CN116843162A - Contradiction reconciliation scheme recommendation and scoring system and method - Google Patents
Contradiction reconciliation scheme recommendation and scoring system and method Download PDFInfo
- Publication number
- CN116843162A CN116843162A CN202311083864.XA CN202311083864A CN116843162A CN 116843162 A CN116843162 A CN 116843162A CN 202311083864 A CN202311083864 A CN 202311083864A CN 116843162 A CN116843162 A CN 116843162A
- Authority
- CN
- China
- Prior art keywords
- data
- solution
- text
- case
- contradiction
- 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 description 29
- 238000012545 processing Methods 0.000 claims abstract description 49
- 230000008094 contradictory effect Effects 0.000 claims abstract description 47
- 238000013136 deep learning model Methods 0.000 claims abstract description 27
- 238000004140 cleaning Methods 0.000 claims abstract description 22
- 230000007246 mechanism Effects 0.000 claims abstract description 22
- 238000013135 deep learning Methods 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 28
- 239000013598 vector Substances 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 16
- 230000011218 segmentation Effects 0.000 claims description 15
- 238000001914 filtration Methods 0.000 claims description 14
- 238000005457 optimization Methods 0.000 claims description 14
- 238000004364 calculation method Methods 0.000 claims description 11
- 238000012795 verification Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 7
- 238000013077 scoring method Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 6
- 239000003550 marker Substances 0.000 claims description 6
- 238000013499 data model Methods 0.000 claims description 3
- 230000006870 function Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 4
- 241000393496 Electra Species 0.000 description 3
- 238000007635 classification algorithm Methods 0.000 description 2
- 238000013500 data storage Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008520 organization Effects 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 229920001971 elastomer Polymers 0.000 description 1
- 239000000806 elastomer Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06311—Scheduling, planning or task assignment for a person or group
-
- 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
- G06F16/3332—Query translation
- G06F16/3334—Selection or weighting of terms from queries, including natural language queries
-
- 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
- G06F16/35—Clustering; Classification
- G06F16/353—Clustering; Classification into predefined classes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/194—Calculation of difference between files
-
- 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
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/0985—Hyperparameter optimisation; Meta-learning; Learning-to-learn
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06395—Quality analysis or management
-
- 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/18—Legal services; Handling legal documents
- G06Q50/182—Alternative dispute resolution
-
- 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
Abstract
The invention discloses a contradiction adjustment scheme recommending and scoring system and method, comprising the following steps: the contradiction case entry module is used for acquiring original data such as case description, solution, processing mechanism, reconciliation type, scoring and the like related to reconciliation of contradiction disputes; the data cleaning module is used for performing cleaning pretreatment operation on the original data; the deep learning modeling module trains a deep learning model based on the pre-processed data; the event mediation allocation processing module allocates a manual processing or automatic recommendation scheme according to the case complexity, and pre-allocates a processing window for the contradictory event principal; a dispute solution recommendation module for providing dispute solution recommendation for a dispute mediator; and the solution scoring module scores the mediation solution given by the personnel. The invention can improve the case processing efficiency of the contradiction mediator and help case parties to solve the problem of contradiction disputes in time.
Description
Technical Field
The invention belongs to the field of big data modeling analysis and application, and relates to a contradictory reconciliation scheme recommendation and scoring system and method.
Background
At present, the method of the contradiction and dispute adjustment system generally focuses on the aspects of information transmission links among modules and terminal scheduling management, but the method has less utilization of historical contradiction and dispute adjustment case data. This allows the contradicting moderator to have a lot of effort to comb the material to make his own judgment while handling the event, despite the availability of the multi-source information. Valuable information in the historical data is mined and deeply fused with a contradiction reconciliation process, so that the contradiction reconciliation bureau and the reconciliation commission can conveniently analyze cases by the contradiction reconciliation bureau at present, and an effective scheme can be adopted for efficiently processing the contradiction disputes.
In terms of a feature vector computing method and a natural language semantic understanding task, the existing scheme only stays on a shallow network and a simple framework based on a statistical model theory, for example, a word2vec algorithm used for computing text vectors in a patent CN 110188092A is a word embedding model based on a shallow language model, and deep semantic relation logic and a paragraph grammar structure cannot be understood. The local outlier factor (CBLOF) anomaly detection method based on the hypothesis clusters used by the method is weak for high-dimensional data processing capacity and cannot process nonlinear relations.
Disclosure of Invention
The invention aims at overcoming the defects of the prior art and provides a contradiction adjustment scheme recommending and scoring system and method.
In order to achieve the above object, the present invention provides a contradictory reconciliation scheme recommendation and scoring system, comprising:
the contradiction case entry module is used for acquiring the original data related to the contradiction dispute adjustment;
the data cleaning module is used for cleaning the original data to obtain effective data;
the deep learning modeling module trains a deep learning model through Chinese word segmentation, text embedding, cross attention mechanism and multi-objective optimization based on the effective data after the cleaning operation;
the event mediation allocation processing module is used for allocating a manual processing or automatic recommendation scheme according to the case complexity and pre-allocating a processing window for the contradictory event parties;
the contradiction dispute solution recommending module is used for building a search engine and recommending a solution for a moderator by combining a deep learning algorithm;
the solution scoring module is used for scoring the solutions given by the mediation staff;
and the data model iteration module is used for continuously accumulating new contradictory disputes and updating the iteration model.
Further, in the contradictory case entry module, the raw data includes a case description, a solution, a principal, a processing organization, a reconciliation type, and a score.
Further, the cleaning operation specifically includes:
removing useless data: the useless data comprises blank data which cannot be added, text data with text length less than 7 characters in a data sample, data with solution description in non-detail and data with completely repeated case and scheme;
and (5) checking text similarity: calculating the similarity of the two text sections, and judging whether the two text sections are similar or not according to a preset threshold value:
data enhancement: randomly deleting and replacing part of stop words in the text sample to obtain a new record, and replacing the entities in the text with similar entities according to the probability; replacing adjectives in the text with similar adjectives according to the probability;
and (3) data sampling: taking part of existing data field case descriptions and solutions as matching positive samples; then randomly extracting a solution from other samples for each case description, ensuring that the randomly found solution is different from the original solution through text similarity verification, and then pairing the case description with the unmatched solution to serve as a negative sample; finally, generating more negative samples based on case records with low solution scores by utilizing data enhancement;
positive sample regularization: removing positive sample data containing keywords or in a specific format according to regular matching; filtering out consultation cases with incomplete contents and arbitration data submitting cases through an event description field; filtering the ambiguous answers, insufficient records and transferring to cases processed by other departments through a solution field;
unified data field: and unifying multiple tables, classifying the fields with the same meaning into one field, and only reserving the fields required by model training.
Further, the processing procedure of the deep learning modeling module is as follows:
chinese word segmentation: dividing characters in a Chinese text into a single character form, and adding a starting and sentence-breaking marker;
text embedding: mapping Chinese text words into a multidimensional vector space by utilizing a pre-training model to obtain text representation;
cross-attention mechanism: embedding the details of the contradiction event description and two sections of texts of the contradiction solution into high-dimensional vector representation through text embedding, and performing attention calculation to obtain the correlation of the two sections of texts;
multi-objective optimization: and (5) optimizing the similarity of the two text sections and the solution score by taking the similarity and the solution score as model output targets.
Further, the processing procedure of the contradiction dispute solution recommendation module is as follows:
preprocessing a new contradiction dispute field: removing stop words and overlong numbers in the case description;
building a search engine: constructing a search engine based on the historical case descriptions and the solutions; searching new case descriptions in the constructed search engine, and finding out similar historical cases and solutions;
and (3) each rough recall result pair returned by the search engine is input into a deep learning model to obtain the matching degree score of each pair, and the first 10 are selected as recommended solutions according to the size sequence.
Further, the processing procedure of the solution scoring module is as follows: the description of the new case, the case type, the processing mechanism, the address information and the new case solution are input into the deep learning model, and then the score of the new case solution is obtained according to the prediction probability output of the deep learning model.
In order to achieve the above object, the present invention further provides a contradictory reconciliation scheme recommendation and scoring method, the method comprising the steps of:
(1) Acquiring original data related to a contradiction dispute adjustment case;
(2) Performing cleaning operation on the original data obtained in the step (1) to obtain effective data;
(3) Training a deep learning model through Chinese word segmentation, text embedding, cross attention mechanism and multi-objective optimization based on the effective data after the cleaning operation in the step (2);
(4) Distributing a manual processing or automatic recommendation scheme according to the case complexity, and pre-distributing a processing window for the contradictory event parties;
(5) Setting up a search engine and recommending a solution for a mediator by combining a deep learning algorithm;
(6) Solutions given by the moderator are scored.
Further, the step (2) includes the following substeps:
(2.1) removing useless data: the useless data comprises blank data which cannot be added, text data with text length less than 7 characters in a data sample, data with solution description in non-detail and data with completely repeated case and scheme;
(2.2) text similarity verification: calculating the similarity of the two text sections, and judging whether the two text sections are similar or not according to a preset threshold value:
(2.3) data enhancement: randomly deleting and replacing part of stop words in the text sample to obtain a new record, and replacing the entities in the text with similar entities according to the probability; replacing adjectives in the text with similar adjectives according to the probability;
(2.4) data sampling: taking part of existing data field case descriptions and solutions as matching positive samples; then randomly extracting a solution from other samples for each case description, ensuring that the randomly found solution is different from the original solution through text similarity verification, and then pairing the case description with the unmatched solution to serve as a negative sample; finally, generating more negative samples based on case records with low solution scores by utilizing data enhancement;
(2.5) positive sample regularization: removing positive sample data containing keywords or in a specific format according to regular matching; filtering out consultation cases with incomplete contents and arbitration data submitting cases through an event description field; filtering the ambiguous answers, insufficient records and transferring to cases processed by other departments through a solution field;
(2.6) unifying data fields: and unifying multiple tables, classifying the fields with the same meaning into one type, and only reserving the fields required by model training.
Further, the step (3) includes the following substeps:
(3.1) Chinese segmentation: dividing characters in a Chinese text into a single character form, and adding a starting and sentence-breaking marker;
(3.2) text embedding: mapping Chinese text words into a multidimensional vector space by utilizing a pre-training model to obtain text representation;
(3.3) Cross-attention mechanism: embedding the details of the contradiction event description and two sections of texts of the contradiction solution into high-dimensional vector representation through text embedding, and performing attention calculation to obtain the correlation of the two sections of texts;
(3.4) Multi-objective optimization: and (5) optimizing the similarity of the two text sections and the solution score by taking the similarity and the solution score as model output targets.
Further, the step (5) includes the sub-steps of:
(5.1) preprocessing a new contradictory dispute field: removing stop words and overlong numbers in the case description;
(5.2) building a search engine: constructing a search engine based on the historical case descriptions and the solutions; searching new case descriptions in the constructed search engine, and finding out similar historical cases and solutions;
and (5.3) each rough recall result pair returned by the search engine is input into the deep learning model to obtain the matching degree score of each pair, and the first 10 are taken as recommended solutions according to the size sequence.
Compared with the prior art, the invention has the beneficial effects that: the data cleaning operation in the invention removes noise aiming at the original data in the contradiction reconciliation field, so that the model can concentrate on main contents in the contradiction reconciliation type text. By introducing a deep language model-based pre-training model ELECTRA, the contradictory reconciliation history event records and the semantic information of the solution can be better characterized, providing an ability support for processing complex text data for downstream classification tasks. The introduced cross attention mechanism can enable the model to learn important information of contradiction solutions and events, so that the expressive capacity of the model is improved. The invention can improve the case processing efficiency of the contradiction mediator, and help case parties to solve the problem of contradiction disputes timely and more quickly.
Drawings
FIG. 1 is a block diagram of a system of the present invention;
FIG. 2 is a scheduling relationship and data flow diagram of the present invention;
FIG. 3 is a flowchart of a recommendation algorithm application in the present invention;
FIG. 4 is a flow chart of the use of the system of the present invention;
FIG. 5 is a block diagram of a deep learning model of the present invention;
FIG. 6 is a specific rule diagram of the contradictory dispute case text regularization in the present invention.
Detailed Description
In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions in the embodiments of the present invention are described in further detail below with reference to the accompanying drawings.
As shown in fig. 1, the algorithmic services in the system provided by the present invention may be provided in the form of network interfaces. The network service is realized by using a flash-based network service architecture by taking Python as a programming realization language. And transmitting contradictory case related information to obtain a required output result by sending an api request to a server from an external terminal.
As shown in fig. 2, the contradiction dispute handling system device provided by the invention provides data and algorithm support for contradiction parties on the terminal equipment and the manual window side, and provides algorithm recommendation service for dispute operators on the reconciliation terminal. For dispute reconciliation cases initiated by the principal on the terminal device, the system will determine the reconciliation terminal assigned to, or access the algorithm and return the results to the principal's terminal device. The event description, record place, type, time, and handling organization stated by the principal are stored in the database in a multi-sheet form, and the data storage part is provided with a data storage management service by the MariaDB database. The search engine in fig. 2 adopts an elastic search, so that rapid full-text search based on word segmentation can be realized.
The invention provides a contradiction reconciliation scheme recommendation and scoring system, comprising:
the contradiction case input module acquires original data such as case description, solution, principal, processing mechanism, mediation type, score and the like related to contradiction dispute mediation from the multi-source multi-terminal; the multi-source multi-terminal comprises off-line windows, network messages, intelligent voice mail channels and the like,
further, in the contradiction case entry module, the original contradiction case description information is recorded in an auxiliary mode through a manual window, internet uploading and a voice recognition algorithm.
The data cleaning module is used for cleaning the original data to obtain effective data; the processing procedure of the data cleaning module is as follows:
removing useless data, wherein the useless data comprise blank data which cannot be supplemented, text data with shorter text length (less than or equal to 6 characters) in a data sample, data with a solution description in no detail, and data with a case and a solution completely repeated;
and (5) checking text similarity: calculating the similarity of the two text sections, and judging whether the two text sections are similar or not according to a preset threshold value;
data enhancement: randomly deleting and replacing part of stop words in the text sample to obtain a new record; replacing the entities in the text with similar entities according to the probability; replacing adjectives in the text with similar adjectives according to the probability;
and (3) data sampling: describing and solving part of existing data field cases, and taking records with higher solution scores as matching positive samples; then randomly extracting a solution from other samples for each case description, ensuring that the randomly found solution is different from the original solution through text similarity verification, and then pairing the case description with the unmatched solution to serve as a negative sample; finally, generating more negative samples based on case records with low solution scores by using a data enhancement method;
positive sample regularization: firstly, carrying out regularization matching and filtering on the original data to obtain partial invalid cases; filtering and eliminating are respectively carried out on the case description field and the solution field by utilizing a specific regular matching rule field and a re.match () function and a re.search () function in the python toolkit re.
Referring to fig. 6, positive sample data containing a certain keyword or a specific format is removed according to regular matching; filtering out wide consultation cases (such as 'legal consultation' recorded only, specific consultation content and contradiction dispute detail missing) and cases such as arbitration data submission through an event description field; filtering ambiguous answers, insufficient records, and cases transferred to other departments for processing through solution fields;
unified data field: unifying multiple tables, classifying the fields with the same meaning into a data field, and only reserving the fields required by model training;
further, in the data sampling part, when a negative example sample of the event description and the unmatched solution is constructed, the system can randomly find other uncorrelated solutions according to one event description to be paired with the uncorrelated solutions to form a negative example sample, so as to avoid that the randomly found solutions are too similar to the proper solutions of the original event, the system calculates the similarity of the two solutions, and the system can be paired as the negative example sample only when the similarity of the two solutions is low. The method is characterized in that the similarity between MinHash and SimHash between the event description text and the randomly found solution is calculated, and two text sections with MinHash score greater than 0.1 or SimHash score greater than 0.65 are judged to be similar texts.
The similarity calculation method of MinHash comprises the following steps: two pieces of text A, B are known, the minimum unique hash codes of the vectorized A_vec and B_vec are calculated and added to the lists signature_A and signature_B respectively, and the hash codes have the following calculation formulas:
where a and b are randomly generated unique numbers,for the sample->For the vectorized coding of the sample i, p is a specified prime number; adding the HashCode to a list of signature_A or signature_B; finally, calculating the Jaccard similarity of the signature_A and the signature_B as the similarity of the texts A and B, wherein the Jaccard similarity calculation formula is as follows:
the similarity calculation method of the SimHash comprises the following steps: two pieces of text A, B are known, text is segmented, vectorized and weighted to represent the importance degree of each segmented segment, A, B of text is further processed by Hash coding to obtain 64-bit binary representation. And replacing 0 in the binary representation of the obtained word segmentation segment with-1, multiplying the binary representation with the weight, and then adding the binary representation with the weight in columns, wherein the positive number is 1, and the negative number is 0, so that the representations a and B of the text A and B are obtained. Further calculating the Hamming distance between a and B to obtain the SimHash similarity of A and B, wherein the formula is as follows:
further, in the data enhancement part, based on a case, the event description and the entity in the solution are subjected to random entity replacement, random synonym replacement, stop word random deletion and random hyponym replacement, and the method is realized by using an open-source Chinese data enhancement tool npcda.
The deep learning modeling module is used for training a deep learning model based on the effective data after the cleaning operation; the deep learning model framework is a model combining a pre-training model ELECTRA model coding, a convolution layer, an attention mechanism and a linear layer; the processing procedure of the deep learning modeling module is as follows:
chinese word segmentation: dividing characters in a section of Chinese text into a single character form, and adding a start and sentence-breaking marker;
text embedding: mapping Chinese text words into a multidimensional vector space by utilizing a pre-training model to obtain text representation;
cross-attention mechanism: embedding the details of the contradiction event description and two sections of texts of the contradiction solution into high-dimensional vector representation through text embedding, and performing attention calculation to obtain the correlation of the two sections of texts;
multi-objective optimization: and (5) optimizing the similarity of the two text sections and the solution score by taking the similarity and the solution score as model output targets.
The event mediation allocation processing module allocates a manual processing or automatic recommendation scheme according to the case complexity and pre-allocates a processing window for the contradictory event principal; wherein, the complexity of the case is automatically judged according to the classification algorithm, and a score is given; if the score is smaller than the preset threshold value of 0.3, judging that the case is low in complexity, and generating a recommended solution for the case for the principal to choose whether to adopt or not; if the principal does not adopt, the case is treated as contradiction with high complexity; if the score is greater than or equal to a preset threshold, judging that the case is high in complexity, and recommending a plurality of departments and window selectable items of the next flow for the principal; then the principal selects the department and window of the next flow; and then flows with the window according to the selected departments of the parties.
Further, the processing procedure of the event mediation distribution processing module is as follows:
contradictory event privacy protection: replacing sensitive words in the contradiction solution, such as personal names, numbers and place names;
contradictory event handling assignment: judging the complexity according to the case description; if the complex event is judged, directly carrying out manual processing; if the case is judged to be the uncomplicated case, automatically recommending a solution;
contradictory event classification: according to the case description, the case types are classified and related department windows are recommended, and the related department windows are transferred to a specific window for processing after being confirmed by a principal.
The contradiction dispute solution recommending module builds a search engine, combines a deep learning algorithm and provides a new dispute solution for a mediator;
the system finds out similar historical reconciliation cases according to event description information provided by the principal in combination with event types and processing mechanisms, and sorts the historical cases by using a model for reference by a mediator.
Further, the processing procedure of the contradiction dispute solution recommendation module is as follows:
preprocessing a new contradiction dispute field: removing stop words and overlong numbers in the case description;
search engine building and application: constructing a search engine based on the historical case descriptions and the solutions; searching new case descriptions in the constructed search engine, and finding out similar historical cases and solutions;
each rough recall result returned by the search engine is paired and input into a deep learning model trained by a deep learning modeling module to obtain a matching degree score (0-1) of each pairing, and the first 10 are selected as recommended schemes according to the size sequence;
the solution scoring module scores the solutions given by the personnel in real time; and inputting the description of the new case, the case type, the processing mechanism, the address information, the new case solution and other information into the deep learning model, and outputting according to the prediction probability of the deep learning model to obtain the score (1-5) of the new case solution.
Specifically, the moderator gives a solution with reference to similar historical reconciliation cases, the system will determine the relevance of the solution to the principal's description of the contradictory event, combine the principal's satisfaction prediction of the solution, and subtract the penalty factor calculated from the length L of the solutionIntegrating the solution scores;
when the solution is smaller than 9 Chinese characters, calculating the score of the solution, multiplying the predicted probability output of the deep learning model by 100 to obtain the score of the deep learning model, and subtracting the penalty factor P from the score of the deep learning model.
The data model iteration module is used for continuously accumulating new contradictory disputes and updating the model at regular time, so that the model effect is improved;
further, the system enters the new contradiction case information collected in one day into a search engine in the non-office time of the moderator, and then adds the data processed by the data cleaning module into a historical data set for training and updates an iteration model.
The processing process of the iteration module is as follows: continuously adding the new daily cases into a contradictory dispute case database; continuously adding new daily cases into a search engine; cleaning and preprocessing newly added data; and adding the data into the deep learning model training data of the deep learning modeling module, retraining and updating the iterative model.
Referring to fig. 2, in the deep learning model, the model training section performs fitting training on three sections:
(1) A contradictory dispute complexity scoring component;
(2) A contradictory dispute type and domain classification section;
(3) And a contradictory dispute solution recommendation and scoring part.
The network structure of the three parts is shown in fig. 5. The contradiction dispute complexity scoring part training data is input into a contradiction dispute description and a place where the contradiction dispute occurs, and is output into a manual label classification label for representing whether the contradiction dispute is complex or not.
The contradiction dispute type and the domain classification model training data are input into a contradiction dispute description and place of occurrence, and are output into a contradiction event type.
The recommendation and scoring part adopts a unified model multi-target optimization method, the structure of the recommendation and scoring part is shown in figure 5, text item contradiction disputes are described, a solution and a processing mechanism make word segmentation by using a pre-training model ELECTRA and encode the word segmentation into a high-dimensional vector, then the high-dimensional vector is input into a one-dimensional convolution layer and a dense layer to obtain a new text encoding vector, the vectorization of two sections of texts of the contradiction description and the solution is represented as cross attention and calculated to obtain relevance, and finally the obtained sequence is connected with other feature encoding vectors and then is input into a multi-layer perceptron and a final output target is obtained through a Softmax function.
The overall goals of the recommendation and scoring section are: by taking the description, the solution, the place of occurrence, the processing mechanism and the case type of a contradictory dispute case record as input features, judging whether the case description record is matched with the solution or not and judging the matching degree (how satisfactory the principal is). The training optimization targets of the recommendation and scoring part are set as the following two classification tasks:
task (1) contradictory matters describe whether or not a solution is relevant;
task (2) taking a score for the current solution for the current description;
referring to fig. 5, the loss functions of task (1) and task (2) both take Cross Entropy Loss Function, where the loss function of task (1) is two-classified, and its formula is:
wherein, the liquid crystal display device comprises a liquid crystal display device,the label of sample i (positive correlation; negative uncorrelation), is +.>Is the probability that sample i is predicted positive. Where L is the character length of the solution.
The loss function of task (2) is multi-classified, and its formula is:
wherein M is the number of scoring categories (1-5);0 or 1, if the class of sample i is equal to c +.>1, otherwiseIs 0; />The probability of category c is predicted for sample i.
The loss function trained by the multi-objective optimization part is the sum of the loss functions of the two classification tasks, and an Adam optimizer is adopted to update parameters. The model is deployed in a web service as shown in fig. 1, the model is preloaded in the service starting process, and the loaded model can be utilized to conduct reasoning prediction when the calling interface uses model prediction. The deep learning model is compatible with CPU deployment and GPU deployment, and the GPU deployment is adopted generally for achieving better reasoning performance.
As shown in FIG. 3, the training and application flow of the recommended algorithm includes collection of historical existing data, data cleansing, modeling training and building a search engine elastomer search. When a new contradictory dispute event scheme recommendation is implemented, firstly searching in a search engine elastic search according to multiple keywords, wherein different characteristic keywords correspond to the following different search clause types:
(1) The case description text is used as a clause type of 'must' to be searched, so that the queried related matters must meet the case description, and the score is calculated;
(2) And (3) taking the name of the processing mechanism and the case type as a 'shield' clause type to search, and calculating the score better for the search result obtained according to the description text.
Referring to fig. 3, when implementing the algorithm fine-ranking part, the solution in the items with the highest matching score of the 20 items retrieved by the search engine is matched with the description items, the processing structure and the case types in the new contradictory disputes to obtain 20 cases; further, the 20 cases are input into a contradictory dispute solution recommendation and scoring model to obtain matching scores of event descriptions and solutions in each case; further, after 20 cases are ranked according to the matching score from big to small, the first 10 cases are taken and the result is output.
The use flow chart of the system is shown in fig. 4, firstly, a party initiates a contradiction reconciliation event, and details of related contradiction disputes are recorded into a system and a database through a voice recognition algorithm or manual text recording. The system can automatically judge the complexity of the current contradictory event by applying a complexity scoring method according to the acquired related item detail information; further, a solution recommendation algorithm is applied to cases judged to be low in complexity, recommending a plurality of suitable solutions to the principal. When the complexity is high or the principal refuses to accept the recommended proposal, the contradiction event type classification algorithm is applied to recommend to the relevant acceptance department, and the contradiction moderator accepts the recommended proposal manually. Further, the new contradiction dispute solution recommendation method in fig. 4 is applied to provide references for the moderator.
In fig. 1, if a recommended scheme is required for a principal, when a corresponding recommended processing scheme is found, the event distribution processing module replaces sensitive information in the text with non-real items, such as person names with Zhang three and Liu four, and the identification card number information with character strings with all numbers 0.
The invention provides a contradiction reconciliation scheme recommendation and scoring method, which comprises the following steps:
(1) Acquiring original data related to a contradiction dispute adjustment case;
(2) Performing cleaning operation on the original data obtained in the step (1) to obtain effective data;
(2.1) removing useless data: the useless data comprises blank data which cannot be added, text data with text length less than 7 characters in a data sample, data with solution description in non-detail and data with completely repeated case and scheme;
(2.2) text similarity verification: calculating the similarity of the two text sections, and judging whether the two text sections are similar or not according to a preset threshold value:
(2.3) data enhancement: randomly deleting and replacing part of stop words in the text sample to obtain a new record, and replacing the entities in the text with similar entities according to the probability; replacing adjectives in the text with similar adjectives according to the probability;
(2.4) data sampling: taking part of existing data field case descriptions and solutions as matching positive samples; then randomly extracting a solution from other samples for each case description, ensuring that the randomly found solution is different from the original solution through text similarity verification, and then pairing the case description with the unmatched solution to serve as a negative sample; finally, generating more negative samples based on case records with low solution scores by utilizing data enhancement;
(2.5) positive sample regularization: removing positive sample data containing keywords or in a specific format according to regular matching; filtering out consultation cases with incomplete contents and arbitration data submitting cases through an event description field; filtering the ambiguous answers, insufficient records and transferring to cases processed by other departments through a solution field;
(2.6) unifying data fields: unifying multiple tables, classifying the fields with the same meaning into one type, and only reserving the fields required by model training;
(3) Training a deep learning model through Chinese word segmentation, text embedding, cross attention mechanism and multi-objective optimization based on the effective data after the cleaning operation in the step (2);
(3.1) Chinese segmentation: dividing characters in a Chinese text into a single character form, and adding a starting and sentence-breaking marker;
(3.2) text embedding: mapping Chinese text words into a multidimensional vector space by utilizing a pre-training model to obtain text representation;
(3.3) Cross-attention mechanism: embedding the details of the contradiction event description and two sections of texts of the contradiction solution into high-dimensional vector representation through text embedding, and performing attention calculation to obtain the correlation of the two sections of texts;
(3.4) Multi-objective optimization: the similarity of the two text sections and the solution score are used as model output targets together to be optimized;
(4) Distributing a manual processing or automatic recommendation scheme according to the case complexity, and pre-distributing a processing window for the contradictory event parties;
(5) Setting up a search engine and recommending a solution for a mediator by combining a deep learning algorithm;
(5.1) preprocessing a new contradictory dispute field: removing stop words and overlong numbers in the case description;
(5.2) building a search engine: constructing a search engine based on the historical case descriptions and the solutions; searching new case descriptions in the constructed search engine, and finding out similar historical cases and solutions;
(5.3) each rough recall result pair returned by the search engine is input into a deep learning model to obtain the matching degree score of each pair, and the first 10 are taken as recommended solutions according to the size sequence;
(6) For scoring solutions given by the moderator.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather to enable any modification, equivalent replacement, improvement or the like to be made within the spirit and principles of the invention.
The above embodiments are merely for illustrating the design concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement the same, the scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications according to the principles and design ideas of the present invention are within the scope of the present invention.
Claims (10)
1. A contradictory reconciliation proposal recommendation and scoring system comprising:
the contradiction case entry module is used for acquiring the original data related to the contradiction dispute adjustment;
the data cleaning module is used for cleaning the original data to obtain effective data;
the deep learning modeling module trains a deep learning model through Chinese word segmentation, text embedding, cross attention mechanism and multi-objective optimization based on the effective data after the cleaning operation;
the event mediation allocation processing module is used for allocating a manual processing or automatic recommendation scheme according to the case complexity and pre-allocating a processing window for the contradictory event parties;
the contradiction dispute solution recommending module is used for building a search engine and recommending a solution for a moderator by combining a deep learning algorithm;
the solution scoring module is used for scoring the solutions given by the mediation staff;
and the data model iteration module is used for continuously accumulating new contradictory disputes and updating the iteration model.
2. The contradictory reconciliation scheme recommendation and scoring system of claim 1, wherein in the contradictory case entry module, the raw data includes a case description, a solution, a principal, a processing agency, a reconciliation type and a score.
3. The contradictory reconciliation regimen recommendation and scoring system of claim 1, wherein the cleansing operation is specifically:
removing useless data: the useless data comprises blank data which cannot be added, text data with text length less than 7 characters in a data sample, data with solution description in non-detail and data with completely repeated case and scheme;
and (5) checking text similarity: calculating the similarity of the two text sections, and judging whether the two text sections are similar or not according to a preset threshold value:
data enhancement: randomly deleting and replacing part of stop words in the text sample to obtain a new record, and replacing the entities in the text with similar entities according to the probability; replacing adjectives in the text with similar adjectives according to the probability;
and (3) data sampling: taking part of existing data field case descriptions and solutions as matching positive samples; then randomly extracting a solution from other samples for each case description, ensuring that the randomly found solution is different from the original solution through text similarity verification, and then pairing the case description with the unmatched solution to serve as a negative sample; finally, generating more negative samples based on case records with low solution scores by utilizing data enhancement;
positive sample regularization: removing positive sample data containing keywords or in a specific format according to regular matching; filtering out consultation cases with incomplete contents and arbitration data submitting cases through an event description field; filtering the ambiguous answers, insufficient records and transferring to cases processed by other departments through a solution field;
unified data field: and unifying multiple tables, classifying the fields with the same meaning into one field, and only reserving the fields required by model training.
4. The contradictory reconciliation scheme recommendation and scoring system of claim 1, wherein the deep learning modeling module processes:
chinese word segmentation: dividing characters in a Chinese text into a single character form, and adding a starting and sentence-breaking marker;
text embedding: mapping Chinese text words into a multidimensional vector space by utilizing a pre-training model to obtain text representation;
cross-attention mechanism: embedding the details of the contradiction event description and two sections of texts of the contradiction solution into high-dimensional vector representation through text embedding, and performing attention calculation to obtain the correlation of the two sections of texts;
multi-objective optimization: and (5) optimizing the similarity of the two text sections and the solution score by taking the similarity and the solution score as model output targets.
5. The contradictory reconciliation scheme recommendation and scoring system of claim 4, wherein the contradictory dispute solution recommendation module processes:
preprocessing a new contradiction dispute field: removing stop words and overlong numbers in the case description;
building a search engine: constructing a search engine based on the historical case descriptions and the solutions; searching new case descriptions in the constructed search engine, and finding out similar historical cases and solutions;
and (3) each rough recall result pair returned by the search engine is input into a deep learning model to obtain the matching degree score of each pair, and the first 10 are selected as recommended solutions according to the size sequence.
6. The contradictory reconciliation solution recommendation and scoring system of claim 5, wherein the solution scoring module processes: the description of the new case, the case type, the processing mechanism, the address information and the new case solution are input into the deep learning model, and then the score of the new case solution is obtained according to the prediction probability output of the deep learning model.
7. A contradictory reconciliation scheme recommendation and scoring method, the method comprising the steps of:
(1) Acquiring original data related to a contradiction dispute adjustment case;
(2) Performing cleaning operation on the original data obtained in the step (1) to obtain effective data;
(3) Training a deep learning model through Chinese word segmentation, text embedding, cross attention mechanism and multi-objective optimization based on the effective data after the cleaning operation in the step (2);
(4) Distributing a manual processing or automatic recommendation scheme according to the case complexity, and pre-distributing a processing window for the contradictory event parties;
(5) Setting up a search engine and recommending a solution for a mediator by combining a deep learning algorithm;
(6) Solutions given by the moderator are scored.
8. The contradictory reconciliation proposal recommendation and scoring method of claim 7, wherein step (2) comprises the sub-steps of:
(2.1) removing useless data: the useless data comprises blank data which cannot be added, text data with text length less than 7 characters in a data sample, data with solution description in non-detail and data with completely repeated case and scheme;
(2.2) text similarity verification: calculating the similarity of the two text sections, and judging whether the two text sections are similar or not according to a preset threshold value:
(2.3) data enhancement: randomly deleting and replacing part of stop words in the text sample to obtain a new record, and replacing the entities in the text with similar entities according to the probability; replacing adjectives in the text with similar adjectives according to the probability;
(2.4) data sampling: taking part of existing data field case descriptions and solutions as matching positive samples; then randomly extracting a solution from other samples for each case description, ensuring that the randomly found solution is different from the original solution through text similarity verification, and then pairing the case description with the unmatched solution to serve as a negative sample; finally, generating more negative samples based on case records with low solution scores by utilizing data enhancement;
(2.5) positive sample regularization: removing positive sample data containing keywords or in a specific format according to regular matching; filtering out consultation cases with incomplete contents and arbitration data submitting cases through an event description field; filtering the ambiguous answers, insufficient records and transferring to cases processed by other departments through a solution field;
(2.6) unifying data fields: and unifying multiple tables, classifying the fields with the same meaning into one type, and only reserving the fields required by model training.
9. The contradictory reconciliation scheme recommendation and scoring method of claim 8, wherein step (3) comprises the sub-steps of:
(3.1) Chinese segmentation: dividing characters in a Chinese text into a single character form, and adding a starting and sentence-breaking marker;
(3.2) text embedding: mapping Chinese text words into a multidimensional vector space by utilizing a pre-training model to obtain text representation;
(3.3) Cross-attention mechanism: embedding the details of the contradiction event description and two sections of texts of the contradiction solution into high-dimensional vector representation through text embedding, and performing attention calculation to obtain the correlation of the two sections of texts;
(3.4) Multi-objective optimization: and (5) optimizing the similarity of the two text sections and the solution score by taking the similarity and the solution score as model output targets.
10. The contradictory reconciliation scheme recommendation and scoring method of claim 9, wherein step (5) comprises the sub-steps of:
(5.1) preprocessing a new contradictory dispute field: removing stop words and overlong numbers in the case description;
(5.2) building a search engine: constructing a search engine based on the historical case descriptions and the solutions; searching new case descriptions in the constructed search engine, and finding out similar historical cases and solutions;
and (5.3) each rough recall result pair returned by the search engine is input into the deep learning model to obtain the matching degree score of each pair, and the first 10 are taken as recommended solutions according to the size sequence.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311083864.XA CN116843162B (en) | 2023-08-28 | 2023-08-28 | Contradiction reconciliation scheme recommendation and scoring system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202311083864.XA CN116843162B (en) | 2023-08-28 | 2023-08-28 | Contradiction reconciliation scheme recommendation and scoring system and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116843162A true CN116843162A (en) | 2023-10-03 |
CN116843162B CN116843162B (en) | 2024-02-09 |
Family
ID=88162010
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202311083864.XA Active CN116843162B (en) | 2023-08-28 | 2023-08-28 | Contradiction reconciliation scheme recommendation and scoring system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116843162B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117390293A (en) * | 2023-12-12 | 2024-01-12 | 之江实验室 | Information recommendation method, device, medium and equipment for dispute cases |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220295A (en) * | 2017-04-27 | 2017-09-29 | 银江股份有限公司 | A kind of people's contradiction reconciles case retrieval and mediation strategy recommends method |
CN109726287A (en) * | 2018-12-25 | 2019-05-07 | 银江股份有限公司 | A kind of people's mediation case classification system and method based on transfer learning and deep learning |
CN109783639A (en) * | 2018-12-24 | 2019-05-21 | 银江股份有限公司 | A kind of conciliation case intelligence allocating method and system based on feature extraction |
CN110188092A (en) * | 2019-04-28 | 2019-08-30 | 浙江工业大学 | The system and method for novel contradiction and disputes in a kind of excavation people's mediation |
US10593431B1 (en) * | 2019-06-03 | 2020-03-17 | Kpn Innovations, Llc | Methods and systems for causative chaining of prognostic label classifications |
CN111709244A (en) * | 2019-11-20 | 2020-09-25 | 中共南通市委政法委员会 | Deep learning method for identifying causal relationship of contradictory dispute events |
US20210012394A1 (en) * | 2019-07-08 | 2021-01-14 | James Tolley | Online mediation platform |
CN114003721A (en) * | 2021-11-02 | 2022-02-01 | 城云科技(中国)有限公司 | Construction method, device and application of dispute event type classification model |
KR102539679B1 (en) * | 2023-02-01 | 2023-06-02 | (주)피플리 | Method, device and system for recommending places tailored to the user based on the user's route |
CN116612281A (en) * | 2023-05-20 | 2023-08-18 | 复旦大学 | Text supervision-based open vocabulary image semantic segmentation system |
-
2023
- 2023-08-28 CN CN202311083864.XA patent/CN116843162B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220295A (en) * | 2017-04-27 | 2017-09-29 | 银江股份有限公司 | A kind of people's contradiction reconciles case retrieval and mediation strategy recommends method |
CN109783639A (en) * | 2018-12-24 | 2019-05-21 | 银江股份有限公司 | A kind of conciliation case intelligence allocating method and system based on feature extraction |
CN109726287A (en) * | 2018-12-25 | 2019-05-07 | 银江股份有限公司 | A kind of people's mediation case classification system and method based on transfer learning and deep learning |
CN110188092A (en) * | 2019-04-28 | 2019-08-30 | 浙江工业大学 | The system and method for novel contradiction and disputes in a kind of excavation people's mediation |
US10593431B1 (en) * | 2019-06-03 | 2020-03-17 | Kpn Innovations, Llc | Methods and systems for causative chaining of prognostic label classifications |
US20210012394A1 (en) * | 2019-07-08 | 2021-01-14 | James Tolley | Online mediation platform |
CN111709244A (en) * | 2019-11-20 | 2020-09-25 | 中共南通市委政法委员会 | Deep learning method for identifying causal relationship of contradictory dispute events |
CN114003721A (en) * | 2021-11-02 | 2022-02-01 | 城云科技(中国)有限公司 | Construction method, device and application of dispute event type classification model |
KR102539679B1 (en) * | 2023-02-01 | 2023-06-02 | (주)피플리 | Method, device and system for recommending places tailored to the user based on the user's route |
CN116612281A (en) * | 2023-05-20 | 2023-08-18 | 复旦大学 | Text supervision-based open vocabulary image semantic segmentation system |
Non-Patent Citations (4)
Title |
---|
ALEXANDER BORGIDA: "Modeling Class Hierarchies with Contradictions", pages 434 - 443 * |
HEIKO MÜLLER, ULF LESER, JOHANN-CHRISTOPH FREYTAG: "Mining for Patterns in Contradictory Data", pages 51 - 58 * |
党振兴;: "人工智能与家事调解融合发展对策研究", 海峡法学, no. 04 * |
金涌涛: "基于深度学习的社会矛盾纠纷事件分类的 研究与应用", 中国优秀学位论文库, pages 1 - 57 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117390293A (en) * | 2023-12-12 | 2024-01-12 | 之江实验室 | Information recommendation method, device, medium and equipment for dispute cases |
CN117390293B (en) * | 2023-12-12 | 2024-04-02 | 之江实验室 | Information recommendation method, device, medium and equipment for dispute cases |
Also Published As
Publication number | Publication date |
---|---|
CN116843162B (en) | 2024-02-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020224097A1 (en) | Intelligent semantic document recommendation method and device, and computer-readable storage medium | |
Zhang et al. | A multi-label classification method using a hierarchical and transparent representation for paper-reviewer recommendation | |
CN111967761A (en) | Monitoring and early warning method and device based on knowledge graph and electronic equipment | |
CN108595525A (en) | A kind of lawyer's information processing method and system | |
CN116843162B (en) | Contradiction reconciliation scheme recommendation and scoring system and method | |
CN113886604A (en) | Job knowledge map generation method and system | |
CN108681548A (en) | A kind of lawyer's information processing method and system | |
CN110222192A (en) | Corpus method for building up and device | |
CN112182248A (en) | Statistical method for key policy of electricity price | |
CN112100396A (en) | Data processing method and device | |
CN111639183A (en) | Financial industry consensus public opinion analysis method and system based on deep learning algorithm | |
Tammina et al. | Sentiment analysis on customer reviews using convolutional neural network | |
Estevez-Velarde et al. | AutoML strategy based on grammatical evolution: A case study about knowledge discovery from text | |
She et al. | Joint learning with BERT-GCN and multi-attention for event text classification and event assignment | |
CN108681977A (en) | A kind of lawyer's information processing method and system | |
CN114048354B (en) | Test question retrieval method, device and medium based on multi-element characterization and metric learning | |
Zhang et al. | A hybrid neural network approach for fine-grained emotion classification and computing | |
Wang et al. | Sotagrec: A combined tag recommendation approach for stack overflow | |
Zhu et al. | Design of knowledge graph retrieval system for legal and regulatory framework of multilevel latent semantic indexing | |
CN113177164B (en) | Multi-platform collaborative new media content monitoring and management system based on big data | |
CN114862006A (en) | Social work service scheme automatic generation method and device based on artificial intelligence | |
CN115034762A (en) | Post recommendation method and device, storage medium, electronic equipment and product | |
Dai et al. | Intelligent audit question answering system based on knowledge graph and semantic similarity | |
Russo et al. | Graph-Based Approach for European Law Classification | |
Sefara et al. | Topic classification of tweets in the broadcasting domain using machine learning methods |
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 |