CN111026880A - Joint learning-based judicial knowledge graph construction method - Google Patents
Joint learning-based judicial knowledge graph construction method Download PDFInfo
- Publication number
- CN111026880A CN111026880A CN201911254309.2A CN201911254309A CN111026880A CN 111026880 A CN111026880 A CN 111026880A CN 201911254309 A CN201911254309 A CN 201911254309A CN 111026880 A CN111026880 A CN 111026880A
- Authority
- CN
- China
- Prior art keywords
- criminal
- neural network
- word
- judgment
- judicial
- 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
- 238000010276 construction Methods 0.000 title claims abstract description 32
- 238000003062 neural network model Methods 0.000 claims abstract description 30
- 238000000034 method Methods 0.000 claims abstract description 24
- 239000013598 vector Substances 0.000 claims description 68
- 238000013528 artificial neural network Methods 0.000 claims description 28
- 238000013527 convolutional neural network Methods 0.000 claims description 27
- 238000012545 processing Methods 0.000 claims description 15
- 230000000306 recurrent effect Effects 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 9
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 125000004122 cyclic group Chemical group 0.000 claims description 6
- 239000003814 drug Substances 0.000 claims description 6
- 229940079593 drug Drugs 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000002372 labelling Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 241000700605 Viruses Species 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 238000002474 experimental method Methods 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000005065 mining Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012800 visualization 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- 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
- G06F16/355—Class or cluster creation or modification
-
- 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
- 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)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- Health & Medical Sciences (AREA)
- Technology Law (AREA)
- Human Resources & Organizations (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Animal Behavior & Ethology (AREA)
- Evolutionary Computation (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a judicial knowledge graph construction method, in particular to a judicial knowledge graph construction method based on joint learning, which comprises the following steps: (1) the method comprises the steps of (1) constructing a criminal judicial field body, (2) building a Seq2Seq neural network model, (3) extracting crime episode triples, and (4) storing in a graph database. The judicial knowledge graph construction method based on joint learning and the body format are feasible, the structure is clear, the reference value is achieved, the correlation information of criminal episodes and criminal results is mined, the accuracy of criminal suggestion can be improved, and the judicial knowledge graph based on the establishment can be applied to the aspects of judicial literature knowledge reasoning and intelligent retrieval of judicial services.
Description
Technical Field
The invention relates to a judicial knowledge graph construction method, in particular to a judicial knowledge graph construction method based on joint learning.
Background
Knowledge maps are effective tools for describing a large number of entities, entity attributes, and relationships between entities. In recent years, with the development of the internet, the knowledge graph is widely concerned, and compared with the knowledge graph in the general field which is subjected to a great amount of analysis and research in academia and industry, the knowledge graph in the vertical field is relatively less in construction method. The basic composition unit of the knowledge graph is an entity-relation-entity triple structure or an entity-attribute value triple structure, and all entities are connected with one another through relations to form a netlike graph structure. Knowledge is expressed in the form of a knowledge graph, and information which is difficult to understand can be displayed in a mining, analyzing and visualization mode, so that a user can conveniently acquire and understand concepts and relations thereof. Knowledge graph construction methods are generally divided into bottom-up methods and top-down methods. The bottom-up method flow is that related entities, attributes and interrelations among the entities are extracted from massive text data to obtain knowledge elements, then ambiguity among the entities is eliminated through processes of entity linking, knowledge merging and the like, a top-level ontology mode is automatically constructed in a data-driven mode for obtaining structured knowledge representation, the key of the bottom-up construction method is also the bottom-up construction method, and most of current general domain knowledge maps are constructed in the bottom-up mode. The top-down method flow is that an ontology and a data mode of a knowledge graph are defined firstly, then information such as entities and the like is filled into a knowledge base according to the defined mode, the top-down construction method is mainly used for constructing the vertical domain knowledge graph, certain domain knowledge is required to be used for guiding and defining an ontology structure, the domain range of data is required to be collected and the like, and then the construction of the vertical domain knowledge graph is realized through information extraction technologies such as named entity recognition, relation extraction and the like. At present, although knowledge map construction methods in many general fields are developed, research on the knowledge map construction method in the criminal judicial field is still in an exploration stage. At present, most of existing information extraction methods aiming at the judicial field obtain basic information of legal documents in a rule construction mode, structurally express each content of the legal documents, do not perform more detailed mining on the episode content of cases, and do not extract relevant information of criminal episodes and criminal results, so that the application in the aspects of criminal suggestion, criminal case recommendation and the like cannot be performed.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention aims to provide a judicial knowledge graph construction method based on joint learning. The method is based on professional knowledge in the judicial field and criminal judgment book text content, utilizes a top-down construction mode and combines a mainstream neural network deep learning algorithm and a joint learning algorithm to extract triples related in the criminal judgment book text, and stores the triples in a graph database mode.
In order to achieve the above purpose and solve the problems existing in the prior art, the invention adopts the technical scheme that: a judicial knowledge graph construction method based on joint learning comprises the following steps:
step 1, constructing a criminal judicial domain body, defining a domain body structure according to specific contents of a criminal judgment book of a drug-related case, extracting corresponding contents in a criminal judgment book text and filling, and specifically comprising the following substeps:
(a) defining a judicial case body structure according to the knowledge of the judicial professional field and the contents set forth in the criminal judgment book text of the drug-related cases, wherein the defined judicial case body structure comprises 10 parts of contents, namely document numbers, criminal judgment book titles, judgment places, judgment time, official complaints, defenders, crime types, crime plots, judgment results and judgment basis;
(b) respectively enabling information to be extracted to pass through manual construction rules according to a defined judicial case body structure, and supplementing and perfecting the existing manual construction rules by adopting an iterative evaluation mode so as to cover all information of each criminal judgment book and extracting contents of each part of the criminal judgment book by using the manual construction rules;
(c) expressing the judicial case body structure by adopting an XML format, filling contents of all parts in the extracted criminal judgment book, and designing a reading module aiming at the judicial case body structure so as to be convenient for subsequent calling and reading;
(a) determining the overall structure of a Seq2Seq neural network model, wherein the model mainly comprises two parts, namely an encoding layer based on a convolutional neural network and a decoding layer based on a cyclic neural network, and a softmax classification layer is arranged behind the decoding layer based on the cyclic neural network to obtain the prediction result of a final label sequence;
(b) the method comprises the following steps of building a coding layer based on a Convolutional Neural Network (CNN), wherein the coding layer comprises two convolutional neural network structures which are respectively used for coding words and words, obtaining coded text characteristic representation through vector splicing, and describing the process of coding the words by the coding layer based on the convolutional neural network through a formula (1) -a formula (3):
wc＝reshape(vc) (2)
where conv () denotes a convolution operation, cjAn initialization vector representing the jth word,
represents the convolution result of the j-th word, reshape () represents the conversion of the matrix shape, representing the character-level vector as vcConversion to word-level vector representation wc，
An initialization vector representing the ith word,
the character feature vector obtained by performing convolution operation on the word of the ith word is represented,
representing a vector splicing operation, wiRepresenting the coding result of the ith word by coding the word;
the process of encoding words based on the encoding layer of the convolutional neural network is described by formula (4) -formula (5):
hi＝conv(wi) (4)
where conv () denotes a convolution operation, wiRepresenting the coding result of the ith word by coding the word, hiRepresenting the word feature vector of the ith word obtained by performing convolution operation on the word,
a vector splicing operation is represented as a vector splicing operation,
representing a feature vector obtained by the ith word through a coding layer based on a convolutional neural network;
(c) and (b) building a decoding layer based on a Recurrent Neural Network (RNN), wherein the decoding layer uses a unidirectional long-short term memory neural network (LSTM), the characteristic vector of the coding layer based on the convolutional neural network obtained in the substep (b) is input, and an output characteristic vector is obtained through decoding of the long-short term memory neural network, and the process is described by a formula (6):
in the formula, LSTM () represents a calculation by a one-way long-short term memory neural network,
a feature vector representing the i-th word passing through the convolutional neural network-based coding layer,
a feature vector representing the i-th word passing through a Recurrent Neural Network (RNN) -based decoding layer;
(d) and performing normalization processing on the feature vector obtained by the decoding layer based on the recurrent neural network through linear mapping operation and by using a softmax function, wherein the normalization processing is described by a formula (7):
where Softmax () denotes a Softmax function, W denotes a parameter matrix of linear mapping,
feature vector, y, representing the i-th word through a Recurrent Neural Network (RNN) based decoding layeriAn output vector representing the i-th word through a Seq2Seq neural network model, each value of the vector representing the probability that the word belongs to each tag, by an output vector y for each wordiPerforming argmax operation to obtain a final prediction result of the tag sequence;
step 3, extracting a crime episode triple, establishing a Seq2Seq neural network model aiming at the criminal decision book text extracted in the step 1, and extracting the crime episode triple, wherein the method specifically comprises the following substeps:
(a) collecting and labeling the text content of the criminal scenario, labeling corresponding entities and the relationship between the entities according to the relationship type between the criminal scenario involved person and the involved articles to construct a data set required by the experiment, dividing the data set, and dividing a training set, a verification set and a test set according to the ratio of 6:2: 2;
(b) preprocessing crime episode text data to form a neural network model and perform vector representation, performing vector representation by adopting a random initialization mode for words, and expressing words by adopting word vectors for word2vec pre-training on criminal decision book texts, and combining a joint learning thought in a label strategy to enable a label to contain two kinds of information of entities and relationship types so as to prevent redundant entities from being identified;
(c) using the divided data set in the substep (a) of the step 3, predicting a label sequence by using a Seq2Seq neural network model built in the supervised learning training step 2 and the trained Seq2Seq neural network model, restoring natural language representation of the predicted label by indexing and inquiring a word list aiming at elements with the predicted label as an entity, determining a relation type according to label information, and finally extracting criminal plots and judgment results in a criminal judgment book text in a triple form;
step 4, storing the crime episode and judgment result triples related to the crime in a graph database Neo4j, and specifically comprising the following substeps:
(a) reading information stored in a body library in the criminal judicial field, and extracting judgment results of the same case corresponding to the criminal scenario extracted by the triple extraction in the step 3;
(b) preprocessing a judgment result of a current criminal suspect, dividing the judgment result into two parts of judgment contents, namely, criminal penalty related to a criminal period, namely, arrest, futile prison, no-term prison and death prison, and aiming at specific criminal period duration, processing the criminal period expressed by Chinese characters into Arabic numerals which are expressed in a form of year, month and day; secondly, penalty related penalty is realized by taking RMB yuan as a unit and processing the penalty expressed by Chinese characters into Arabic numbers;
(c) respectively processing the two parts of judgment contents into a triple form, and corresponding the triple with the crime festival to form association through a criminal suspect; the criminal plot triple and the judgment result triple are stored by adopting a graph database Neo4j, the triple is firstly processed into a csv format in consideration of storage efficiency, and then a graph database is imported to form a judicial knowledge graph of the virus-related cases in the criminal judicial field.
The invention has the beneficial effects that: a judicial knowledge graph construction method based on joint learning comprises the following steps: (1) the method comprises the steps of (1) constructing a criminal judicial field body, (2) building a Seq2Seq neural network model, (3) extracting crime episode triples, and (4) storing in a graph database. The judicial knowledge graph construction method based on joint learning and the body format are feasible, the structure is clear, the reference value is achieved, the correlation information of criminal episodes and criminal results is mined, the criminal suggestion accuracy can be improved, and the judicial service oriented application in the aspects of intelligent retrieval of official documents, classification recommendation and the like can be realized based on the established judicial knowledge graph.
Drawings
FIG. 1 is a flow chart of the method steps of the present invention.
FIG. 2 is a diagram of a Seq2Seq neural network model in accordance with the present invention.
Detailed Description
The invention will be further explained with reference to the drawings.
A judicial knowledge graph construction method based on joint learning comprises the following steps:
step 1, constructing a criminal judicial domain body, defining a domain body structure according to specific contents of a criminal judgment book of a drug-related case, extracting corresponding contents in a criminal judgment book text and filling, and specifically comprising the following substeps:
(a) defining a judicial case body structure according to the knowledge of the judicial professional field and the contents set forth in the criminal judgment book text of the drug-related cases, wherein the defined judicial case body structure comprises 10 parts of contents, namely document numbers, criminal judgment book titles, judgment places, judgment time, official complaints, defenders, crime types, crime plots, judgment results and judgment basis;
(b) respectively enabling information to be extracted to pass through manual construction rules according to a defined judicial case body structure, and supplementing and perfecting the existing manual construction rules by adopting an iterative evaluation mode so as to cover all information of each criminal judgment book and extracting contents of each part of the criminal judgment book by using the manual construction rules;
(c) expressing the judicial case body structure by adopting an XML format, filling contents of all parts in the extracted criminal judgment book, and designing a reading module aiming at the judicial case body structure so as to be convenient for subsequent calling and reading;
(a) determining the overall structure of a Seq2Seq neural network model, wherein the model mainly comprises two parts, namely an encoding layer based on a convolutional neural network and a decoding layer based on a cyclic neural network, and a softmax classification layer is arranged behind the decoding layer based on the cyclic neural network to obtain a final prediction result of a label sequence, as shown in fig. 2;
(b) the method comprises the following steps of building a coding layer based on a Convolutional Neural Network (CNN), wherein the coding layer comprises two convolutional neural network structures which are respectively used for coding words and words, obtaining coded text characteristic representation through vector splicing, and describing the process of coding the words by the coding layer based on the convolutional neural network through a formula (1) -a formula (3):
wc＝reshape(vc) (2)
where conv () denotes a convolution operation, cjAn initialization vector representing the jth word,
represents the convolution result of the j-th word, reshape () represents the conversion of the matrix shape, representing the character-level vector as vcConversion to word levelVector representation wc，
An initialization vector representing the ith word,
the character feature vector obtained by performing convolution operation on the word of the ith word is represented,
representing a vector splicing operation, wiRepresenting the coding result of the ith word by coding the word;
the process of encoding words based on the encoding layer of the convolutional neural network is described by formula (4) -formula (5):
hi＝conv(wi) (4)
where conv () denotes a convolution operation, wiRepresenting the coding result of the ith word by coding the word, hiRepresenting the word feature vector of the ith word obtained by performing convolution operation on the word,
a vector splicing operation is represented as a vector splicing operation,
representing a feature vector obtained by the ith word through a coding layer based on a convolutional neural network;
(c) and (b) building a decoding layer based on a Recurrent Neural Network (RNN), wherein the decoding layer uses a unidirectional long-short term memory neural network (LSTM), the characteristic vector of the coding layer based on the convolutional neural network obtained in the substep (b) is input, and an output characteristic vector is obtained through decoding of the long-short term memory neural network, and the process is described by a formula (6):
in the formula, LSTM () represents a calculation by a one-way long-short term memory neural network,
a feature vector representing the i-th word passing through the convolutional neural network-based coding layer,
a feature vector representing the i-th word passing through a Recurrent Neural Network (RNN) -based decoding layer;
(d) and performing normalization processing on the feature vector obtained by the decoding layer based on the recurrent neural network through linear mapping operation and by using a softmax function, wherein the normalization processing is described by a formula (7):
where Softmax () denotes a Softmax function, W denotes a parameter matrix of linear mapping,
feature vector, y, representing the i-th word through a Recurrent Neural Network (RNN) based decoding layeriAn output vector representing the i-th word through a Seq2Seq neural network model, each value of the vector representing the probability that the word belongs to each tag, by an output vector y for each wordiPerforming argmax operation to obtain a final prediction result of the tag sequence;
step 3, extracting a crime episode triple, establishing a Seq2Seq neural network model aiming at the criminal decision book text extracted in the step 1, and extracting the crime episode triple, wherein the method specifically comprises the following substeps:
(a) collecting and labeling the text content of the criminal scenario, labeling corresponding entities and the relationship between the entities according to the relationship type between the criminal scenario involved person and the involved articles to construct a data set required by the experiment, dividing the data set, and dividing a training set, a verification set and a test set according to the ratio of 6:2: 2;
(b) preprocessing crime episode text data to form a neural network model and perform vector representation, performing vector representation by adopting a random initialization mode for words, and expressing words by adopting word vectors for word2vec pre-training on criminal decision book texts, and combining a joint learning thought in a label strategy to enable a label to contain two kinds of information of entities and relationship types so as to prevent redundant entities from being identified;
(c) using the divided data set in the substep (a) of the step 3, predicting a label sequence by using a Seq2Seq neural network model built in the supervised learning training step 2 and the trained Seq2Seq neural network model, restoring natural language representation of the predicted label by indexing and inquiring a word list aiming at elements with the predicted label as an entity, determining a relation type according to label information, and finally extracting criminal plots and judgment results in a criminal judgment book text in a triple form;
step 4, storing the crime episode and judgment result triples related to the crime in a graph database Neo4j, and specifically comprising the following substeps:
(a) reading information stored in a body library in the criminal judicial field, and extracting judgment results of the same case corresponding to the criminal scenario extracted by the triple extraction in the step 3;
(b) preprocessing a judgment result of a current criminal suspect, dividing the judgment result into two parts of judgment contents, namely, criminal penalty related to a criminal period, namely, arrest, futile prison, no-term prison and death prison, and aiming at specific criminal period duration, processing the criminal period expressed by Chinese characters into Arabic numerals which are expressed in a form of year, month and day; secondly, penalty related penalty is realized by taking RMB yuan as a unit and processing the penalty expressed by Chinese characters into Arabic numbers;
(c) respectively processing the two parts of judgment contents into a triple form, and corresponding the triple with the crime festival to form association through a criminal suspect; the criminal plot triple and the judgment result triple are stored by adopting a graph database Neo4j, the triple is firstly processed into a csv format in consideration of storage efficiency, and then a graph database is imported to form a judicial knowledge graph of the virus-related cases in the criminal judicial field.
Claims (1)
1. A judicial knowledge graph construction method based on joint learning is characterized by comprising the following steps:
step 1, constructing a criminal judicial domain body, defining a domain body structure according to specific contents of a criminal judgment book of a drug-related case, extracting corresponding contents in a criminal judgment book text and filling, and specifically comprising the following substeps:
(a) defining a judicial case body structure according to the knowledge of the judicial professional field and the contents set forth in the criminal judgment book text of the drug-related cases, wherein the defined judicial case body structure comprises 10 parts of contents, namely document numbers, criminal judgment book titles, judgment places, judgment time, official complaints, defenders, crime types, crime plots, judgment results and judgment basis;
(b) respectively enabling information to be extracted to pass through manual construction rules according to a defined judicial case body structure, and supplementing and perfecting the existing manual construction rules by adopting an iterative evaluation mode so as to cover all information of each criminal judgment book and extracting contents of each part of the criminal judgment book by using the manual construction rules;
(c) expressing the judicial case body structure by adopting an XML format, filling contents of all parts in the extracted criminal judgment book, and designing a reading module aiming at the judicial case body structure so as to be convenient for subsequent calling and reading;
step 2, building a Seq2Seq neural network model, determining a model structure of the Seq2Seq neural network model, and initializing each parameter of the Seq2Seq neural network model, specifically comprising the following substeps:
(a) determining the overall structure of a Seq2Seq neural network model, wherein the model mainly comprises two parts, namely an encoding layer based on a convolutional neural network and a decoding layer based on a cyclic neural network, and a softmax classification layer is arranged behind the decoding layer based on the cyclic neural network to obtain the prediction result of a final label sequence;
(b) the method comprises the following steps of building a coding layer based on a Convolutional Neural Network (CNN), wherein the coding layer comprises two convolutional neural network structures which are respectively used for coding words and words, obtaining coded text characteristic representation through vector splicing, and describing the process of coding the words by the coding layer based on the convolutional neural network through a formula (1) -a formula (3):
wc＝reshape(vc) (2)
where conv () denotes a convolution operation, cjAn initialization vector representing the jth word,
represents the convolution result of the j-th word, reshape () represents the conversion of the matrix shape, representing the character-level vector as vcConversion to word-level vector representation wc，
An initialization vector representing the ith word,
the character feature vector obtained by performing convolution operation on the word of the ith word is represented,
representing a vector splicing operation, wiCoding by encoding words for the ith wordA code result;
the process of encoding words based on the encoding layer of the convolutional neural network is described by formula (4) -formula (5):
hi＝conv(wi) (4)
where conv () denotes a convolution operation, wiRepresenting the coding result of the ith word by coding the word, hiRepresenting the word feature vector of the ith word obtained by performing convolution operation on the word,
a vector splicing operation is represented as a vector splicing operation,
representing a feature vector obtained by the ith word through a coding layer based on a convolutional neural network;
(c) and (b) building a decoding layer based on a Recurrent Neural Network (RNN), wherein the decoding layer uses a unidirectional long-short term memory neural network (LSTM), the characteristic vector of the coding layer based on the convolutional neural network obtained in the substep (b) is input, and an output characteristic vector is obtained through decoding of the long-short term memory neural network, and the process is described by a formula (6):
in the formula, LSTM () represents a calculation by a one-way long-short term memory neural network,
a feature vector representing the i-th word passing through the convolutional neural network-based coding layer,
a feature vector representing the i-th word passing through a Recurrent Neural Network (RNN) -based decoding layer;
(d) and performing normalization processing on the feature vector obtained by the decoding layer based on the recurrent neural network through linear mapping operation and by using a softmax function, wherein the normalization processing is described by a formula (7):
where Softmax () denotes a Softmax function, W denotes a parameter matrix of linear mapping,
feature vector, y, representing the i-th word through a Recurrent Neural Network (RNN) based decoding layeriAn output vector representing the i-th word through a Seq2Seq neural network model, each value of the vector representing the probability that the word belongs to each tag, by an output vector y for each wordiPerforming argmax operation to obtain a final prediction result of the tag sequence;
step 3, extracting a crime episode triple, establishing a Seq2Seq neural network model aiming at the criminal decision book text extracted in the step 1, and extracting the crime episode triple, wherein the method specifically comprises the following substeps:
(a) collecting and labeling the text content of the criminal scenario, labeling corresponding entities and the relationship between the entities according to the relationship type between the criminal scenario involved person and the involved articles to construct a data set required by the experiment, dividing the data set, and dividing a training set, a verification set and a test set according to the ratio of 6:2: 2;
(b) preprocessing crime episode text data to form a neural network model and perform vector representation, performing vector representation by adopting a random initialization mode for words, and expressing words by adopting word vectors for word2vec pre-training on criminal decision book texts, and combining a joint learning thought in a label strategy to enable a label to contain two kinds of information of entities and relationship types so as to prevent redundant entities from being identified;
(c) using the divided data set in the substep (a) of the step 3, predicting a label sequence by using a Seq2Seq neural network model built in the supervised learning training step 2 and the trained Seq2Seq neural network model, restoring natural language representation of the predicted label by indexing and inquiring a word list aiming at elements with the predicted label as an entity, determining a relation type according to label information, and finally extracting criminal plots and judgment results in a criminal judgment book text in a triple form;
step 4, storing the crime episode and judgment result triples related to the crime in a graph database Neo4j, and specifically comprising the following substeps:
(a) reading information stored in a body library in the criminal judicial field, and extracting judgment results of the same case corresponding to the criminal scenario extracted by the triple extraction in the step 3;
(b) preprocessing a judgment result of a current criminal suspect, dividing the judgment result into two parts of judgment contents, namely, criminal penalty related to a criminal period, namely, arrest, futile prison, no-term prison and death prison, and aiming at specific criminal period duration, processing the criminal period expressed by Chinese characters into Arabic numerals which are expressed in a form of year, month and day; secondly, penalty related penalty is realized by taking RMB yuan as a unit and processing the penalty expressed by Chinese characters into Arabic numbers;
(c) respectively processing the two parts of judgment contents into a triple form, and corresponding the triple with the crime festival to form association through a criminal suspect; the criminal plot triple and the judgment result triple are stored by adopting a graph database Neo4j, the triple is firstly processed into a csv format in consideration of storage efficiency, and then a graph database is imported to form a judicial knowledge graph of the virus-related cases in the criminal judicial field.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911254309.2A CN111026880B (en) | 2019-12-08 | 2019-12-08 | Joint learning-based judicial knowledge graph construction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911254309.2A CN111026880B (en) | 2019-12-08 | 2019-12-08 | Joint learning-based judicial knowledge graph construction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111026880A true CN111026880A (en) | 2020-04-17 |
| CN111026880B CN111026880B (en) | 2021-06-08 |
Family
ID=70208228
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911254309.2A Active CN111026880B (en) | 2019-12-08 | 2019-12-08 | Joint learning-based judicial knowledge graph construction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111026880B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111597352A (en) * | 2020-05-18 | 2020-08-28 | 中国人民解放军国防科技大学 | Network space knowledge graph reasoning method and device combining ontology concept and example |
| CN112417155A (en) * | 2020-11-27 | 2021-02-26 | 浙江大学 | Court trial query generation method, device and medium based on pointer-generation Seq2Seq model |
| CN112632223A (en) * | 2020-12-29 | 2021-04-09 | 天津汇智星源信息技术有限公司 | Case and event knowledge graph construction method and related equipment |
| CN113239208A (en) * | 2021-05-06 | 2021-08-10 | 广东博维创远科技有限公司 | Mark training model based on knowledge graph |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107908671A (en) * | 2017-10-25 | 2018-04-13 | 南京擎盾信息科技有限公司 | Knowledge mapping construction method and system based on law data |
| CN108875051A (en) * | 2018-06-28 | 2018-11-23 | 中译语通科技股份有限公司 | Knowledge mapping method for auto constructing and system towards magnanimity non-structured text |
| CN109977237A (en) * | 2019-05-27 | 2019-07-05 | 南京擎盾信息科技有限公司 | A kind of dynamic law occurrence diagram spectrum construction method towards legal field |
| US20190332985A1 (en) * | 2017-09-22 | 2019-10-31 | 1Nteger, Llc | Systems and methods for investigating and evaluating financial crime and sanctions-related risks |
-
2019
- 2019-12-08 CN CN201911254309.2A patent/CN111026880B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190332985A1 (en) * | 2017-09-22 | 2019-10-31 | 1Nteger, Llc | Systems and methods for investigating and evaluating financial crime and sanctions-related risks |
| CN107908671A (en) * | 2017-10-25 | 2018-04-13 | 南京擎盾信息科技有限公司 | Knowledge mapping construction method and system based on law data |
| CN108875051A (en) * | 2018-06-28 | 2018-11-23 | 中译语通科技股份有限公司 | Knowledge mapping method for auto constructing and system towards magnanimity non-structured text |
| CN109977237A (en) * | 2019-05-27 | 2019-07-05 | 南京擎盾信息科技有限公司 | A kind of dynamic law occurrence diagram spectrum construction method towards legal field |
Non-Patent Citations (2)
| Title |
|---|
| GUODONG LI等: "Combining Domain Knowledge Extraction with Graph Long Short-Term Memory for Learning Classification of Chinese Legal Documents", 《IEEE》 * |
| 陈彦光等: "基于刑事案例的知识图谱构建技术", 《郑州大学学报（理学版）》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111597352A (en) * | 2020-05-18 | 2020-08-28 | 中国人民解放军国防科技大学 | Network space knowledge graph reasoning method and device combining ontology concept and example |
| CN111597352B (en) * | 2020-05-18 | 2023-05-09 | 中国人民解放军国防科技大学 | Network space knowledge graph reasoning method and device combining ontology concepts and instances |
| CN112417155A (en) * | 2020-11-27 | 2021-02-26 | 浙江大学 | Court trial query generation method, device and medium based on pointer-generation Seq2Seq model |
| CN112632223A (en) * | 2020-12-29 | 2021-04-09 | 天津汇智星源信息技术有限公司 | Case and event knowledge graph construction method and related equipment |
| CN112632223B (en) * | 2020-12-29 | 2023-01-20 | 天津汇智星源信息技术有限公司 | Case and event knowledge graph construction method and related equipment |
| CN113239208A (en) * | 2021-05-06 | 2021-08-10 | 广东博维创远科技有限公司 | Mark training model based on knowledge graph |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111026880B (en) | 2021-06-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021147726A1 (en) | Information extraction method and apparatus, electronic device and storage medium | |
| CN111026880B (en) | Joint learning-based judicial knowledge graph construction method | |
| CN111783394B (en) | Training method of event extraction model, event extraction method, system and equipment | |
| CN108182295A (en) | A kind of Company Knowledge collection of illustrative plates attribute extraction method and system | |
| CN107871158A (en) | A kind of knowledge mapping of binding sequence text message represents learning method and device | |
| CN113722490B (en) | Visual rich document information extraction method based on key value matching relation | |
| CN112749562A (en) | Named entity identification method, device, storage medium and electronic equipment | |
| CN112464669B (en) | Stock entity word disambiguation method, computer device, and storage medium | |
| CN113191148A (en) | Rail transit entity identification method based on semi-supervised learning and clustering | |
| CN112182248A (en) | Statistical method for key policy of electricity price | |
| CN116383399A (en) | Event public opinion risk prediction method and system | |
| CN113742733A (en) | Reading understanding vulnerability event trigger word extraction and vulnerability type identification method and device | |
| CN115292568B (en) | Civil news event extraction method based on joint model | |
| CN115292490A (en) | Analysis algorithm for policy interpretation semantics | |
| CN116245107A (en) | Electric power audit text entity identification method, device, equipment and storage medium | |
| CN115934883A (en) | Entity relation joint extraction method based on semantic enhancement and multi-feature fusion | |
| CN113505222A (en) | Government affair text classification method and system based on text circulation neural network | |
| CN116629258B (en) | Structured analysis method and system for judicial document based on complex information item data | |
| CN117033626A (en) | Text auditing method, device, equipment and storage medium | |
| CN117216617A (en) | Text classification model training method, device, computer equipment and storage medium | |
| CN115964497A (en) | Event extraction method integrating attention mechanism and convolutional neural network | |
| Mars et al. | Combination of DE-GAN with CNN-LSTM for Arabic OCR on Images with Colorful Backgrounds | |
| CN115204140A (en) | Legal provision prediction method based on attention mechanism and knowledge graph | |
| CN114911940A (en) | Text emotion recognition method and device, electronic equipment and storage medium | |
| CN115309862A (en) | Causal relationship identification method and device based on graph convolution network and contrast learning |
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 |