CN109933789A - A kind of judicial domain Relation extraction method and system neural network based - Google Patents
A kind of judicial domain Relation extraction method and system neural network based Download PDFInfo
- Publication number
- CN109933789A CN109933789A CN201910145396.1A CN201910145396A CN109933789A CN 109933789 A CN109933789 A CN 109933789A CN 201910145396 A CN201910145396 A CN 201910145396A CN 109933789 A CN109933789 A CN 109933789A
- Authority
- CN
- China
- Prior art keywords
- vector
- corpus
- judicial domain
- relation extraction
- feature set
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 68
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 34
- 239000013598 vector Substances 0.000 claims abstract description 125
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 16
- 230000001537 neural effect Effects 0.000 claims description 6
- 238000013480 data collection Methods 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 3
- 238000005457 optimization Methods 0.000 abstract 1
- 230000009466 transformation Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000003058 natural language processing Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000005086 pumping Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Machine Translation (AREA)
Abstract
A kind of judicial domain Relation extraction method and system neural network based, the present invention is in original open neural network Relation extraction frame foundation, construct judicial domain dedicated data set, and form judicial domain charge feature set, the method for improving Relation extraction accuracy rate by optimization neural network: first, it is netted from Chinese judgement document and obtains a large amount of judicial domain correlation non-structured texts, and with Word2Vec model, the vector that the vectors transformation model such as Bert model obtains text is indicated；Secondly, by non-structured text carry out TF-IDF word frequency statistics, obtain different charges and case by feature set, and obtain vector expression；Then optimize OpenNRE model and JointNRE model, obtain the higher JudNRE model of accuracy；Finally, charge feature vector vector is handled using JudNRE model to text vector, judicial domain Relation extraction model is obtained, for carrying out judicial domain Relation extraction to judicial domain unstructured text data to be processed, obtains corresponding entity triple.
Description
Technical field
The present invention relates to Relation extraction fields, more specifically to a kind of judicial domain relationship neural network based
Abstracting method and system.
Background technique
In recent years, with the rapid development of artificial intelligence technology, more and more research work are put to practical application
?.A kind of strong form of expression of the Relation extraction as natural language processing technique, also along with knowledge mapping research
Intimately it is concerned.For practicability, Relation extraction extracts useful information as one from non-structured text, quickly
The accurate technology for obtaining structured data information, can effectively mitigate the burden of manual analysis mass data text.
Instantly, it is based primarily upon single language text for the Relation extraction of specific area, and research surrounds English exhibition mostly
It opens.Procuratorate handles a case at present, and there is still a need for carry out artificial screening to a large amount of judgement document's text and criminal case folder text and divide
Analysis, and merit text and folder text are presented mostly in the form of half structure or pure urtext, and text categories are various, public procurator
The process checked takes time and effort and working efficiency is lower, and Relation extraction technology be to aid in user's rapid and convenient obtain information can
By medium.Therefore, judicial domain text is realized in conjunction with judicial domain text feature using existing Relation extraction the relevant technologies
Relation extraction, handling a case procuratorate's high efficiency high quality has great meaning.
Problems Existing:
Although the process of Relation extraction be obtain text data, text participle, natural language processing, entity to predict,
Relation inference (having used remote supervisory mostly), relationship probabilistic forecasting and etc., and training data more standardizes.But by
Some judicial domain judgement document texts are analyzed, it is known that it has following features:
1, text includes information multiplicity；
2, text locating is strong；
3, without external data as reference.
Complete judicial domain criminal case folder text is analyzed, it is known that it there are following features:
1, it is complete to cover information for folder；
2, information is interrelated between text in folder；
3, information to be extracted is more, only relies on nlp processing relatively difficult to achieve.
Whether therefore unpredictable judicial domain text data can be handled by conventional Relation extraction mode.
Summary of the invention
The technical problem to be solved in the present invention is that whether for unpredictable in the prior art conventional relationship can be passed through
Extraction mode handles the technological deficiency of judicial domain text data, provides a kind of judicial domain relationship neural network based
Abstracting method and system.
The present invention solves judicial domain Relation extraction method neural network based used by its technical problem, includes:
S1, judicial domain unstructured text data is obtained, then by Text Pretreatment, obtains the conclusion of each text
Part as corpus after participle, and generates the corresponding entity triple of each corpus；
S2, TF-IDF word frequency statistics are carried out to the judicial domain unstructured text data, obtains different charges and case
By the feature set constituted together；
S3, it respectively obtains and expects that the vector of collection and feature set indicates, then carry out corpus vector sum feature set vector
Splicing, so that obtaining the final sequence vector of corpus indicates；
S4, final sequence vector is indicated to carry out neural metwork training, obtains the Relation extraction model towards judicial domain；
S5, using the Relation extraction model, judicial neck is carried out to judicial domain unstructured text data to be processed
Domain Relation extraction obtains corresponding entity triple.
Further, it in the step S1 of judicial domain Relation extraction method neural network based of the invention, obtains
Judicial domain unstructured text data, which refers to from Chinese judgement document's net, crawls unstructured text data collection, sentences comprising criminal
Certainly two kinds of texts of book and criminal written verdict.
Further, in judicial domain Relation extraction method neural network based of the invention, feature set vector is
The term vector * (1+TF-IDF word frequency weight) of each word of corpus.
Further, it in the step S3 of judicial domain Relation extraction method neural network based of the invention, obtains
Expect that the method that the vector of collection and feature set indicates is to obtain vector using Word2Vec model or Bert model to indicate.
Further, in judicial domain Relation extraction method neural network based of the invention, corpus vector by
Word Embeddings and Position Embeddings are spliced, and Word Embeddings generates every corpus
Term vector, dimension are denoted as dw, Position Embeddings generates the position vector of every corpus, and dimension is denoted as dp；Feature
Integrate vector as the term vector * (1+TF-IDF word frequency weight) of each word of corpus, dimension is denoted as dti, the vector of feature set vector
Sequence is expressed as
The final sequence vector of corpus is expressed as w={ w1,w2,...,wm}；
Wherein, wi=Rd, d=dw+dp*2+dti。
The present invention is to solve its technical problem, additionally provides a kind of judicial domain Relation extraction neural network based system
System includes:
Corpus obtains module, for obtaining judicial domain unstructured text data, then by Text Pretreatment, obtains
It takes the conclusion part of each text, as corpus after participle, and generates the corresponding entity triple of each corpus；
Feature set obtains module, for carrying out TF-IDF word frequency statistics to the judicial domain unstructured text data,
Different charges and case are obtained by the feature set that constitutes together；
Final sequence vector representation module expects that the vector of collection and feature set indicates for respectively obtaining, then by corpus
Collection vector sum feature set vector is spliced, so that obtaining the final sequence vector of corpus indicates；
Model training module carries out neural metwork training for indicating final sequence vector, obtains towards judicial domain
Relation extraction model；
Relation extraction module, for utilizing the Relation extraction model, to judicial domain non-structured text to be processed
Data carry out judicial domain Relation extraction, obtain corresponding entity triple.
Further, module is obtained in the corpus of judicial domain Relation extraction system neural network based of the invention
In, acquisition judicial domain unstructured text data, which refers to from Chinese judgement document's net, crawls unstructured text data collection, wraps
Containing two kinds of texts of criminal judgment and criminal written verdict.
Further, in judicial domain Relation extraction system neural network based of the invention, feature set vector is
The term vector * (1+TF-IDF word frequency weight) of each word of corpus.
Further, in judicial domain Relation extraction system neural network based of the invention, final sequence vector
It obtains expecting that the method that the vector of collection and feature set indicates is to obtain using Word2Vec model or Bert model in representation module
It is indicated to vector.
Further, in judicial domain Relation extraction system neural network based of the invention, corpus vector by
Word Embeddings and Position Embeddings are spliced, and Word Embeddings generates every corpus
Term vector, dimension are denoted as dw, Position Embeddings generates the position vector of every corpus, and dimension is denoted as dp；Feature
Integrate vector as the term vector * (1+TF-IDF word frequency weight) of each word of corpus, dimension is denoted as dti, the vector of feature set vector
Sequence is expressed as
The final sequence vector of corpus is expressed as w={ w1,w2,...,wm}；
Wherein, wi=Rd, d=dw+dp*2+dti。
Implement judicial domain Relation extraction method and system neural network based of the invention, has below beneficial to effect
Fruit: present invention firstly provides open neural network Relation extraction is applied in judicial domain work, it is intended to pass through analysis department
Method field text semantic feature realizes the pumping to key message in extensive judicial domain text by the method for machine learning
It takes, meanwhile, charge feature vector is added in the present invention in Relation extraction neural network model for the first time, and multi-angle optimizes Relation extraction
Accurate rate.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples, in attached drawing:
Fig. 1 is the flow chart of one embodiment of judicial domain Relation extraction method neural network based of the invention；
Fig. 2 is the position vector schematic diagram of corpus；
Fig. 3 is the schematic diagram of one embodiment of judicial domain Relation extraction system neural network based.
Specific embodiment
For a clearer understanding of the technical characteristics, objects and effects of the present invention, now control attached drawing is described in detail
A specific embodiment of the invention.
It is the process of one embodiment of judicial domain Relation extraction method neural network based of the invention with reference to Fig. 1, Fig. 1
Figure.The present embodiment judicial domain Relation extraction method neural network based comprises the following steps:
S1, judicial domain unstructured text data is obtained, then by Text Pretreatment, obtains the conclusion of each text
Part as corpus after participle, and generates the corresponding entity triple of each corpus.
In this step, frame (Wenshu_Spider) is crawled first with open source judicial domain case document, is cut out from China
Sentence document and obtains two kinds of texts of criminal judgment and criminal written verdict on the net, it is main to extract the text containing defendant and the injured party
(can according to case by or charge define, common charge such as table 1).
The charge of 1 corresponding relationship containing PER2PER of table
The conclusion part in text is intercepted, using jieba, the participles tool such as KCWS is segmented, and extracts text based on NLP
The entity pair of PERSON type in this constitutes entity triple as label in conjunction with charge, and then forms corpus (label form
Such as entity triple " defendant-swindle-injured party ", in instances, defendant and the injured party are indicated by specific name, such as
" Mr. Wang-swindle-Lee ").
S2, TF-IDF word frequency statistics are carried out to the judicial domain unstructured text data, obtains different charges and case
By the feature set constituted together.Feature set mainly include every kind of charge and case by common vocabulary and word-building rule.
In this step, TF-IDF word frequency statistics are carried out to the urtext crawled, assesses the words in text for every
Class case by or charge text significance level, by crucial words obtain different charges and case by feature set.
For in the word in a certain specific file, its importance (i.e. word frequency) is indicated are as follows:
Wherein, ni,jIndicate the number that word i occurs in specific file j, ∑knk,jIndicate all in specific file j
The total degree that word occurs.
S3, it respectively obtains and expects that the vector of collection and feature set indicates, then carry out corpus vector sum feature set vector
Splicing, so that obtaining the final sequence vector of corpus indicates.
In step, model (such as Word2Vec model or Bert model) is generated using vector and obtains corpus and feature set
Vector indicate.Wherein Word2Vec model can directly train the vector for the fixed dimension for obtaining each word to indicate, using CBOW
Method (one kind of Continuous Bag-of-Words, Word2Vec model) does term vector to the description text of each entity
It generates.When model training, the dimension size that each term vector is arranged is 100, and suitable min-count and sliding is arranged
(the training input of CBOW model is the corresponding term vector of context-sensitive word of some Feature Words, Min-cout to window value
With sliding window value for controlling context length), it obtains indicating term vector.And the training of BERT model is divided into pre-training
(Pre-training) and fine tuning (Fine-tuning) two step, the present invention only use pre-training, generate the Chinese words of 768 dimensions to
Amount indicates.
Corpus vector sum feature set vector is spliced, obtaining the final sequence vector of corpus indicates, is denoted as w
(using Word2Vec model respectively, Bert model generates vector and indicates, compares for experimental result).Wherein, corpus vector
It is spliced by Word Embeddings and Position Embeddings, Word Embeddings generates every corpus
Term vector, dimension is denoted as dw, Position Embeddings generates the position vector of every corpus, and dimension is denoted as dp, often
The relative position of a word and entity pair, such as " president " in Fig. 2, the relative distance with head entity " king AA " is 4, with tail reality
The relative position of body " king BB " is -3)；Feature set vector is the term vector * (1+TF-IDF word frequency weight) of each word of corpus,
Dimension is denoted as dti(sequence vector of feature set is expressed as
The final sequence vector of corpus is expressed as w={ w1,w2,...,wm}；
Wherein, wi=Rd, d=dw+dp*2+dti。
S4, final sequence vector is indicated to carry out neural metwork training, obtains the Relation extraction model towards judicial domain.
For above-mentioned implementation steps S4, completed by step in detail below:
Feature extraction is carried out using the convolutional layer of convolutional neural networks.Convolutional layer sliding window is denoted as l, sentence
Embeddings length is denoted as dc, convolutional layer matrix-vector is expressed asBias vector is expressed as b.I-th of cunning
The sequence vector of dynamic window may be expressed as:
qi=wi-l+1:i(1≤i≤m+l-1)
The calculation formula of i-th of filter of convolutional layer are as follows:
pi=[Wq+b]i
Followed by maximum pond:
[x]i=max (pi)
Finally by non-linear layer tanh function and full articulamentum, the output vector o of convolutional layer is obtained.Non-linear layer are as follows: g
=tanh ([x]i), full articulamentum are as follows: o=Wg+b.
By the output vector o of convolutional layer by one softmax layers, the relationship probability of entity pair is predicted:
Wherein, nrFor relationship quantity, o is convolutional layer output vector.
In addition, the sequence vector of feature set is carried out the charge prediction of each corpus by attention mechanism.It is first
First calculate the attention weight of each charge:The charge possibility distrabtion of each corpus is calculated again:
zi=softmax (Wai+b)
Finally obtain charge prediction probability:
Optimal Parameters.The loss function setup of Relationship Prediction are as follows:
The loss function setup of charge prediction are as follows:
Final loss function setup are as follows:
L=Lrelation+α·Lcharge
Wherein, s is corpus quantity, and θ is all parameters of model, and α is hyper parameter, is set as 1 herein.
S5, using the Relation extraction model, judicial neck is carried out to judicial domain unstructured text data to be processed
Domain Relation extraction obtains corresponding entity triple.
With reference to Fig. 3, the present invention is to solve its technical problem, additionally provides a kind of judicial domain relationship neural network based
Extraction system obtains module 31 comprising corpus, feature set obtains module 32, final sequence vector representation module 33, model instruction
Practice module 34 and Relation extraction module 35.
Corpus obtains module 31 for obtaining judicial domain unstructured text data, then passes through Text Pretreatment,
It obtains the conclusion part of each text, as corpus after participle, and generates the corresponding entity triple of each corpus；
Feature set obtains module 32 and is used to carry out TF-IDF word frequency statistics to the judicial domain unstructured text data, obtains difference
Charge and case are by the feature set that constitutes together；Final sequence vector representation module 33 is for respectively obtaining expectation collection and feature set
Vector indicate, then corpus vector sum feature set vector is spliced, to obtain the final sequence vector of corpus
It indicates；Model training module 34 is used to indicate final sequence vector to carry out neural metwork training, obtains towards judicial domain
Relation extraction model；Relation extraction module 35 is used to utilize the Relation extraction model, non-structural to judicial domain to be processed
Change text data and carry out judicial domain Relation extraction, obtains corresponding entity triple.
Wherein, acquisition judicial domain unstructured text data, which refers to from Chinese judgement document's net, crawls non-structured text
Data set includes two kinds of texts of criminal judgment and criminal written verdict；Obtain expecting in final sequence vector representation module collection and
The method that the vector of feature set indicates is to obtain vector using Word2Vec model or Bert model to indicate.
Corpus vector is spliced by Word Embeddings and Position Embeddings, Word
Embeddings generates the term vector of every corpus, and dimension is denoted as dw, Position Embeddings generates every corpus
Position vector, dimension is denoted as dp；Feature set vector is the term vector * (1+TF-IDF word frequency weight) of each word of corpus, dimension
Degree is denoted as dti, the sequence vector of feature set vector is expressed as
The final sequence vector of corpus is expressed as w={ w1,w2,...,wm}；
Wherein, wi=Rd, d=dw+dp*2+dti。
The embodiment of the present invention is described with above attached drawing, but the invention is not limited to above-mentioned specific
Embodiment, the above mentioned embodiment is only schematical, rather than restrictive, those skilled in the art
Under the inspiration of the present invention, without breaking away from the scope protected by the purposes and claims of the present invention, it can also make very much
Form, all of these belong to the protection of the present invention.
Claims (10)
1. a kind of judicial domain Relation extraction method neural network based, characterized by comprising:
S1, judicial domain unstructured text data is obtained, then by Text Pretreatment, obtains the conclusion portion of each text
Point, as corpus after participle, and generate the corresponding entity triple of each corpus；
S2, TF-IDF word frequency statistics are carried out to the judicial domain unstructured text data, obtains different charges and case by institute
The feature set constituted together；
S3, it respectively obtains and expects that the vector of collection and feature set indicates, then splice corpus vector sum feature set vector,
It is indicated to obtain the final sequence vector of corpus；
S4, final sequence vector is indicated to carry out neural metwork training, obtains the Relation extraction model towards judicial domain；
S5, using the Relation extraction model, judicial domain pass is carried out to judicial domain unstructured text data to be processed
System extracts, and obtains corresponding entity triple.
2. judicial domain Relation extraction method neural network based according to claim 1, which is characterized in that step S1
In, acquisition judicial domain unstructured text data, which refers to from Chinese judgement document's net, crawls unstructured text data collection, wraps
Containing two kinds of texts of criminal judgment and criminal written verdict.
3. judicial domain Relation extraction method neural network based according to claim 1, which is characterized in that feature set
Vector is the term vector * (1+TF-IDF word frequency weight) of each word of corpus.
4. judicial domain Relation extraction method neural network based according to claim 1, which is characterized in that step S3
In obtain expecting that the method that the vector of collection and feature set indicates is using Word2Vec model or Bert model to obtain vector table
Show.
5. judicial domain Relation extraction method neural network based according to claim 1, which is characterized in that corpus
Vector is spliced by Word Embeddings and Position Embeddings, and Word Embeddings generates every language
Expect the term vector of collection, dimension is denoted as dw, Position Embeddings generates the position vector of every corpus, and dimension is denoted as
dp；Feature set vector is the term vector * (1+TF-IDF word frequency weight) of each word of corpus, and dimension is denoted as dti, feature set vector
Sequence vector be expressed as
The final sequence vector of corpus is expressed as w={ w1,w2,...,wm}；
Wherein, wi=Rd, d=dw+dp*2+dti。
6. a kind of judicial domain Relation extraction system neural network based, characterized by comprising:
Corpus obtains module, for obtaining judicial domain unstructured text data, then by Text Pretreatment, obtains every
The conclusion part of a text as corpus after participle, and generates the corresponding entity triple of each corpus；
Feature set obtains module, for carrying out TF-IDF word frequency statistics to the judicial domain unstructured text data, obtains
Different charges and case are by the feature set that constitutes together；
Final sequence vector representation module expects that the vector of collection and feature set indicates for respectively obtaining, then by corpus to
Amount and feature set vector are spliced, so that obtaining the final sequence vector of corpus indicates；
Model training module carries out neural metwork training for indicating final sequence vector, obtains the pass towards judicial domain
It is extraction model；
Relation extraction module, for utilizing the Relation extraction model, to judicial domain unstructured text data to be processed
Judicial domain Relation extraction is carried out, corresponding entity triple is obtained.
7. judicial domain Relation extraction system neural network based according to claim 6, which is characterized in that corpus
It obtains in module, acquisition judicial domain unstructured text data, which refers to from Chinese judgement document's net, crawls non-structured text number
It include two kinds of texts of criminal judgment and criminal written verdict according to collection.
8. judicial domain Relation extraction system neural network based according to claim 6, which is characterized in that feature set
Vector is the term vector * (1+TF-IDF word frequency weight) of each word of corpus.
9. judicial domain Relation extraction system neural network based according to claim 6, which is characterized in that finally to
It obtains expecting that the method that the vector of collection and feature set indicates is using Word2Vec model or Bert in amount sequence representation module
Model obtains vector expression.
10. judicial domain Relation extraction system neural network based according to claim 6, which is characterized in that corpus
Collection vector is spliced by Word Embeddings and Position Embeddings, and Word Embeddings generates every
The term vector of corpus, dimension are denoted as dw, Position Embeddings generates the position vector of every corpus, dimension note
For dp；Feature set vector is the term vector * (1+TF-IDF word frequency weight) of each word of corpus, and dimension is denoted as dti, feature set to
The sequence vector of amount is expressed as
The final sequence vector of corpus is expressed as w={ w1,w2,...,wm}；
Wherein, wi=Rd, d=dw+dp*2+dti。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910145396.1A CN109933789B (en) | 2019-02-27 | 2019-02-27 | Neural network-based judicial domain relation extraction method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910145396.1A CN109933789B (en) | 2019-02-27 | 2019-02-27 | Neural network-based judicial domain relation extraction method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109933789A true CN109933789A (en) | 2019-06-25 |
| CN109933789B CN109933789B (en) | 2021-04-13 |
Family
ID=66986028
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910145396.1A Expired - Fee Related CN109933789B (en) | 2019-02-27 | 2019-02-27 | Neural network-based judicial domain relation extraction method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109933789B (en) |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110570920A (en) * | 2019-08-20 | 2019-12-13 | 华东理工大学 | Entity and relationship joint learning method based on attention focusing model |
| CN110598846A (en) * | 2019-08-15 | 2019-12-20 | 北京航空航天大学 | Hierarchical recurrent neural network decoder and decoding method |
| CN110837558A (en) * | 2019-11-07 | 2020-02-25 | 成都星云律例科技有限责任公司 | Judgment document entity relation extraction method and system |
| CN111260114A (en) * | 2020-01-08 | 2020-06-09 | 昆明理工大学 | Low-frequency confusable criminal name prediction method for integrating case auxiliary sentence |
| CN111444704A (en) * | 2020-03-27 | 2020-07-24 | 中南大学 | Network security keyword extraction method based on deep neural network |
| CN111476035A (en) * | 2020-05-06 | 2020-07-31 | 中国人民解放军国防科技大学 | Chinese open relation prediction method and device, computer equipment and storage medium |
| CN111563374A (en) * | 2020-03-23 | 2020-08-21 | 北京交通大学 | Personnel social relationship extraction method based on judicial official documents |
| CN111680504A (en) * | 2020-08-11 | 2020-09-18 | 四川大学 | Legal information extraction model, method, system, device and auxiliary system |
| CN111798847A (en) * | 2020-06-22 | 2020-10-20 | 广州小鹏车联网科技有限公司 | Voice interaction method, server and computer-readable storage medium |
| CN112347780A (en) * | 2020-11-27 | 2021-02-09 | 浙江大学 | Judicial fact finding generation method, device and medium based on deep neural network |
| CN112396201A (en) * | 2019-07-30 | 2021-02-23 | 北京国双科技有限公司 | Criminal name prediction method and system |
| CN112579748A (en) * | 2019-09-30 | 2021-03-30 | 北京国双科技有限公司 | Method and device for extracting specific event relation from inquiry record |
| CN112764024A (en) * | 2020-12-29 | 2021-05-07 | 杭州电子科技大学 | Radar target identification method based on convolutional neural network and Bert |
| CN113011185A (en) * | 2020-07-17 | 2021-06-22 | 上海浦东华宇信息技术有限公司 | Legal field text analysis and identification method, system, storage medium and terminal |
| CN113779260A (en) * | 2021-08-12 | 2021-12-10 | 华东师范大学 | Domain map entity and relationship combined extraction method and system based on pre-training model |
| CN113901827A (en) * | 2021-12-10 | 2022-01-07 | 北京邮电大学 | Entity identification and relation extraction method and device, electronic equipment and storage medium |
| CN116304035A (en) * | 2023-02-28 | 2023-06-23 | 中国司法大数据研究院有限公司 | Multi-notice multi-crime name relation extraction method and device in complex case |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150169563A1 (en) * | 2013-12-13 | 2015-06-18 | National Cheng Kung University | Method and system for recommending research information news |
| JP2017091270A (en) * | 2015-11-11 | 2017-05-25 | 大日本印刷株式会社 | Information processing device, information processing system, and program |
| CN106952193A (en) * | 2017-03-23 | 2017-07-14 | 北京华宇信息技术有限公司 | A kind of criminal case aid decision-making method based on fuzzy depth belief network |
| CN108073673A (en) * | 2017-05-15 | 2018-05-25 | 北京华宇元典信息服务有限公司 | A kind of legal knowledge map construction method, apparatus, system and medium based on machine learning |
| CN108280058A (en) * | 2018-01-02 | 2018-07-13 | 中国科学院自动化研究所 | Relation extraction method and apparatus based on intensified learning |
| CN108763485A (en) * | 2018-05-25 | 2018-11-06 | 南京大学 | A kind of chain of evidence relational model construction method of the judgement document based on text similarity |
| CN108763483A (en) * | 2018-05-25 | 2018-11-06 | 南京大学 | A kind of Text Information Extraction method towards judgement document |
| CN109145071A (en) * | 2018-08-06 | 2019-01-04 | 中国地质大学（武汉） | A kind of automated construction method and system towards geophysics field knowledge mapping |
| CN109241285A (en) * | 2018-08-29 | 2019-01-18 | 东南大学 | A kind of device of the judicial decision in a case of auxiliary based on machine learning |
| WO2019023412A1 (en) * | 2017-07-26 | 2019-01-31 | Siuvo Inc. | Semantic classification numerical data in natural language context based on machine learning |
| CN109376963A (en) * | 2018-12-10 | 2019-02-22 | 杭州世平信息科技有限公司 | A kind of criminal case charge law article unified prediction neural network based |
-
2019
- 2019-02-27 CN CN201910145396.1A patent/CN109933789B/en not_active Expired - Fee Related
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150169563A1 (en) * | 2013-12-13 | 2015-06-18 | National Cheng Kung University | Method and system for recommending research information news |
| JP2017091270A (en) * | 2015-11-11 | 2017-05-25 | 大日本印刷株式会社 | Information processing device, information processing system, and program |
| CN106952193A (en) * | 2017-03-23 | 2017-07-14 | 北京华宇信息技术有限公司 | A kind of criminal case aid decision-making method based on fuzzy depth belief network |
| CN108073673A (en) * | 2017-05-15 | 2018-05-25 | 北京华宇元典信息服务有限公司 | A kind of legal knowledge map construction method, apparatus, system and medium based on machine learning |
| WO2019023412A1 (en) * | 2017-07-26 | 2019-01-31 | Siuvo Inc. | Semantic classification numerical data in natural language context based on machine learning |
| CN108280058A (en) * | 2018-01-02 | 2018-07-13 | 中国科学院自动化研究所 | Relation extraction method and apparatus based on intensified learning |
| CN108763485A (en) * | 2018-05-25 | 2018-11-06 | 南京大学 | A kind of chain of evidence relational model construction method of the judgement document based on text similarity |
| CN108763483A (en) * | 2018-05-25 | 2018-11-06 | 南京大学 | A kind of Text Information Extraction method towards judgement document |
| CN109145071A (en) * | 2018-08-06 | 2019-01-04 | 中国地质大学（武汉） | A kind of automated construction method and system towards geophysics field knowledge mapping |
| CN109241285A (en) * | 2018-08-29 | 2019-01-18 | 东南大学 | A kind of device of the judicial decision in a case of auxiliary based on machine learning |
| CN109376963A (en) * | 2018-12-10 | 2019-02-22 | 杭州世平信息科技有限公司 | A kind of criminal case charge law article unified prediction neural network based |
Non-Patent Citations (2)
| Title |
|---|
| 王林玉 等: "基于卷积神经网络和关键词策略的实体关系抽取方法", 《模式识别与人工智能》 * |
| 高丹 等: "海量法律文书中基于 CNN 的实体关系抽取技术", 《小型微型计算机系统》 * |
Cited By (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112396201A (en) * | 2019-07-30 | 2021-02-23 | 北京国双科技有限公司 | Criminal name prediction method and system |
| CN110598846A (en) * | 2019-08-15 | 2019-12-20 | 北京航空航天大学 | Hierarchical recurrent neural network decoder and decoding method |
| CN110598846B (en) * | 2019-08-15 | 2022-05-03 | 北京航空航天大学 | Hierarchical recurrent neural network decoder and decoding method |
| CN110570920B (en) * | 2019-08-20 | 2023-07-14 | 华东理工大学 | Entity and relationship joint learning method based on concentration model |
| CN110570920A (en) * | 2019-08-20 | 2019-12-13 | 华东理工大学 | Entity and relationship joint learning method based on attention focusing model |
| CN112579748A (en) * | 2019-09-30 | 2021-03-30 | 北京国双科技有限公司 | Method and device for extracting specific event relation from inquiry record |
| CN110837558A (en) * | 2019-11-07 | 2020-02-25 | 成都星云律例科技有限责任公司 | Judgment document entity relation extraction method and system |
| CN110837558B (en) * | 2019-11-07 | 2022-04-15 | 成都星云律例科技有限责任公司 | Judgment document entity relation extraction method and system |
| CN111260114A (en) * | 2020-01-08 | 2020-06-09 | 昆明理工大学 | Low-frequency confusable criminal name prediction method for integrating case auxiliary sentence |
| CN111260114B (en) * | 2020-01-08 | 2022-06-17 | 昆明理工大学 | Low-frequency confusable criminal name prediction method for integrating case auxiliary sentence |
| CN111563374A (en) * | 2020-03-23 | 2020-08-21 | 北京交通大学 | Personnel social relationship extraction method based on judicial official documents |
| CN111444704B (en) * | 2020-03-27 | 2023-09-19 | 中南大学 | Network safety keyword extraction method based on deep neural network |
| CN111444704A (en) * | 2020-03-27 | 2020-07-24 | 中南大学 | Network security keyword extraction method based on deep neural network |
| CN111476035A (en) * | 2020-05-06 | 2020-07-31 | 中国人民解放军国防科技大学 | Chinese open relation prediction method and device, computer equipment and storage medium |
| CN111476035B (en) * | 2020-05-06 | 2023-09-05 | 中国人民解放军国防科技大学 | Chinese open relation prediction method, device, computer equipment and storage medium |
| CN111798847A (en) * | 2020-06-22 | 2020-10-20 | 广州小鹏车联网科技有限公司 | Voice interaction method, server and computer-readable storage medium |
| CN113011185A (en) * | 2020-07-17 | 2021-06-22 | 上海浦东华宇信息技术有限公司 | Legal field text analysis and identification method, system, storage medium and terminal |
| CN111680504A (en) * | 2020-08-11 | 2020-09-18 | 四川大学 | Legal information extraction model, method, system, device and auxiliary system |
| CN112347780A (en) * | 2020-11-27 | 2021-02-09 | 浙江大学 | Judicial fact finding generation method, device and medium based on deep neural network |
| CN112347780B (en) * | 2020-11-27 | 2023-09-12 | 浙江大学 | Judicial fact finding generation method, device and medium based on deep neural network |
| CN112764024A (en) * | 2020-12-29 | 2021-05-07 | 杭州电子科技大学 | Radar target identification method based on convolutional neural network and Bert |
| CN113779260B (en) * | 2021-08-12 | 2023-07-18 | 华东师范大学 | Pre-training model-based domain map entity and relationship joint extraction method and system |
| CN113779260A (en) * | 2021-08-12 | 2021-12-10 | 华东师范大学 | Domain map entity and relationship combined extraction method and system based on pre-training model |
| CN113901827A (en) * | 2021-12-10 | 2022-01-07 | 北京邮电大学 | Entity identification and relation extraction method and device, electronic equipment and storage medium |
| CN116304035A (en) * | 2023-02-28 | 2023-06-23 | 中国司法大数据研究院有限公司 | Multi-notice multi-crime name relation extraction method and device in complex case |
| CN116304035B (en) * | 2023-02-28 | 2023-11-03 | 中国司法大数据研究院有限公司 | Multi-notice multi-crime name relation extraction method and device in complex case |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109933789B (en) | 2021-04-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109933789A (en) | A kind of judicial domain Relation extraction method and system neural network based | |
| Singh et al. | Vectorization of text documents for identifying unifiable news articles | |
| CN107168945B (en) | Bidirectional cyclic neural network fine-grained opinion mining method integrating multiple features | |
| WO2018218705A1 (en) | Method for recognizing network text named entity based on neural network probability disambiguation | |
| Zou et al. | A lexicon-based supervised attention model for neural sentiment analysis | |
| CN107818164A (en) | A kind of intelligent answer method and its system | |
| CN107679039A (en) | The method and apparatus being intended to for determining sentence | |
| CN110929030A (en) | Text abstract and emotion classification combined training method | |
| CN108255805A (en) | The analysis of public opinion method and device, storage medium, electronic equipment | |
| CN103631859A (en) | Intelligent review expert recommending method for science and technology projects | |
| CN112001186A (en) | Emotion classification method using graph convolution neural network and Chinese syntax | |
| CN108563703A (en) | A kind of determination method of charge, device and computer equipment, storage medium | |
| CN110597998A (en) | Military scenario entity relationship extraction method and device combined with syntactic analysis | |
| CN110851601A (en) | Cross-domain emotion classification system and method based on layered attention mechanism | |
| Ji et al. | SKGSUM: Abstractive document summarization with semantic knowledge graphs | |
| Feifei et al. | Bert-based Siamese network for semantic similarity | |
| Ji et al. | A deep learning method for named entity recognition in bidding document | |
| Wu et al. | TW-TGNN: Two windows graph-based model for text classification | |
| CN109117471A (en) | A kind of calculation method and terminal of the word degree of correlation | |
| CN115359323A (en) | Image text information generation method and deep learning model training method | |
| Ji et al. | Research on semantic similarity calculation methods in Chinese financial intelligent customer service | |
| Huang et al. | Structure graph refined information propagate network for aspect-based sentiment analysis | |
| Li et al. | DTGCN: a method combining dependency tree and graph convolutional networks for Chinese long-interval named entity relationship extraction | |
| CN112347251A (en) | Emotion analysis system based on attention mechanism and bidirectional LSTM (least squares Transmission technology) | |
| Rodrigues et al. | Domain adaptation of POS taggers without handcrafted features |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210413 |