CN114625860A - Contract clause identification method, device, equipment and medium - Google Patents
Contract clause identification method, device, equipment and medium Download PDFInfo
- Publication number
- CN114625860A CN114625860A CN202210305314.7A CN202210305314A CN114625860A CN 114625860 A CN114625860 A CN 114625860A CN 202210305314 A CN202210305314 A CN 202210305314A CN 114625860 A CN114625860 A CN 114625860A
- Authority
- CN
- China
- Prior art keywords
- answer
- contract
- model
- standard
- training
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000002372 labelling Methods 0.000 claims abstract description 52
- 238000012549 training Methods 0.000 claims description 114
- 239000013598 vector Substances 0.000 claims description 62
- 230000006870 function Effects 0.000 claims description 26
- 238000004364 calculation method Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 6
- 230000000644 propagated effect Effects 0.000 description 4
- 239000013307 optical fiber Substances 0.000 description 2
- 230000002093 peripheral effect Effects 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000003491 array Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001902 propagating effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/338—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/103—Formatting, i.e. changing of presentation of documents
- G06F40/117—Tagging; Marking up; Designating a block; Setting of attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/18—Legal services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Mathematical Physics (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Technology Law (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Human Computer Interaction (AREA)
- Evolutionary Biology (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a method, a device, equipment and a medium for identifying contract clauses. The identification method of the contract clauses comprises the following steps: acquiring a contract text and at least one standard question matched with contract terms to be identified; combining each standard question with a contract text respectively to obtain a plurality of input data; respectively inputting the input data into a reading understanding model which is trained, and acquiring answer labeling positions of the standard questions in a contract text; and forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text. The technical scheme of the embodiment of the invention can greatly reduce the requirement on the labeled sample and improve the data identification accuracy rate under the condition of variable contract data identification.
Description
Technical Field
The embodiment of the invention relates to the technical field of data processing, in particular to a method, a device, equipment and a medium for identifying contract clauses.
Background
The identification of contract clauses is a technical point which is widely applied in the technical field of data processing, and the identification efficiency of the contract clauses directly influences the data processing time.
At present, contract clauses are generally identified by a keyword matching method, but the keyword matching method needs to manually maintain a large keyword library and cannot identify the keywords which do not appear, and when the contract clauses are identified by an entity extraction method of a multi-layer attention mechanism, although keyword matching is not relied on, the entities cannot be identified into categories which are not seen. For example, if only the "name of party a" and the "name of party b" are involved in the training data, it is not possible for the model to complete the correct classification when the "name of party a" is encountered.
Disclosure of Invention
The embodiment of the invention provides a contract clause identification method, a contract clause identification device, contract clause identification equipment and a contract clause identification medium, which can greatly reduce the requirement on a labeling sample and improve the data identification accuracy rate in a changeable contract data identification environment.
In a first aspect, an embodiment of the present invention provides a method for identifying contract terms, including:
acquiring a contract text and at least one standard question matched with contract terms to be identified;
combining each standard question with a contract text respectively to obtain a plurality of input data;
respectively inputting the input data into a reading understanding model which completes training, and acquiring the answer labeling positions of the standard questions in the contract text;
and forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text.
In a second aspect, an embodiment of the present invention further provides an apparatus for identifying contract terms, including:
the first data acquisition module is used for acquiring a contract text and at least one standard question matched with contract clauses to be identified;
the data combination module is used for combining each standard question with the contract text respectively to obtain a plurality of input data;
the second data acquisition module is used for respectively inputting all input data into the reading understanding model which completes training and acquiring answer marking positions of all standard questions in the contract text;
and the data identification module is used for forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, combining the standard questions with the matched standard answers and forming a contract clause identification result matched with the contract text.
In a third aspect, an embodiment of the present invention further provides an electronic device, where the electronic device includes:
one or more processors;
storage means for storing one or more programs;
when executed by the one or more processors, cause the one or more processors to implement a method of identifying contract terms as provided by any of the embodiments of the invention.
In a fourth aspect, an embodiment of the present invention further provides a computer storage medium, on which a computer program is stored, where the computer program is executed by a processor, and the computer program implements the method for identifying contract terms provided in any embodiment of the present invention.
According to the technical scheme of the embodiment, by obtaining the contract text and at least one standard question matched with contract terms to be identified, each standard question is combined with the contract text respectively to obtain a plurality of input data, the input data are input into a reading understanding model after training respectively to obtain answer marking positions of the standard questions in the contract text, standard answers corresponding to the standard questions respectively are further formed according to the answer marking positions, and the standard questions are combined with the matched standard answers to form a contract term identification result matched with the contract text. The answer labeling positions of the standard questions matched with the contract text in the contract text can be obtained based on the reading understanding model, so that the standard questions and the matched standard answers can be combined to form a contract clause recognition result matched with the contract text, a keyword bank does not need to be maintained, the limitation of entity type recognition can be broken, zero sample data can be recognized more accurately, the problems that in the prior art, the keyword bank is maintained and unseen entity types cannot be recognized are solved, the requirements for labeling samples can be greatly reduced, and the data recognition accuracy under a changeable contract data recognition environment is improved.
Drawings
FIG. 1 is a flowchart of a method for identifying terms of a contract according to an embodiment of the present invention;
FIG. 2 is a flowchart of a method for identifying terms of a contract according to a second embodiment of the present invention;
FIG. 3 is a flowchart of model training of a method for identifying contract terms according to a second embodiment of the present invention;
FIG. 4 is a data flow diagram illustrating a reading understanding model training process according to a second embodiment of the present invention;
fig. 5 is a schematic diagram of a contract clause recognition device provided in the third embodiment of the present invention;
fig. 6 is a schematic structural diagram of an electronic device according to a fourth embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the invention and are not limiting of the invention.
It should be further noted that, for the convenience of description, only some but not all of the relevant aspects of the present invention are shown in the drawings. Before discussing exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart may describe the operations (or steps) as a sequential process, many of the operations can be performed in parallel, concurrently, or simultaneously. In addition, the order of the operations may be re-arranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figure. The processes may correspond to methods, functions, procedures, subroutines, and the like.
Example one
Fig. 1 is a flowchart of a contract term identification method according to an embodiment of the present invention, which is applicable to a case where a desired content is accurately identified in a multivariate contract data identification environment, and the method can be performed by a contract term identification apparatus, which can be implemented by software and/or hardware, and can be generally integrated in an electronic device. The electronic device may be a terminal device, a server device, or the like, and the embodiment of the present invention does not limit the type of the electronic device that executes the method for identifying contract terms. Accordingly, as shown in fig. 1, the method comprises the following operations:
s110, acquiring a contract text and at least one standard question matched with contract terms to be identified.
The contract text can be a text format of the party or an agreement between the parties for setting up, changing and terminating the civil relationship. The standard question may be a question set for the content of the contract term for obtaining the desired content in the contract term. For example, a standard question may include, but is not limited to, "is the name of first company? "," the name of the second company is? "," delivery mode is? "and the like. The last "? "may be replaced with" XX ". The embodiment of the invention does not limit the question content of the standard questions.
In the embodiment of the invention, the contract text can be firstly obtained, then the contract text is analyzed to obtain each contract clause to be identified, and then at least one standard question matched with each contract clause to be identified is respectively determined. Optionally, the contract text may be obtained by converting the data format of the contract in the non-text format (e.g., the contract in the picture format, etc.).
And S120, combining each standard question with the contract text to obtain a plurality of input data.
Wherein the input data may be the result of combining each standard question with the contract text, respectively.
In the embodiment of the present invention, one standard question may be randomly selected from at least one standard question matched with each contract term to be identified, and combined with the corresponding contract term to be identified, so that each contract term to be identified and the matched standard question are combined to obtain a plurality of input data.
And S130, respectively inputting the input data into the reading understanding model which is trained, and acquiring the answer marking positions of the standard questions in the contract text.
Wherein, the reading understanding model can be a model for classifying and identifying the content in the same clause according to the standard inquiry. The answer annotation location may be the location of the answer in the terms of the contract corresponding to the standard question.
In the embodiment of the invention, after the plurality of input data are obtained, the reading understanding model which is trained can be further obtained, and then the input data are respectively input into the reading understanding model which is trained, so that the answer marking positions of the standard questions forming the input data in the contract clauses of the contract text are determined according to the output result of the reading understanding model. Specifically, the reading understanding model can label the answers of the standard questions in contract terms matched with the standard questions according to the standard questions in the input data to obtain answer labeling positions, the reading understanding model has greater openness to entity types capable of being identified, and entity types other than the reading understanding model can be identified for training, namely the entity types identified during training are not relied on, so that the requirements for labeling samples are greatly reduced.
And S140, forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text.
The standard answer may be information corresponding to the position of the answer label in the contract clause. The contract term identification result may be a combination of the standard question and the matched standard answer.
In the embodiment of the invention, the contract clauses matched with the answer marking positions can be respectively determined, then the standard answers respectively corresponding to the standard questions are obtained in the contract clauses matched with the answer marking positions according to the answer marking positions, and the standard questions and the matched standard answers are further combined to generate the contract clause identification result matched with the contract text.
According to the technical scheme of the embodiment, by obtaining the contract text and at least one standard question matched with contract terms to be identified, each standard question is combined with the contract text respectively to obtain a plurality of input data, the input data are input into a reading understanding model after training respectively to obtain answer marking positions of the standard questions in the contract text, standard answers corresponding to the standard questions respectively are further formed according to the answer marking positions, and the standard questions are combined with the matched standard answers to form a contract term identification result matched with the contract text. The answer labeling positions of the standard questions matched with the contract text in the contract text can be obtained based on the reading understanding model, so that the standard questions and the matched standard answers can be combined to form a contract clause recognition result matched with the contract text, a keyword bank does not need to be maintained, the limitation of entity type recognition can be broken, zero sample data can be recognized more accurately, the problems that in the prior art, the keyword bank is maintained and unseen entity types cannot be recognized are solved, the requirements for labeling samples can be greatly reduced, and the data recognition accuracy under a changeable contract data recognition environment is improved.
Example two
Fig. 2 is a flowchart of a method for identifying contract terms according to a second embodiment of the present invention, which is embodied on the basis of the second embodiment, and in this embodiment, a specific alternative implementation manner for determining the reading understanding model before inputting each input data into the reading understanding model after completing training is given. Before inputting each input data into the reading understanding model which is trained, the specific process of determining the reading understanding model is as follows: acquiring a pre-training model, and splicing the pre-training model with a prompt learning model to form a model to be trained; the pre-training model is obtained by training a set number of Chinese reading understanding data; and performing model training on the model to be trained by using the training sample set, and adjusting model parameters in the prompt learning model in the training process to form a reading understanding model. Accordingly, as shown in fig. 2, the method includes the following operations:
s210, acquiring a contract text and at least one standard question matched with contract terms to be identified.
And S220, combining the standard questions with the contract texts respectively to obtain a plurality of input data.
And S230, acquiring a pre-training model, and splicing the pre-training model and the prompt learning model to form a model to be trained.
The pre-training model can be obtained by training a set number of Chinese reading understanding data. Illustratively, the pre-trained model may include the roberta _ wwm _ ext _ large model. The set number may be a preset number of chinese reading comprehension data for training the pre-training model. The prompt learning model may be a model that adds prompt information to the input without changing the pre-trained model structure and parameters. The model to be trained can be obtained by splicing the pre-training model and the prompt learning model. The training sample set may be a set of samples used to train the model to be trained.
In the embodiment of the invention, a pre-training model obtained by training a set number of Chinese reading understanding data can be obtained, a prompt learning model is created according to the model structure of the pre-training model, and the pre-training model and the prompt learning model are spliced to generate the model to be trained.
S240, performing model training on the model to be trained by using the training sample set, and adjusting model parameters in the prompt learning model in the training process to form a reading understanding model.
In the embodiment of the invention, after the model to be trained is obtained, a training sample set can be further obtained, the model to be trained is trained by using the training sample set, model parameters in the pre-training model are frozen in the model training process, only the parameters in the prompt learning model are adjusted, and when the parameters are adjusted to meet the model prediction requirements (such as the preset prediction accuracy), the reading understanding model is generated. Only a few parameters (only parameters in the prompt learning model) need to be optimized in the training process, so that the model training is more efficient, and the requirement on labeling expectation is greatly reduced. The model has the reasoning ability of a large model and the efficient training ability of a small model simultaneously due to the fusion of the pre-training model and the prompt learning model, and the extracted problems are converted into reading comprehension, so that the model to be trained does not depend on a fixed label system any more, has the zero-sample reasoning ability, and is better suitable for the characteristic of rapid iterative update of the current business.
In an optional embodiment of the present invention, performing model training on a model to be trained by using a training sample set may include: sequentially obtaining at least one training sample; respectively carrying out word segmentation processing and integer format conversion on the question and the text to obtain integer input data respectively corresponding to the question and the text; sequentially inputting all integer input data into a prompt learning model and a pre-training model, and acquiring output word vectors corresponding to all integer input data; inputting each output word vector into a full connection layer, and acquiring an answer initial position description vector and an answer ending position description vector; respectively inputting the answer initial position description vector and the answer ending position description vector into a classification network to obtain an answer initial position probability set and an answer ending position probability set; calculating a loss function corresponding to the model to be trained according to the answer initial position probability set, the answer end position probability set and the labeling result of the question answer in the text; and adjusting model parameters in the prompt learning model according to the loss function calculation result, and returning to execute the operation of sequentially obtaining a training sample until the training finishing condition is met.
The training samples may be samples in a training sample set, and the training samples may include: the questions, the texts and the labeling results of the answers to the questions in the texts. The labeling result may be a result of labeling the starting position and the ending position of the answer to the question in the text. Illustratively, the starting position and the ending position of the answer to the question may be labeled in the text in a one-hot coding manner, so as to obtain vectors corresponding to the starting position and the ending position of the answer to the question, respectively. The integer format conversion may be an operation of converting a participle into an integer, and an example may be to determine an integer corresponding to a participle by referring to a form of a dictionary table. The integer input data may be the result of word segmentation processing of questions and text, as well as integer format conversion. The output word vector may be an output vector determined from the integer input data.
The answer starting position description vector can be a vector for representing the corresponding answer starting position of each output word vector. The answer ending position description vector may be a vector for characterizing the position of the answer ending corresponding to each output word vector. The classification network may be a network capable of outputting classification probabilities. Illustratively, the classification network may include, but is not limited to, a Softmax classifier. The answer starting position probability set may be a set of probability values matched with an answer starting position description vector, and elements in the answer starting position description vector have a one-to-one correspondence relationship with the probability values in the answer starting position probability set. The answer end position probability set may be a set of probability values matched with an answer end position description vector, and elements in the answer end position description vector have a one-to-one correspondence relationship with the probability values in the answer end position probability set. The end training condition may be a condition for stopping training of the model to be trained using the training samples. For example, the ending training condition may include, but is not limited to, training the model to be trained by using all training samples in the training sample set, or the prediction accuracy of the model to be trained reaches a preset prediction accuracy, and the like.
In the embodiment of the present invention, at least one training sample including a question, a text, and a result of marking a question answer in the text may be sequentially obtained from a training sample set, and then the question and the text are subjected to word segmentation according to a preset text division rule, and an integer format conversion is performed on a word segmentation processing result to determine an integer corresponding to each word in the word segmentation processing result, that is, to obtain integer input data corresponding to the question and the text, respectively. After integer input data are obtained, sequentially inputting the integer input data into a prompt learning model and a pre-training model according to the order of word segmentation in a training sample to obtain output word vectors corresponding to the integer input data respectively, further inputting the output word vectors into a full connection layer (a splicing layer spliced after the prompt learning model), further determining an answer initial position description vector and an answer ending position description vector according to the output of the full connection layer, further respectively inputting the answer initial position description vector and the answer ending position description vector into a classification network, calculating an answer initial position probability set matched with the answer initial position description vector through the classification network, and calculating an answer ending position probability set matched with the answer ending position description vector, thereby obtaining an answer initial position probability set, an answer ending position probability set and a labeling result of a question answer in a text according to the answer initial position probability set, the answer ending position probability set and the question answer, calculating a loss function corresponding to the model to be trained to obtain a loss function calculation result, further utilizing an optimizer (such as an Adam optimizer and the like) to reversely deduce the loss function calculation result aiming at the model to be trained, optimizing and prompting model parameters in the learning model, and returning to execute the operation of obtaining a training sample until the training ending condition is met.
For example, to ensure the validity of the validation set, i.e. to ensure that the model performing well in the training sample set can have good effect in the actual test, only the first 10% of the data in each problem can be extracted for training, and the rest data can be used for validation. The design can ensure the uniformity of the verification sample, and the distribution of the verification sample is closer to the actual situation.
For example, the probability value in the answer start position probability set, the probability value in the answer end position probability set, and the labeling result of the question answer in the text may be substituted into the standard formula of the existing negative log-likelihood function to obtain a loss function corresponding to the model to be trained, and the calculation result of the loss function substituted into a specific value is used as the calculation result of the loss function.
In an optional embodiment of the present invention, before inputting each output word vector into the full-connected layer, the method may further include: and creating a full-connection layer according to the dimension of each output word vector and the labeling result of the question answer in the text.
In the embodiment of the invention, the dimension of each output vector can be used as a row of the target matrix, the labeling quantity of the labeling result of the question answer in the text is used as a column of the target matrix, the target matrix is created, the target matrix is further used as a full connection layer, and elements in the target matrix can be randomly generated.
For example, assuming that an output word vector is a 1x1024 vector, the dimension of the output vector is 1024, and the number of labels of the labeling result of the question answer in the text is 2 when the question answer is labeled head and tail in the text, a 1024x2 matrix can be constructed as the full connection layer.
In an optional embodiment of the present invention, adjusting the model parameters in the prompt learning model according to the loss function calculation result may include: calculating an initial gradient according to the loss function calculation result, and outputting a disturbance vector; calculating a confrontation gradient according to the output disturbance vector and the initial gradient; and performing countermeasure training on the model to be trained according to the countermeasure gradient.
Wherein the initial gradient may be a gradient propagated back to the loss function calculation. The output perturbation vector may be a vector determined from the initial gradient and the output word vector. The opposing gradient may be a gradient calculated from a loss function of the output perturbation vector and the initial gradient.
In the embodiment of the invention, the initial gradient can be obtained by reversely propagating the calculation result of the loss function, further calculating the matrix norm of the initial gradient, thereby calculating the input perturbation according to the matrix norm of the initial gradient, and adds the input disturbance to the output word vector to obtain an output disturbance vector, calculates the loss function of the output disturbance vector, and the calculation result of the loss function of the output disturbance vector is reversely propagated to obtain the confrontation gradient, thereby carrying out the confrontation training on the model to be trained according to the confrontation gradient to realize the adjustment of the model parameters in the prompt learning model, the training against the model to be trained increases the loss of the model to be trained, ensures that in the case of increasing the probability of making a wrong judgment on the model, the contract clause recognition result can be accurately output, and the anti-interference performance and the generalization performance of the reading understanding model are improved, so that the method is more suitable for the situation of more interference in the actual service.
And S250, respectively inputting the input data into the reading understanding model which is trained, and acquiring the answer labeling positions of the standard questions in the contract text.
In an optional embodiment of the present invention, the respectively inputting each input data into the reading understanding model completing the training, and obtaining the answer labeling position of each standard question in the contract text may include: acquiring currently input target input data, and inputting the target input data into a reading understanding model; calculating to obtain an answer initial position probability set and an answer end position probability set which are matched with the target input data through reading the understanding model; respectively filtering probability values which do not reach a preset threshold value threshold in an answer initial position probability set and an answer ending position probability set by reading an understanding model to obtain a target answer initial position probability set and a target answer ending position probability set; and acquiring an answer marking position of the target standard question corresponding to the target input data according to a preset answer length limit value, a target answer initial position probability set and a target answer final position probability set by reading the understanding model.
Wherein the target input data may be a piece of input data. The preset threshold may be a preset probability threshold. The target answer starting position probability set may be a set of probability values greater than or equal to a preset threshold value in the answer starting position probability set. The target answer end position probability set may be a set of probability values greater than or equal to a preset threshold value in the answer end position probability set. The answer length limit value may be an integer that characterizes the length of the answer. For example, the answer length limit value may be the number of segments of all the segments included in the answer.
In the embodiment of the invention, target input data which needs to be input into a reading understanding model at present can be determined firstly, then the target input data is input into the reading understanding model which is trained, an answer starting position probability set and an answer ending position probability set which are matched with the target input data are determined according to an output result of the reading understanding model, then the probability value of the answer starting position probability set and the probability value of the answer ending position probability set are compared with a preset threshold through the reading understanding model, the probability value which does not reach the preset threshold is filtered, and then the target answer starting position probability set and the target answer ending position probability set are obtained. After the target answer starting position probability set and the target answer ending position probability set are obtained, the reading understanding model can further screen the target answer starting position probability set and the probability values in the target answer ending position probability set according to preset answer length limiting values, and the answer labeling positions of the target standard questions corresponding to the target input data are determined. The probability value which does not reach the preset threshold value threshold is filtered through the preset threshold value threshold, so that the frequency of determining the target standard corresponding to the target input data can be reduced, and the data processing efficiency is improved.
In an optional embodiment of the present invention, the reading the understanding model, and according to a preset answer length limit value, a target answer start position probability set, and a target answer end position probability set, obtaining an answer labeling position of a target standard question corresponding to the target input data may include: determining a target answer starting position probability set and at least one probability combination to be processed which meets a preset answer length limit value in the target answer ending position probability set; and determining the answer labeling position of the target standard question corresponding to the target input data according to the target answer starting position probability and the target answer ending position probability matched with the probability combination to be processed.
The probability combination to be processed may be a combination formed by randomly selecting a probability value from a target answer starting position probability set and a target answer ending position probability set according to a preset answer length limit value. The target answer starting position probability may be a probability value in a target answer starting position probability set constituting the probability combination to be processed. The probability of the ending position of the target answer may be a probability value in a probability set of ending positions of the target answer in the probability combination to be processed.
In the embodiment of the invention, an answer initial position corresponding to a probability value in a target answer initial position probability set can be selected according to a target answer ending position probability set, an answer ending position corresponding to a probability value in a target answer ending position probability set is selected according to a target answer ending position probability set, the number of participles of the answer initial position and the answer ending position corresponding to two probability values is determined, when the number of the participles is larger than or equal to an answer length limiting value, the probability value in the selected target answer initial position probability set and the probability value in the target answer ending position probability set can be combined to obtain at least one probability combination to be processed, then the product of the target answer initial position probability and the target answer ending position probability forming each probability combination to be processed is calculated, and the target answer initial position probability and the target answer ending position probability corresponding to the maximum value of the product are matched in target input data The position of the participle is used as the answer marking position of the target standard question corresponding to the target input data. The probability combination to be processed is determined through the answer length limiting value, the combination of probability values which do not meet the answer length limiting value can be effectively excluded, and therefore the entity which needs to be identified actually is obtained while the efficiency of determining the target standard questions corresponding to the target input data is improved.
And S260, forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text.
Fig. 3 is a flowchart of model training of a contract term identification method according to a second embodiment of the present invention, and as shown in fig. 3, a pre-training model may be first obtained, a prompt learning model is further created, and labeled contract data (training samples) is obtained as read understanding contract data, and then the pre-training model and the prompt learning model are fused and spliced to obtain a model to be trained, and the model to be trained is further countertrained by using the read understanding contract data, and output results of the model to be trained that completes the countertraining are verified and screened to optimize parameters of the prompt learning model, so as to obtain the read understanding model. The data flow in the training process of reading the understanding model can be seen in fig. 4.
The technical scheme of this embodiment includes obtaining a contract text and at least one standard question matched with contract terms to be recognized, further combining each standard question with the contract text to obtain a plurality of input data, thereby obtaining a pre-training model, splicing the pre-training model with a prompt learning model to form a model to be trained, further performing model training on the model to be trained by using a training sample set, adjusting model parameters in the prompt learning model during training to form a reading understanding model, further inputting each input data into the reading understanding model after training, obtaining answer marking positions of each standard question in the contract text, further forming standard answers corresponding to each standard question according to the answer marking positions, and combining each standard question with the matched standard answers, and forming a contract clause identification result matched with the contract text. The answer labeling positions of the standard questions matched with the contract text in the contract text can be obtained based on the reading understanding model, so that the standard questions and the matched standard answers can be combined to form a contract clause recognition result matched with the contract text, a keyword bank does not need to be maintained, the limitation of entity type recognition can be broken, zero sample data can be recognized more accurately, the problems that in the prior art, the keyword bank is maintained and unseen entity types cannot be recognized are solved, the requirements for labeling samples can be greatly reduced, and the data recognition accuracy under a changeable contract data recognition environment is improved.
It should be noted that any permutation and combination between the technical features in the above embodiments also belong to the scope of the present invention.
EXAMPLE III
Fig. 5 is a schematic diagram of an apparatus for identifying contract terms according to a third embodiment of the present invention, as shown in fig. 5, the apparatus includes: a first data obtaining module 310, a data combining module 320, a second data obtaining module 330, and a data identifying module 340, wherein:
a first data acquisition module 310, configured to acquire a contract text and at least one standard question matching contract terms to be identified;
the data combination module 320 is used for combining each standard question with the contract text respectively to obtain a plurality of input data;
the second data acquisition module 330 is configured to input each input data into the reading understanding model that completes training, and acquire an answer labeling position of each standard question in the contract text;
and the data identification module 340 is configured to form standard answers corresponding to the standard questions according to the answer labeling positions, and combine the standard questions and the matched standard answers to form a contract clause identification result matched with the contract text.
According to the technical scheme of the embodiment, by obtaining the contract text and at least one standard question matched with contract terms to be identified, each standard question is combined with the contract text respectively to obtain a plurality of input data, the input data are input into a reading understanding model after training respectively to obtain answer marking positions of the standard questions in the contract text, standard answers corresponding to the standard questions respectively are further formed according to the answer marking positions, and the standard questions are combined with the matched standard answers to form a contract term identification result matched with the contract text. The answer labeling positions of the standard questions matched with the contract text in the contract text can be obtained based on the reading understanding model, so that the standard questions and the matched standard answers can be combined to form a contract clause recognition result matched with the contract text, a keyword library does not need to be maintained, the limitation of entity category recognition can be broken, zero sample data can be recognized more accurately, the problems that in the prior art, the keyword library is maintained and the entity category which is not seen can not be recognized are solved, the requirement on labeling samples can be greatly reduced, and the data recognition accuracy under a changeable contract data recognition environment is improved.
Optionally, the contract clause recognition device further includes a reading understanding model generation model, configured to obtain a pre-training model, and splice the pre-training model and the prompt learning model to form a model to be trained; the pre-training model is obtained by training a set number of Chinese reading understanding data; and performing model training on the model to be trained by using a training sample set, and adjusting model parameters in the prompt learning model in the training process to form the reading understanding model.
Optionally, the reading understanding model generation model is specifically used for: sequentially obtaining a training sample, wherein the training sample comprises: the method comprises the following steps of marking results of questions, texts and answers of the questions in the texts; respectively carrying out word segmentation processing and integer format conversion on the question and the text to obtain integer input data respectively corresponding to the question and the text; sequentially inputting the integer input data into the prompt learning model and the pre-training model to obtain output word vectors corresponding to the integer input data respectively; inputting each output word vector into a full connection layer, and obtaining an answer initial position description vector and an answer ending position description vector; respectively inputting the answer starting position description vector and the answer ending position description vector into a classification network to obtain an answer starting position probability set and an answer ending position probability set; calculating a loss function corresponding to the model to be trained according to the answer starting position probability set, the answer ending position probability set and the labeling result of the question answer in the text; and adjusting the model parameters in the prompt learning model according to the loss function calculation result, and returning to execute the operation of sequentially obtaining a training sample until the training finishing condition is met.
Optionally, the reading understanding model generation model is specifically used for: calculating an initial gradient according to the loss function calculation result, and outputting a disturbance vector; calculating a confrontation gradient according to the output disturbance vector and the initial gradient; and carrying out countermeasure training on the model to be trained according to the countermeasure gradient.
Optionally, the second data obtaining module 330 is specifically configured to obtain currently input target input data, and input the target input data into the reading understanding model; calculating to obtain an answer starting position probability set and an answer ending position probability set which are matched with the target input data through the reading understanding model; respectively filtering probability values which do not reach a preset threshold value threshold from the answer starting position probability set and the answer ending position probability set through the reading understanding model to obtain a target answer starting position probability set and a target answer ending position probability set; and acquiring the answer marking position of the target standard question corresponding to the target input data through the reading understanding model according to a preset answer length limiting value, the target answer initial position probability set and the target answer final position probability set.
Optionally, the second data obtaining module 330 is specifically configured to determine at least one to-be-processed probability combination that satisfies a preset answer length limit value in the target answer starting position probability set and the target answer ending position probability set; and determining the answer labeling position of the target standard question corresponding to the target input data according to the target answer starting position probability and the target answer ending position probability matched with the probability combination to be processed.
Optionally, the apparatus for identifying contract terms further includes a fully-connected layer creating module, configured to create a fully-connected layer according to the dimension of each output word vector and the labeling result of the question answer in the text.
The contract clause identification device can execute the contract clause identification method provided by any embodiment of the invention, and has corresponding functional modules and beneficial effects of the execution method. For technical details that are not described in detail in this embodiment, reference may be made to a method for identifying contract terms provided in any embodiment of the present invention.
Since the identification device of contract terms described above is a device capable of executing the identification method of contract terms in the embodiment of the present invention, a person skilled in the art can understand the specific implementation of the identification device of contract terms in the embodiment of the present invention and various modifications thereof based on the identification method of contract terms described in the embodiment of the present invention, and therefore, how the identification device of contract terms implements the identification method of contract terms in the embodiment of the present invention is not described in detail herein. The device used by those skilled in the art to implement the method for identifying contract terms in the embodiments of the present invention is within the scope of the protection of the present application.
Example four
Fig. 6 is a schematic structural diagram of an electronic device according to a fourth embodiment of the present invention. FIG. 6 illustrates a block diagram of an electronic device 412 that is suitable for use in implementing embodiments of the present invention. The electronic device 412 shown in fig. 6 is only an example and should not bring any limitations to the functionality and scope of use of the embodiments of the present invention.
As shown in fig. 6, the electronic device 412 is in the form of a general purpose computing device. The components of the electronic device 412 may include, but are not limited to: one or more processors 416, a storage device 428, and a bus 418 that couples the various system components including the storage device 428 and the processors 416.
The electronic device 412 may also communicate with one or more external devices 414 (e.g., keyboard, pointing device, camera, display 424, etc.), with one or more devices that enable a user to interact with the electronic device 412, and/or with any devices (e.g., network card, modem, etc.) that enable the electronic device 412 to communicate with one or more other computing devices. Such communication may occur via I/O interface 422. Also, the electronic device 412 may communicate with one or more networks (e.g., a Local Area Network (LAN), Wide Area Network (WAN), etc.) and/or a public Network, such as the internet, via the Network adapter 420. As shown, network adapter 420 communicates with the other modules of electronic device 412 over bus 418. It should be appreciated that although not shown in the figures, other hardware and/or software modules may be used in conjunction with the electronic device 412, including but not limited to: microcode, device drivers, Redundant processing units, external disk drive Arrays, disk array (RAID) systems, tape drives, and data backup storage systems, to name a few.
The processor 416 executes various functional applications and data processing by running programs stored in the storage device 428, for example, implementing the contract clause identification method provided by the above-described embodiment of the present invention, including: acquiring a contract text and at least one standard question matched with contract terms to be identified; combining each standard question with the contract text to obtain a plurality of input data; respectively inputting the input data into a reading understanding model which completes training, and acquiring the answer labeling positions of the standard questions in the contract text; and forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text.
According to the technical scheme of the embodiment, by obtaining the contract text and at least one standard question matched with contract terms to be identified, each standard question is combined with the contract text respectively to obtain a plurality of input data, the input data are input into a reading understanding model after training respectively to obtain answer marking positions of the standard questions in the contract text, standard answers corresponding to the standard questions respectively are further formed according to the answer marking positions, and the standard questions are combined with the matched standard answers to form a contract term identification result matched with the contract text. The answer labeling positions of the standard questions matched with the contract text in the contract text can be obtained based on the reading understanding model, so that the standard questions and the matched standard answers can be combined to form a contract clause recognition result matched with the contract text, a keyword bank does not need to be maintained, the limitation of entity type recognition can be broken, zero sample data can be recognized more accurately, the problems that in the prior art, the keyword bank is maintained and unseen entity types cannot be recognized are solved, the requirements for labeling samples can be greatly reduced, and the data recognition accuracy under a changeable contract data recognition environment is improved.
EXAMPLE five
An embodiment of the present invention further provides a computer storage medium storing a computer program, which when executed by a computer processor is configured to perform the method for identifying contract terms according to any one of the above embodiments of the present invention, including: acquiring a contract text and at least one standard question matched with contract terms to be identified; combining each standard question with a contract text respectively to obtain a plurality of input data; respectively inputting the input data into a reading understanding model which is trained, and acquiring answer labeling positions of the standard questions in a contract text; and forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text.
Computer storage media for embodiments of the invention may employ any combination of one or more computer-readable media. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM) or flash Memory), an optical fiber, a portable compact disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, Radio Frequency (RF), etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider).
It is to be noted that the foregoing description is only exemplary of the invention and that the principles of the technology may be employed. It will be understood by those skilled in the art that the present invention is not limited to the particular embodiments described herein, but is capable of various obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the invention. Therefore, although the present invention has been described in greater detail by the above embodiments, the present invention is not limited to the above embodiments, and may include other equivalent embodiments without departing from the spirit of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims (10)
1. A method for identifying terms of a contract, comprising:
acquiring a contract text and at least one standard question matched with contract terms to be identified;
combining each standard question with a contract text respectively to obtain a plurality of input data;
respectively inputting the input data into a reading understanding model which completes training, and acquiring the answer labeling positions of the standard questions in the contract text;
and forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, and combining the standard questions and the matched standard answers to form a contract clause recognition result matched with the contract text.
2. The method of claim 1, further comprising, prior to separately entering each input datum into the trained reading understanding model:
acquiring a pre-training model, and splicing the pre-training model with a prompt learning model to form a model to be trained; the pre-training model is obtained by training a set number of Chinese reading understanding data;
and performing model training on the model to be trained by using a training sample set, and adjusting model parameters in the prompt learning model in the training process to form the reading understanding model.
3. The method of claim 2, wherein model training the model to be trained using a training sample set comprises:
sequentially obtaining at least one training sample, wherein the training sample comprises: the method comprises the following steps of marking results of questions, texts and answers to the questions in the texts;
respectively carrying out word segmentation processing and integer format conversion on the question and the text to obtain integer input data respectively corresponding to the question and the text;
sequentially inputting each integer input data into the prompt learning model and the pre-training model to obtain output word vectors corresponding to each integer input data;
inputting each output word vector into a full connection layer, and obtaining an answer initial position description vector and an answer ending position description vector;
respectively inputting the answer starting position description vector and the answer ending position description vector into a classification network to obtain an answer starting position probability set and an answer ending position probability set;
calculating a loss function corresponding to the model to be trained according to the answer starting position probability set, the answer ending position probability set and the labeling result of the question answer in the text;
and adjusting the model parameters in the prompt learning model according to the loss function calculation result, and returning to execute the operation of sequentially obtaining a training sample until the training finishing condition is met.
4. The method of claim 3, wherein the adjusting model parameters in the prompt learning model according to the loss function calculation result comprises:
calculating an initial gradient according to the loss function calculation result, and outputting a disturbance vector;
calculating a confrontation gradient according to the output disturbance vector and the initial gradient;
and carrying out countermeasure training on the model to be trained according to the countermeasure gradient.
5. The method of claim 1, wherein inputting each input data into the reading understanding model after completing training respectively, and obtaining the answer labeling position of each standard question in the contract text comprises:
acquiring currently input target input data, and inputting the target input data into the reading understanding model;
calculating to obtain an answer starting position probability set and an answer ending position probability set which are matched with the target input data through the reading understanding model;
respectively filtering probability values which do not reach a preset threshold value threshold from the answer starting position probability set and the answer ending position probability set through the reading understanding model to obtain a target answer starting position probability set and a target answer ending position probability set;
and acquiring the answer marking position of the target standard question corresponding to the target input data through the reading understanding model according to a preset answer length limiting value, the target answer initial position probability set and the target answer final position probability set.
6. The method of claim 5, wherein obtaining, by the reading understanding model, answer annotation positions of the target standard questions corresponding to the target input data according to preset answer length limit values, the target answer starting position probability set, and the target answer ending position probability set comprises:
determining at least one probability combination to be processed which meets a preset answer length limit value in the target answer starting position probability set and the target answer ending position probability set;
and determining the answer labeling position of the target standard question corresponding to the target input data according to the target answer starting position probability and the target answer ending position probability matched with the probability combination to be processed.
7. The method of claim 3, further comprising, prior to inputting each of the output word vectors into a fully-connected layer:
and creating a full-connection layer according to the dimension of each output word vector and the labeling result of the question answer in the text.
8. An apparatus for identifying terms of a contract, comprising:
the first data acquisition module is used for acquiring a contract text and at least one standard question matched with contract clauses to be identified;
the data combination module is used for combining each standard question with the contract text respectively to obtain a plurality of input data;
the second data acquisition module is used for respectively inputting all input data into the reading understanding model which completes training and acquiring answer marking positions of all standard questions in the contract text;
and the data identification module is used for forming standard answers corresponding to the standard questions respectively according to the answer labeling positions, combining the standard questions with the matched standard answers and forming a contract clause identification result matched with the contract text.
9. An electronic device, characterized in that the electronic device comprises:
one or more processors;
storage means for storing one or more programs;
when executed by the one or more processors, cause the one or more processors to implement the method of identifying contract terms of any of claims 1-7.
10. A computer storage medium having stored thereon a computer program, characterized in that the program, when being executed by a processor, implements the method of identification of contract terms according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210305314.7A CN114625860A (en) | 2022-03-25 | 2022-03-25 | Contract clause identification method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210305314.7A CN114625860A (en) | 2022-03-25 | 2022-03-25 | Contract clause identification method, device, equipment and medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114625860A true CN114625860A (en) | 2022-06-14 |
Family
ID=81904518
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210305314.7A Pending CN114625860A (en) | 2022-03-25 | 2022-03-25 | Contract clause identification method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114625860A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116384382A (en) * | 2023-01-04 | 2023-07-04 | 深圳擎盾信息科技有限公司 | Automatic long contract element identification method and device based on multi-round interaction |
-
2022
- 2022-03-25 CN CN202210305314.7A patent/CN114625860A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116384382A (en) * | 2023-01-04 | 2023-07-04 | 深圳擎盾信息科技有限公司 | Automatic long contract element identification method and device based on multi-round interaction |
| CN116384382B (en) * | 2023-01-04 | 2024-03-22 | 深圳擎盾信息科技有限公司 | Automatic long contract element identification method and device based on multi-round interaction |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107293296B (en) | Voice recognition result correction method, device, equipment and storage medium | |
| CN108985358B (en) | Emotion recognition method, device, equipment and storage medium | |
| WO2021051560A1 (en) | Text classification method and apparatus, electronic device, and computer non-volatile readable storage medium | |
| CN110276023B (en) | POI transition event discovery method, device, computing equipment and medium | |
| CN110415679B (en) | Voice error correction method, device, equipment and storage medium | |
| CN112528637B (en) | Text processing model training method, device, computer equipment and storage medium | |
| CN110750993A (en) | Word segmentation method, word segmentation device, named entity identification method and system | |
| CN111079432B (en) | Text detection method and device, electronic equipment and storage medium | |
| EP3940693A1 (en) | Voice interaction-based information verification method and apparatus, and device and computer storage medium | |
| CN112883968B (en) | Image character recognition method, device, medium and electronic equipment | |
| CN111312209A (en) | Text-to-speech conversion processing method and device and electronic equipment | |
| CN113158656B (en) | Ironic content recognition method, ironic content recognition device, electronic device, and storage medium | |
| CN112825114A (en) | Semantic recognition method and device, electronic equipment and storage medium | |
| CN114067786A (en) | Voice recognition method and device, electronic equipment and storage medium | |
| CN113053367A (en) | Speech recognition method, model training method and device for speech recognition | |
| CN114218945A (en) | Entity identification method, device, server and storage medium | |
| CN113743101A (en) | Text error correction method and device, electronic equipment and computer storage medium | |
| CN116955699A (en) | Video cross-mode search model training method, searching method and device | |
| CN112214595A (en) | Category determination method, device, equipment and medium | |
| CN114625860A (en) | Contract clause identification method, device, equipment and medium | |
| CN113160820A (en) | Speech recognition method, and training method, device and equipment of speech recognition model | |
| CN113361523A (en) | Text determination method and device, electronic equipment and computer readable storage medium | |
| CN113761845A (en) | Text generation method and device, storage medium and electronic equipment | |
| CN112487813A (en) | Named entity recognition method and system, electronic equipment and storage medium | |
| CN113420121B (en) | Text processing model training method, voice text processing method and device |
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 |