CN109376964A - A kind of criminal case charge prediction technique based on Memory Neural Networks - Google Patents
A kind of criminal case charge prediction technique based on Memory Neural Networks Download PDFInfo
- Publication number
- CN109376964A CN109376964A CN201811505665.2A CN201811505665A CN109376964A CN 109376964 A CN109376964 A CN 109376964A CN 201811505665 A CN201811505665 A CN 201811505665A CN 109376964 A CN109376964 A CN 109376964A
- Authority
- CN
- China
- Prior art keywords
- charge
- neural networks
- memory neural
- key
- criminal case
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 title claims abstract description 17
- 238000003062 neural network model Methods 0.000 claims abstract description 7
- 238000013527 convolutional neural network Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 238000003780 insertion Methods 0.000 claims description 3
- 230000037431 insertion Effects 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 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
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
-
- 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)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Tourism & Hospitality (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- General Health & Medical Sciences (AREA)
- Marketing (AREA)
- Data Mining & Analysis (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Technology Law (AREA)
- Primary Health Care (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The criminal case charge prediction technique based on Memory Neural Networks that the invention discloses a kind of, comprising the following steps: Step 1: building training dataset, obtains merit description and charge as training data;Step 2: constructive memory neural network model and being trained by training data;The key-value stored in Memory Neural Networks model is right to i.e. " merit Expressive Features vector "-" charge coding ";Step 3: being judged by Memory Neural Networks model after training to criminal case charge.Model of the invention also can be carried out good prediction to low frequency charge, and the present invention provides a set of charge prediction frameworks end to end, and strong reference can be provided to judge, improve judicial automation and intelligence degree.
Description
Technical field
The present invention relates to judicial intelligent fields, and in particular to a kind of criminal case charge based on Memory Neural Networks is pre-
Survey method.
Background technique
Currently, the charge prediction of criminal case is generally seen as a text classification problem: merit description is as to be sorted
Text, and then its corresponding charge trains a SVM or neural network model to be divided as corresponding tag along sort
Class.
Training dataset of the existing general Text Classification System dependent on sufficient training data and categorical measure balance,
However, there are natural unbalanced phenomenas for the distribution of different classes of case, for example " theft " crime might have in the database
Thousands of examples, and " crime of concealing deposits abroad " may only have several in the database, and its frequency distribution is a long-tail
Distribution, i.e. high frequency charge is actually less, and most of charge is that frequency is relatively low.Categorical data amount is very few to will cause mould
Type study it is insufficient, and data distribution imbalance then makes model tend to the classification more than forecast sample quantity, the two are former
Because making the file classification method of current main-stream that can only obtain considerable effect in a small amount of high frequency charge in such a scenario.
Summary of the invention
It is an object of the invention to be directed to above-mentioned the problems of the prior art, a kind of punishment based on Memory Neural Networks is provided
Thing case charge prediction technique is also able to carry out good prediction to low frequency charge, promotes judicial automation and intelligence degree.
To achieve the goals above, the technical solution adopted by the present invention the following steps are included:
Step 1: building training dataset, obtains merit description and charge as training data;
Step 2: constructive memory neural network model and being trained by training data;
The key-value stored in Memory Neural Networks model is right to i.e. " merit Expressive Features vector "-" charge coding ";
Step 3: being judged by Memory Neural Networks model after training to criminal case charge.
The step 1 crawls criminal written verdict from Chinese judgement document net, obtains merit description and charge as instruction
Practice data.
The specific steps of constructive memory neural network model include: in the step 2
Step 2-1) merit is described to segment, and it is mapped as term vector sequence;
Step 2-2) do one-dimensional convolutional neural networks in term vector sequence and extract feature, obtain the feature of merit description to
Measure q;
Step 2-3) by feature vector q input key-value memory module progress K- neighborhood matching, calculate separately q and each key
Cosine similarity, select the maximum key of K cosine similarity simultaneously to take its encoded radio, select K be worth in the most value of quantity;
Step 2-4) by obtained encoded radio be converted to its correspond to charge.
The step 2-1) in participle using Tsinghua University Open-Source Tools thulac, mapping follows directly after model instruction
Practice.
When being trained in the step 2 by training data, for a training sample, if it predicts charge and mark
Quasi- charge is inconsistent, then using its feature vector as key, charge coding as value insertion key-value memory module;If predicting charge
It is consistent with standard charge, then it is sought averagely by its feature vector and with its most like corresponding key of identical charge, and turn back to original
Position.
When being trained in the step 2 by training data, when being inserted into a new feature vector, but remember
When module has taken, selection abandons the position not updated at most in memory module, replaces into new key-value pair.
Compared with prior art, the technical solution adopted by the present invention are as follows: using the description of the merit of standard and its charge as instructing
Practice data and build training dataset, the Memory Neural Networks model constructed is trained by training dataset, " merit description
Feature vector "-" charge coding " is to the key-value pair stored in Memory Neural Networks model is converted to, using multi-layer perception (MLP) point
Class device judges criminal case charge, and model of the invention also can be carried out good prediction to low frequency charge, provides a set of
Charge prediction framework end to end can provide strong reference to judge, improve judicial automation and intelligence degree.
Detailed description of the invention
Fig. 1 criminal case charge prediction technique flow diagram of the present invention.
Specific embodiment
Present invention will be described in further detail below with reference to the accompanying drawings.
Referring to Fig. 1, the present invention is based on the criminal case charge prediction technique of Memory Neural Networks the following steps are included:
Step 1) constructs training dataset;
Criminal written verdict is crawled from Chinese judgement document net, obtains merit description and charge as training data;If
Merit, which describes corresponding charge, to be had multinomial, takes its main charge, each charge is mapped as a unique integer as it later
Coding.
Step 2) constructs and training Memory Neural Networks model;
A. building neural network model specifically includes:
Step 2-1) merit is described to segment, and it is mapped as term vector;Participle is using Tsinghua University's Open-Source Tools
Thulac, since term vector is related with field in the task, without using the term vector of pre-training, which follows directly after model instruction
Practice.
Step 2-2) one-dimensional convolutional neural networks extraction feature is done in the term vector sequence that step 2-1) is obtained, it must appear in court
The feature vector q of feelings description;Other feature extraction networks can also be used, for example extract temporal aspect using LSTM.But at this
In task, compared to other network structures, such as Recognition with Recurrent Neural Network, convolutional neural networks are extracting feature more directly effectively.
Step 2-3) the feature vector q that step 2-2) is obtained enter key-value memory module carry out K- neighborhood matching, matching
Method is that q and each key calculate cosine similarity respectively, selects this maximum key of K cosine similarity, obtains its value (herein i.e.
For the coding of charge classification), select the value that quantity is most in this K value;
The key-value stored in memory module is right to actually " merit Expressive Features vector "-" charge coding ".
Step 2-4) its correspondence charge will be converted in encoded radio obtained in step 2-3).
B. the memory module training step of neural network includes:
Step 3-1) to a training sample, if it predicts that charge and standard charge are inconsistent, its feature vector is made
For key, charge coding is as value insertion key-value memory module;If predicting, charge is consistent with standard charge, by its feature vector
It is averaged with its most like corresponding key of identical charge, and turns back to original position;
Step 3-2) in step 3-1), when being inserted into a new feature vector, and when memory module has taken, selection
The position not updated at most in memory module is abandoned, is replaced into new key-value pair.
Claims (7)
1. a kind of criminal case charge prediction technique based on Memory Neural Networks, which comprises the following steps:
Step 1: building training dataset, obtains merit description and charge as training data;
Step 2: constructive memory neural network model and being trained by training data;
The key-value stored in Memory Neural Networks model is right to i.e. " merit Expressive Features vector "-" charge coding ";
Step 3: being judged by Memory Neural Networks model after training to criminal case charge.
2. the criminal case charge prediction technique based on Memory Neural Networks according to claim 1, it is characterised in that: described
The step of one crawl criminal written verdict from Chinese judgement document net, obtain merit description and charge as training data.
3. the criminal case charge prediction technique based on Memory Neural Networks according to claim 1, it is characterised in that: if case
Feelings, which describe corresponding charge, to be had multinomial, takes its main charge, and each charge is mapped as a unique integer as its volume later
Code.
4. the criminal case charge prediction technique based on Memory Neural Networks according to claim 1, which is characterized in that described
The specific steps of constructive memory neural network model include: in step 2
Step 2-1) merit is described to segment, and it is mapped as term vector sequence;
Step 2-2) one-dimensional convolutional neural networks extraction feature is done in term vector sequence, obtain the feature vector q of merit description;
Step 2-3) feature vector q input key-value memory module is subjected to K- neighborhood matching, calculate separately the remaining of q and each key
String similarity selects the maximum key of K cosine similarity and takes its encoded radio, selects the value that quantity is most in K value;
Step 2-4) by obtained encoded radio be converted to its correspond to charge.
5. the criminal case charge prediction technique based on Memory Neural Networks according to claim 4, it is characterised in that: described
Step 2-1) in participle using Tsinghua University Open-Source Tools thulac, mapping follows directly after model training.
6. the criminal case charge prediction technique based on Memory Neural Networks according to claim 1, which is characterized in that described
When being trained in step 2 by training data, for a training sample, if it predicts that charge and standard charge are inconsistent,
Then using its feature vector as key, charge coding as value insertion key-value memory module;If predicting charge and standard charge one
It causes, is then sought averagely by its feature vector and with its most like corresponding key of identical charge, and turn back to original position.
7. the criminal case charge prediction technique based on Memory Neural Networks according to claim 1, which is characterized in that described
When being trained in step 2 by training data, when being inserted into a new feature vector, but when memory module has taken,
Selection abandons the position not updated at most in memory module, replaces into new key-value pair.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811505665.2A CN109376964B (en) | 2018-12-10 | 2018-12-10 | Criminal case criminal name prediction method based on memory neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811505665.2A CN109376964B (en) | 2018-12-10 | 2018-12-10 | Criminal case criminal name prediction method based on memory neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109376964A true CN109376964A (en) | 2019-02-22 |
CN109376964B CN109376964B (en) | 2021-11-12 |
Family
ID=65373147
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811505665.2A Active CN109376964B (en) | 2018-12-10 | 2018-12-10 | Criminal case criminal name prediction method based on memory neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109376964B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN112396201A (en) * | 2019-07-30 | 2021-02-23 | 北京国双科技有限公司 | Criminal name prediction method and system |
CN113515631A (en) * | 2021-06-18 | 2021-10-19 | 深圳大学 | Method, device, terminal equipment and storage medium for predicting criminal name |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108846384A (en) * | 2018-07-09 | 2018-11-20 | 北京邮电大学 | Merge the multitask coordinated recognition methods and system of video-aware |
-
2018
- 2018-12-10 CN CN201811505665.2A patent/CN109376964B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108846384A (en) * | 2018-07-09 | 2018-11-20 | 北京邮电大学 | Merge the multitask coordinated recognition methods and system of video-aware |
Non-Patent Citations (6)
Title |
---|
ALEXANDER H. MILLER 等: "Key-Value Memory Networks for Directly Reading Documents", 《ARXIV》 * |
BINGFENG LUO 等: "Learning to Predict Charges for Criminal Cases with Legal Basis", 《ARXIV》 * |
HAOXI ZHONG 等: "Legal Judgment Prediction via Topological Learning", 《RESEARCHGATE》 * |
YATIAN SHEN 等: "Legal Article-Aware End-To-End Memory Network for Charge Prediction", 《CSAE2018》 * |
ZIKUN HU 等: "Few-Shot Charge Prediction with Discriminative Legal Attributes", 《PROCEEDINGS OF THE 27TH INTERNATIONAL CONFERENCE ON COMPUTATIONAL LINGUISTICS》 * |
邹晓玫 等: "基于神经网络的刑事案件量刑决策系统", 《软件时空》 * |
Cited By (5)
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 |
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 |
CN113515631A (en) * | 2021-06-18 | 2021-10-19 | 深圳大学 | Method, device, terminal equipment and storage medium for predicting criminal name |
CN113515631B (en) * | 2021-06-18 | 2024-05-17 | 深圳大学 | Method, device, terminal equipment and storage medium for predicting crime name |
Also Published As
Publication number | Publication date |
---|---|
CN109376964B (en) | 2021-11-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109376963A (en) | A kind of criminal case charge law article unified prediction neural network based | |
CN107330396B (en) | Pedestrian re-identification method based on multi-attribute and multi-strategy fusion learning | |
CN104965905B (en) | A kind of method and apparatus of Web page classifying | |
CN110442684A (en) | A kind of class case recommended method based on content of text | |
CN109376964A (en) | A kind of criminal case charge prediction technique based on Memory Neural Networks | |
CN109413028A (en) | SQL injection detection method based on convolutional neural networks algorithm | |
CN110334272A (en) | The intelligent answer method, apparatus and computer storage medium of knowledge based map | |
CN109471938A (en) | A kind of file classification method and terminal | |
CN109522304A (en) | Exception object recognition methods and device, storage medium | |
CN103970733B (en) | A kind of Chinese new word identification method based on graph structure | |
CN110532417A (en) | Image search method, device and terminal device based on depth Hash | |
CN109840322A (en) | It is a kind of based on intensified learning cloze test type reading understand analysis model and method | |
CN110188209A (en) | Cross-module state Hash model building method, searching method and device based on level label | |
CN108629326A (en) | The action behavior recognition methods of objective body and device | |
CN106778876A (en) | User classification method and system based on mobile subscriber track similitude | |
CN109933656A (en) | Public sentiment polarity prediction technique, device, computer equipment and storage medium | |
CN103488789A (en) | Recommendation method, recommendation device and search engine | |
CN108804516A (en) | Similar users search device, method and computer readable storage medium | |
CN111695019B (en) | Method and device for identifying associated account | |
CN113792212B (en) | Multimedia resource recommendation method, device, equipment and storage medium | |
CN110210550A (en) | Image fine granularity recognition methods based on integrated study strategy | |
CN109871749A (en) | A kind of pedestrian based on depth Hash recognition methods and device, computer system again | |
CN111626357B (en) | Image identification method based on neural network model | |
CN108984514A (en) | Acquisition methods and device, storage medium, the processor of word | |
Ding et al. | A semantics-guided graph convolutional network for skeleton-based action recognition |
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 |