure is for determining that each word is to the attention rate of each word in urtext data in user's edit commands.The branch includes:
Two input layer S1-Sq, the second hidden layer set and coding layer.Model is determined compared to order word disclosed in a upper embodiment, this
Order word in embodiment determines model due to increasing the second input layer, and in training, training sample needs further to increase
Add: the term vector for the participle that training text data include.
Model is determined based on the order word of the present embodiment, and in conjunction with Fig. 8, the process for determining command object word is illustrated,
The process may include:
Step S600, by the first input layer, it is special that the term vector of each word and correlation in user's edit commands are inputted
Sign；
Step S610, by the first hidden layer set, the correlative character of each word is converted, correlation after being converted
Feature；
Wherein, the step S600-S610 of the present embodiment and step S500-S510 in previous embodiment is corresponded, in detail
Referring to foregoing description, details are not described herein again.
Step S620, by the second input layer, the term vector of each word in the urtext data is inputted；
Specifically, the second input layer inputs the term vector s of each word in urtext data1,…sq, wherein q indicates original text
Word sum in notebook data.
Step S630, the second hidden layer set that model is determined by the order word, to each in the urtext data
The term vector of word is converted, and term vector after the transformation of each word is obtained；
Wherein, h can be used in term vector after the transformation that the second hidden layer set converts term vector1,…hqTable
Show.
It is understood that between step S620-S630 and step S600-S610 and there is no stringent sequences to limit,
Liang Ge branch can synchronize execution, can also any sequencing execution.Fig. 8 merely illustrates a kind of optional executive mode.
Step S640, by coding layer, by k-th of word in pre-set user edit commands to each word in urtext data
Attention rate weight obtain k-th of word to the attention rate of each word in urtext data multiplied by term vector after the transformation of equivalent
Weight vectors；
Wherein, k value [1, m], m are the word sum that user's edit commands includes.
Define pkjIndicate that k-th of word is to the attention rate of j-th of word in urtext data in preset user's edit commands
Weight.hjTerm vector after the transformation that expression converts the term vector of j-th of word in urtext data.It then presets and closes
Note degree weight pkjWith term vector h after the transformationjIt is multiplied, obtains multiplied result ckIt indicates are as follows:
Wherein, pkjIt can be obtained by collecting mass data training in advance.It is used as a known constant in the present embodiment.
Step S650, by coding layer, by k-th of word to the attention rate weight of each word in urtext data to
Amount is added with correlative character after the transformation of k-th of word of the first hidden layer set output, obtains addition result；
Step S660, the output layer that model is determined by the order word according to the addition result determination and exports institute
State the command object word in user's edit commands.
Compared to a upper embodiment, order word determines that model further comprises another branch in the present embodiment, which uses
In determine each word in user's edit commands to the attention rate of each word in urtext data, and it is determining attention rate and first is hidden
Correlative character is added after the transformation of each word in the determining user's edit commands of layer set, is finally come according to the addition result
Determine command object word.Summarize concern of each word to each word in urtext data due to further contemplating user's edit commands
Degree, so that command object word prediction accuracy more improves.
Text editing apparatus provided by the embodiments of the present application is described below, text editing apparatus described below with
Above-described method for editing text can correspond to each other reference.
Referring to Fig. 9, Fig. 9 is a kind of text editing apparatus structural schematic diagram disclosed in the embodiment of the present application.As shown in figure 9,
The apparatus may include:
User's edit commands acquiring unit 11, for obtaining urtext data and user's edit commands to be edited；
Edit operation determination unit 12, for determining the corresponding edit operation of user's edit commands；
Command object word determination unit 13, for being wrapped with user's edit commands according to the urtext data
The semantic dependency of each word contained determines command object word from user's edit commands；
Editing and processing unit 14, for ordering the target in the urtext data according to the edit operation
Word is enabled to be edited.
Optionally, the command object word determination unit may include:
Correlative character determination unit, for according to each word in user's edit commands and the urtext data
Semantic dependency, determine the correlative character of equivalent；
Model prediction unit, for being determined in model and user's edit commands using order word trained in advance
The correlative character of each word determines command object word from each word that user's edit commands includes；
The order word determines that training sample when model pre-training includes: that user corresponding with training text data edits
Order includes the term vector of training word, and is determined according to the trained word and the semantic dependency of the training text data
Training word correlative character；Sample label include: the trained word whether be command object word annotation results.
Optionally, the correlative character determination unit may include:
Term vector determination unit, for user's edit commands and the urtext data are segmented and word to
Quantization, the term vector of the participle respectively contained；
I member entry determination unit, for determining that user's edit commands and the urtext data are respectively wrapped respectively
The i member entry of the participle contained, i value [1, N], N are setting constant；
Matching score processing unit, the i member entry of the participle for including according to user's edit commands, with the original
The match condition of the i member entry of each participle, determines the correlation for the participle that user's edit commands includes in beginning text data
Property feature.
Optionally, the matching score processing unit may include:
Matching score computing unit, for calculating the i member entry for the participle that user's edit commands includes, with the original
The match condition of the i member entry of each participle in beginning text data；
Coverage determination unit is determined for the size relation according to the matching score and setting matching score threshold value
The i member entry for the participle that user's edit commands includes, the coverage in the urtext data；
Integrated treatment unit, for determining user's edit commands according to the matching score and/or the coverage
The correlative character for the participle for including.
Optionally, the model prediction unit may include:
First input unit inputs the user and edits for determining the first input layer of model by the order word
The term vector and correlative character of each word in order；
Fisrt feature converter unit, for determining the first hidden layer set of model by the order word, to the phase of each word
Closing property feature is converted, and correlative character after the transformation of each word is obtained；
Output unit, for determining the output layer of model by the order word, according to related after the transformation of each word
Property feature it is determining and export the command object word in user's edit commands.
Optionally, the model prediction unit can also include:
Second input unit inputs the urtext for determining the second input layer of model by the order word
The term vector of each word in data；
Second feature converter unit, for determining the second hidden layer set of model by the order word, to described original
The term vector of each word is converted in text data, obtains term vector after the transformation of each word；
Coding unit will be k-th in pre-set user edit commands for determining the coding layer of model by the order word
Word obtains k-th of word to original multiplied by term vector after the transformation of equivalent to the attention rate weight of each word in urtext data
The attention rate weight vectors of each word in beginning text data, k value [1-m], m are the word sum that user's edit commands includes；By institute
State k-th word of k-th of word to the attention rate weight vectors of each word in urtext data, with the first hidden layer set output
Transformation after correlative character be added, input of the addition result as the output layer, so that the output layer is according to the phase
Add result determining and exports the command object word in user's edit commands.
Optionally, user's edit commands acquiring unit may include:
First user's edit commands obtains subelement, and for obtaining user's edit commands of speech form, and transcription is text
User's edit commands of this form；
Or,
Second user edit commands obtains subelement, for obtaining user's edit commands of textual form.
Text editing apparatus provided by the embodiments of the present application can be applied to server, as PC terminal, cloud platform, server and
Server cluster etc..Optionally, Figure 10 shows the hardware block diagram of server, and referring to Fig.1 0, the hardware configuration of server
It may include: at least one processor 1, at least one communication interface 2, at least one processor 3 and at least one communication bus
4；
In the embodiment of the present application, processor 1, communication interface 2, memory 3, communication bus 4 quantity be at least one,
And processor 1, communication interface 2, memory 3 complete mutual communication by communication bus 4；
Processor 1 may be a central processor CPU or specific integrated circuit ASIC (Application
Specific Integrated Circuit), or be arranged to implement the integrated electricity of one or more of the embodiment of the present invention
Road etc.；
Memory 3 may include high speed RAM memory, it is also possible to further include nonvolatile memory (non-volatile
Memory) etc., a for example, at least magnetic disk storage；
Wherein, memory is stored with program, the program that processor can call memory to store, and described program is used for:
Obtain urtext data and user's edit commands to be edited；
Determine the corresponding edit operation of user's edit commands；
According to the urtext data, the semantic dependency for each word for being included with user's edit commands, from institute
It states and determines command object word in user's edit commands；
According to the edit operation, the command object word in the urtext data is edited.
Optionally, the refinement function of described program and extension function can refer to above description.
The embodiment of the present application also provides a kind of storage medium, which can be stored with the journey executed suitable for processor
Sequence, described program are used for:
Obtain urtext data and user's edit commands to be edited；
Determine the corresponding edit operation of user's edit commands；
According to the urtext data, the semantic dependency for each word for being included with user's edit commands, from institute
It states and determines command object word in user's edit commands；
According to the edit operation, the command object word in the urtext data is edited.
Optionally, the refinement function of described program and extension function can refer to above description.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that
A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or apparatus that includes the element.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.
The foregoing description of the disclosed embodiments makes professional and technical personnel in the field can be realized or use the application.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the application.Therefore, the application
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (16)
1. a kind of method for editing text characterized by comprising
Obtain urtext data and user's edit commands to be edited；
Determine the corresponding edit operation of user's edit commands；
According to the urtext data, the semantic dependency for each word for being included with user's edit commands, from the use
Command object word is determined in the edit commands of family；
According to the edit operation, the command object word in the urtext data is edited.
2. the method according to claim 1, wherein described according to the urtext data, with the user
The semantic dependency for each word that edit commands is included determines command object word from user's edit commands, comprising:
According to the semantic dependency of each word and the urtext data in user's edit commands, the phase of equivalent is determined
Closing property feature；
The correlative character that each word in model and user's edit commands is determined using order word trained in advance, from institute
It states and determines command object word in each word that user's edit commands includes；
The order word determines that training sample when model pre-training includes: user's edit commands corresponding with training text data
Include the term vector of training word, and is instructed according to determined by the trained word and the semantic dependency of the training text data
Practice the correlative character of word；Sample label include: the trained word whether be command object word annotation results.
3. according to the method described in claim 2, it is characterized in that, described according to each word and institute in user's edit commands
The semantic dependency for stating urtext data determines the correlative character of equivalent, comprising:
To user's edit commands and the urtext data segment and term vector, the participle respectively contained
Term vector；
The i member entry for the participle that user's edit commands and the urtext data respectively contain, i value are determined respectively
[1, N], N are setting constant；
I member according to the i member entry for the participle that user's edit commands includes, with each participle in the urtext data
The match condition of entry determines the correlative character for the participle that user's edit commands includes.
4. according to the method described in claim 3, it is characterized in that, the participle for including according to user's edit commands
The match condition of the i member entry of each participle, determines user's edit commands in i member entry, with the urtext data
The correlative character for the participle for including, comprising:
The i member entry for calculating the participle that user's edit commands includes, the i member with each participle in the urtext data
The matching score of entry；
According to the size relation of the matching score and setting matching score threshold value, point that user's edit commands includes is determined
The i member entry of word, the coverage in the urtext data；
According to the matching score and/or the coverage, determine that the correlation for the participle that user's edit commands includes is special
Sign.
5. according to the method described in claim 2, it is characterized in that, described determine model using order word trained in advance, with
And in user's edit commands each word correlative character, from each word that user's edit commands includes determine target life
Enable word, comprising:
The first input layer that model is determined by the order word inputs the term vector and phase of each word in user's edit commands
Closing property feature；
The first hidden layer set that model is determined by the order word, converts the correlative character of each word, obtains each word
Transformation after correlative character；
The output layer that model is determined by the order word according to correlative character determination after the transformation of each word and exports institute
State the command object word in user's edit commands.
6. according to the method described in claim 5, it is characterized by further comprising:
The second input layer that model is determined by the order word inputs the term vector of each word in the urtext data；
The second hidden layer set that model is determined by the order word carries out the term vector of each word in the urtext data
Transformation, obtains term vector after the transformation of each word；
The coding layer that model is determined by the order word, by k-th of word in pre-set user edit commands to urtext data
In the attention rate weight of each word obtain k-th of word to each word in urtext data multiplied by term vector after the transformation of equivalent
Attention rate weight vectors, k value [1-m], m be user's edit commands include word sum；
The coding layer that model is determined by the order word, by k-th of word to the attention rate of each word in urtext data
Weight vectors are added with correlative character after the transformation of k-th of word of the first hidden layer set output, and addition result is as institute
The input of output layer is stated, so that the output layer is determining according to the addition result and exports the mesh in user's edit commands
Mark order word.
7. the method according to claim 1, wherein obtaining the process of user's edit commands, comprising:
User's edit commands of speech form is obtained, and transcription is user's edit commands of textual form；
Or,
Obtain user's edit commands of textual form.
8. a kind of text editing apparatus characterized by comprising
User's edit commands acquiring unit, for obtaining urtext data and user's edit commands to be edited；
Edit operation determination unit, for determining the corresponding edit operation of user's edit commands；
Command object word determination unit, for according to the urtext data, with user's edit commands included it is each
The semantic dependency of word determines command object word from user's edit commands；
Editing and processing unit, for according to the edit operation, to the command object word in the urtext data into
Edlin.
9. device according to claim 8, which is characterized in that the command object word determination unit includes:
Correlative character determination unit, for the language according to each word and the urtext data in user's edit commands
Adopted correlation determines the correlative character of equivalent；
Model prediction unit, for determining each word in model and user's edit commands using order word trained in advance
Correlative character, from each word that user's edit commands includes determine command object word；
The order word determines that training sample when model pre-training includes: user's edit commands corresponding with training text data
Include the term vector of training word, and is instructed according to determined by the trained word and the semantic dependency of the training text data
Practice the correlative character of word；Sample label include: the trained word whether be command object word annotation results.
10. device according to claim 9, which is characterized in that the correlative character determination unit includes:
Term vector determination unit, for user's edit commands and the urtext data segment and term vector
Change, the term vector of the participle respectively contained；
I member entry determination unit, for determining what user's edit commands and the urtext data respectively contained respectively
The i member entry of participle, i value [1, N], N are setting constant；
Matching score processing unit, the i member entry of the participle for including according to user's edit commands, with the original text
The match condition of the i member entry of each participle in notebook data determines that the correlation for the participle that user's edit commands includes is special
Sign.
11. device according to claim 10, which is characterized in that the matching score processing unit includes:
Matching score computing unit, for calculating the i member entry for the participle that user's edit commands includes, with the original text
The match condition of the i member entry of each participle in notebook data；
Coverage determination unit, described in determining according to the matching score and the size relation for setting matching score threshold value
The i member entry for the participle that user's edit commands includes, the coverage in the urtext data；
Integrated treatment unit, for determining that user's edit commands includes according to the matching score and/or the coverage
Participle correlative character.
12. device according to claim 9, which is characterized in that the model prediction unit includes:
First input unit inputs user's edit commands for determining the first input layer of model by the order word
In each word term vector and correlative character；
Fisrt feature converter unit, for determining the first hidden layer set of model by the order word, to the correlation of each word
Feature is converted, and correlative character after the transformation of each word is obtained；
Output unit, it is special according to correlation after the transformation of each word for determining the output layer of model by the order word
Sign determines and exports the command object word in user's edit commands.
13. device according to claim 12, which is characterized in that the model prediction unit further include:
Second input unit inputs the urtext data for determining the second input layer of model by the order word
In each word term vector；
Second feature converter unit, for determining the second hidden layer set of model by the order word, to the urtext
The term vector of each word is converted in data, obtains term vector after the transformation of each word；
Coding unit, for determining the coding layer of model by the order word, by k-th of word pair in pre-set user edit commands
The attention rate weight of each word in urtext data obtains k-th of word to original text multiplied by term vector after the transformation of equivalent
The attention rate weight vectors of each word in notebook data, k value [1-m], m are the word sum that user's edit commands includes；By the kth
Attention rate weight vectors of a word to each word in urtext data, the change of k-th of the word exported with the first hidden layer set
It changes rear correlative character to be added, input of the addition result as the output layer, so that the output layer is tied according to the addition
Fruit is determining and exports the command object word in user's edit commands.
14. device according to claim 8, which is characterized in that user's edit commands acquiring unit includes:
First user's edit commands obtains subelement, and for obtaining user's edit commands of speech form, and transcription is text shape
User's edit commands of formula；
Or,
Second user edit commands obtains subelement, for obtaining user's edit commands of textual form.
15. a kind of text editing apparatus characterized by comprising memory and processor；
The memory, for storing program；
The processor is realized for executing described program such as method for editing text of any of claims 1-7
Each step.
16. a kind of readable storage medium storing program for executing, is stored thereon with computer program, which is characterized in that the computer program is processed
When device executes, each step such as method for editing text of any of claims 1-7 is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810262255.3A CN110321534B (en) | 2018-03-28 | 2018-03-28 | Text editing method, device, equipment and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810262255.3A CN110321534B (en) | 2018-03-28 | 2018-03-28 | Text editing method, device, equipment and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110321534A true CN110321534A (en) | 2019-10-11 |
| CN110321534B CN110321534B (en) | 2023-11-24 |
Family
ID=68109605
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810262255.3A Active CN110321534B (en) | 2018-03-28 | 2018-03-28 | Text editing method, device, equipment and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110321534B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139368A (en) * | 2021-05-18 | 2021-07-20 | 清华大学 | Text editing method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1708784A (en) * | 2002-10-29 | 2005-12-14 | 诺基亚有限公司 | Method and system for editing text in a handheld electronic device |
| CN1726488A (en) * | 2002-05-07 | 2006-01-25 | 国际商业机器公司 | Integrated development tool for building a natural language understanding application |
| KR20160090743A (en) * | 2015-01-22 | 2016-08-01 | 삼성전자주식회사 | A text editing appratus and a text editing method based on sppech signal |
| CN107480118A (en) * | 2017-08-16 | 2017-12-15 | 科大讯飞股份有限公司 | Method for editing text and device |
-
2018
- 2018-03-28 CN CN201810262255.3A patent/CN110321534B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1726488A (en) * | 2002-05-07 | 2006-01-25 | 国际商业机器公司 | Integrated development tool for building a natural language understanding application |
| CN1708784A (en) * | 2002-10-29 | 2005-12-14 | 诺基亚有限公司 | Method and system for editing text in a handheld electronic device |
| KR20160090743A (en) * | 2015-01-22 | 2016-08-01 | 삼성전자주식회사 | A text editing appratus and a text editing method based on sppech signal |
| CN107480118A (en) * | 2017-08-16 | 2017-12-15 | 科大讯飞股份有限公司 | Method for editing text and device |
Non-Patent Citations (1)
| Title |
|---|
| 杜慧等: "基于情感词向量的微博情感分类", 《中文信息学报》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113139368A (en) * | 2021-05-18 | 2021-07-20 | 清华大学 | Text editing method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110321534B (en) | 2023-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107563498B (en) | Image description method and system based on combined visual and semantic attention strategy | |
| CN107330011B (en) | The recognition methods of the name entity of more strategy fusions and device | |
| CN102004792B (en) | Method and system for generating hot-searching word | |
| CN115630640B (en) | Intelligent writing method, device, equipment and medium | |
| CN110147451B (en) | Dialogue command understanding method based on knowledge graph | |
| CN108765383B (en) | Video description method based on deep migration learning | |
| CN111666427A (en) | Entity relationship joint extraction method, device, equipment and medium | |
| CN109408833A (en) | A kind of interpretation method, device, equipment and readable storage medium storing program for executing | |
| CN108491421B (en) | Method, device and equipment for generating question and answer and computing storage medium | |
| CN109858041A (en) | A kind of name entity recognition method of semi-supervised learning combination Custom Dictionaries | |
| CN108052499A (en) | Text error correction method, device and computer-readable medium based on artificial intelligence | |
| CN109446505A (en) | Model essay generation method and system | |
| CN110287482A (en) | Semi-automation participle corpus labeling training device | |
| CN109326280B (en) | Singing synthesis method and device and electronic equipment | |
| CN108875090A (en) | A kind of song recommendations method, apparatus and storage medium | |
| Dethlefs et al. | Conditional random fields for responsive surface realisation using global features | |
| CN111563158A (en) | Text sorting method, sorting device, server and computer-readable storage medium | |
| CN112765983A (en) | Entity disambiguation method based on neural network combined with knowledge description | |
| CN108304424A (en) | Text key word extracting method and text key word extraction element | |
| CN107221344A (en) | A kind of speech emotional moving method | |
| CN106297766B (en) | Phoneme synthesizing method and system | |
| CN112541347A (en) | Machine reading understanding method based on pre-training model | |
| WO2023029354A1 (en) | Text information extraction method and apparatus, and storage medium and computer device | |
| CN107329951A (en) | Build name entity mark resources bank method, device, storage medium and computer equipment | |
| CN110321534A (en) | A kind of method for editing text, device, equipment and readable storage medium storing program for executing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |