CN112100212A - Case scenario extraction method based on machine learning and rule matching - Google Patents
Case scenario extraction method based on machine learning and rule matching Download PDFInfo
- Publication number
- CN112100212A CN112100212A CN202010920756.3A CN202010920756A CN112100212A CN 112100212 A CN112100212 A CN 112100212A CN 202010920756 A CN202010920756 A CN 202010920756A CN 112100212 A CN112100212 A CN 112100212A
- Authority
- CN
- China
- Prior art keywords
- text
- matching
- training
- word
- machine learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 55
- 238000010801 machine learning Methods 0.000 title claims abstract description 22
- 239000013598 vector Substances 0.000 claims abstract description 33
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000013135 deep learning Methods 0.000 claims abstract description 26
- 230000014509 gene expression Effects 0.000 claims abstract description 21
- 230000011218 segmentation Effects 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 6
- 238000012545 processing Methods 0.000 claims abstract description 5
- 238000012549 training Methods 0.000 claims description 31
- 238000013527 convolutional neural network Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 8
- 238000011161 development Methods 0.000 claims description 8
- 230000018109 developmental process Effects 0.000 claims description 8
- 238000002474 experimental method Methods 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000004140 cleaning Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 abstract description 9
- 239000000284 extract Substances 0.000 abstract description 8
- 238000004458 analytical method Methods 0.000 abstract description 6
- 230000009471 action Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000005406 washing 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
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2455—Query execution
- G06F16/24564—Applying rules; Deductive queries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Technology Law (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a case scenario extraction method based on machine learning and rule matching, wherein the case scenario extraction method comprises the following steps: the keyword matching and regularization matching method comprises the following steps: extracting description sentences containing specified keywords or conforming to the regular expression in the paragraphs of the referee document as features; searching an episode corresponding to the characteristics in a pre-constructed episode library; the deep learning process comprises the following steps: performing word segmentation processing on the text to obtain a word sequence after word segmentation; vectorizing the word sequence after word segmentation to obtain a text vector of the text to be extracted; and inputting the text vector of the text to be extracted into a pre-constructed deep learning extraction model, and obtaining a result according to the output of the extraction model. The invention can extract not only explicit plots with strong interpretability, but also implicit plots with not so strong interpretability. And the analysis accuracy of case facts of low-frequency criminal names is improved by using different deep neural networks for different criminal names.
Description
Technical Field
The invention relates to the electronic technology of legal documents, in particular to a case scenario extraction method based on machine learning and rule matching.
Background
The legal scenario extraction task aims to automatically extract the most important scenarios from the case fact description part in the legal document. On the one hand, help people without legal basis understand important stories; on the other hand, legal references are provided for professional legal personnel. In recent years, China is continuously and deeply promoted to build an intelligent judicial law, and extraction of legal plots is an important link. The key word matching algorithm is utilized to research the plot extraction of cases as early as the last century. There have also been some studies related to legal plot extraction in recent years. With the rapid development of the deep learning technology, some scholars extract case scenarios in legal documents by using a deep neural network, and a good effect is achieved.
The chinese patent "109285094 legal document processing method and device" provides a method for extracting crime keywords in a target legal document and determining the crime episodes of case crime names in the target legal document in a pre-constructed crime database according to the crime keywords.
The Chinese patent '110032721 a referee document pushing method and device' provides a method and device for pushing referee documents by obtaining case plot characteristics through keyword matching and regular expression matching and searching similar characteristics.
Chinese patent '110263323 keyword extraction method and system based on a fence type long-time memory neural network' provides a keyword extraction method and system based on a neural network. : inputting legal text corpora of keywords to be extracted into a text coding model of a neural network to obtain a text semantic feature vector sequence; and inputting the text semantic feature vector sequence into a keyword recognition model to obtain a keyword extraction result.
The keyword matching and regular expression matching method can simply and effectively extract the obvious plot with high confidence coefficient, but the result is easy to make mistakes due to neglecting a plurality of semantic subtleties. But also has a low recall rate due to the expression of the regular expression without keywords and mismatch. On the basis of certain performance, even if a small promotion is required, enormous manpower is required to design a tighter regular expression. The deep learning method can learn some scenes which are difficult to match by regular expressions. But has the disadvantage that a large amount of labeled data is typically required for training. Furthermore, due to the problem of data imbalance, the deep learning method has poor analysis accuracy when dealing with cases of low-frequency guilt names. And the deep learning method lacks a certain interpretability.
The keyword matching and regular expression matching method is simple and efficient, has good interpretability, can extract explicit plots, and has the defects of low recall rate and labor consumption. The deep learning technology can extract some implicit plots through some implicit expressions, but has poor interpretability, needs a large amount of data, and has low analysis accuracy and coverage rate on case facts of some low-frequency crime names.
Disclosure of Invention
The invention aims to provide a case plot extraction method based on machine learning and rule matching, which is used for solving the problem that the deep learning method has poor analysis accuracy when processing cases of low-frequency criminal names by using different deep neural networks for different criminal names.
The invention relates to a case scenario extraction method based on machine learning and rule matching, which comprises the following steps: the keyword matching and regularization matching method comprises the following steps: extracting description sentences containing specified keywords or conforming to the regular expression in the paragraphs of the referee document as features; searching an episode corresponding to the characteristics in a pre-constructed episode library; the deep learning process comprises the following steps: performing word segmentation processing on the text to obtain a word sequence after word segmentation; vectorizing the word sequence after word segmentation to obtain a text vector of the text to be extracted; and inputting the text vector of the text to be extracted into a pre-constructed deep learning extraction model, and obtaining a result according to the output of the extraction model.
According to an embodiment of the case scenario extraction method based on machine learning and rule matching, the method further comprises the following steps: an episode base is constructed in advance.
According to an embodiment of the case scenario extraction method based on machine learning and rule matching, the method further comprises the following steps: a deep learning extraction model is constructed in advance: collecting official documents aiming at different crimes; cleaning data of the referee document, and dividing and extracting a part of case fact description according to keywords; manually calibrating the plot corresponding to the case fact; and (5) training the model.
According to an embodiment of the case scenario extraction method based on machine learning and rule matching, the pre-constructing scenario library comprises the following steps: (1) determining general plots and exclusive plots of various criminal names; (2) making regular expressions and matching rules for the determined universal plots and the special plots of the various criminal names; (3) and testing the various criminal names by using massive actual cases, and modifying the regular expression and the matching rule according to the test result.
According to an embodiment of the case scenario extraction method based on machine learning and rule matching, the method for constructing the deep learning extraction model further comprises the following steps: the method comprises the steps of dividing a referee document according to the names of the crimes, dividing the referee document with each name of the crimes into a training set, a testing set and a development set according to a certain proportion, wherein the training set is used for training a model, the development set is used for adjusting model parameters, and the testing set is used for finally evaluating the performance of the model.
According to an embodiment of the case scenario extraction method based on machine learning and rule matching, the training model comprises: an input layer, a hidden layer, and an output layer, wherein: an input layer: inputting a word vector two-dimensional matrix of a training text; order to
Representing a k-dimensional word vector corresponding to the ith word in a sentence, the sentence of length n is represented as:
wherein
Is a concatenation operator; hiding the layer: the system comprises a convolutional neural network, a word vector, a maximum pool and a vector matrix, wherein the convolutional neural network is used for abstracting a text input vector matrix to obtain deeper text information, extracting features for classification by using a binary classification task with plot extraction regarded as different criminals, performing convolution operation on the word vector by using convolution kernels with different sizes, and splicing the obtained features together to obtain final features through maximum pool pooling; an output layer: and (3) passing the obtained feature vector through one or more full-connection layers and activation function layers, and then passing through a Sigmoid activation function to obtain a text-based feature prediction classification result.
According to one embodiment of the case scenario extraction method based on machine learning and rule matching, the experiment is set, a jieba word segmentation component is used for Chinese word segmentation in the experiment, and a pre-training word vector of an Tencent AI laboratory is used; the experimental hidden layer adopts a convolutional neural network, the convolutional neural network uses convolutional kernels with the window sizes of 1, 2, 3 and 4, each convolutional kernel has 64, the output layer adopts a structure of two linear layers, and the characteristic size is SfThe size of the hidden layer is ShThe number of labels is SlFirst by Sf×ShOf (2) isLayer, then by Tanh function:
Tanh(x)＝(ex-e-x)/(ex+e-x)
by Sh×SlIn which the layer size S is hiddenh256 and linear layer dropout, probability 0.5; the learning rate for the training was set to 0.001, the training used Adam optimizer and BCELoss as loss function and Sigmoid activation function.
The case plot extraction method based on machine learning and rule matching can extract not only explicit plots with strong interpretability, but also some implicit plots with not so strong interpretability. And the analysis accuracy of case facts of low-frequency criminal names is improved by using different deep neural networks for different criminal names.
Drawings
FIG. 1 is a main flow chart of a case scenario extraction method based on machine learning and rule matching according to the present invention;
FIG. 2 is a diagram of the steps of pre-building a deep learning extraction model;
fig. 3 is a specific topology structure diagram of the deep learning extraction model.
Detailed Description
In order to make the objects, contents, and advantages of the present invention clearer, the following detailed description of the embodiments of the present invention will be made in conjunction with the accompanying drawings and examples.
Fig. 1 is a main flow chart of a case scenario extraction method based on machine learning and rule matching according to the present invention, and as shown in fig. 1, the method for extracting a scenario according to the present invention includes two modules after a text of a scenario to be extracted is input, one of which is a keyword matching and regularization matching flow on the left side of fig. 1 and a deep learning flow on the right side of fig. 1.
The keyword matching and regularization matching process comprises the following steps:
(1) in a paragraph of the referee document, only the keyword or the descriptive sentence conforming to the regular expression is included and extracted as the feature.
(2) And searching the plot corresponding to the characteristic in a pre-constructed plot library.
For example, the expression "(not | not) {0,4} antecedent" has a feature corresponding to the scenario "non-criminal antecedent".
The deep learning process comprises the following steps:
(1) and performing word segmentation processing on the text to obtain a word sequence after word segmentation.
(2) And vectorizing the word sequence after word segmentation to obtain a text vector of the text to be extracted.
(3) And inputting the text vector of the text to be extracted into a pre-constructed deep learning extraction model, and obtaining a result according to the output of the extraction model.
In order to make the technical field of the embodiment of the invention better understand, the embodiment of the invention is further described in detail below.
The method comprises the following steps of constructing an episode base in advance according to the criminal instruction opinions in various regions of the people's republic of China criminal law and the criminal action law of the people related to law, wherein the method comprises the following steps:
(1) the related people of law determine the universal plot and the exclusive plot of each criminal name according to the criminal instruction opinions in various regions of the criminal law of the people's republic of China and the criminal action law of the people's republic of China.
(2) Under the guidance of law-related people, regular expressions and matching rules are formulated for the determined general plots and the special plots of the various criminal names.
(3) And testing the various criminal names by using massive actual cases, and modifying the regular expressions and the matching rules under the guidance of persons related to law.
Fig. 2 is a diagram of steps for constructing a deep learning extraction model in advance, and the steps shown in fig. 2 are as follows:
(1) a large number of official documents for different names of crimes are collected. A total of 2426402 official documents were collected all over the country since 2000 years.
(2) And (4) cleaning data of the referee document, and dividing and extracting a part of the case fact description according to the keywords. 889690 parts of official documents remained after washing.
(3) The corresponding plot of the case fact is manually calibrated according to the criminal law of the people's republic of China and the criminal action of the people's republic of China in various regions of the sentencing guidance opinions.
(4) The referee documents are divided according to the names of the crimes, and a training set, a testing set and a development set are divided according to a certain proportion for the referee documents with the names of the crimes. The training set is used for training the model, the development set is used for adjusting the model parameters, and the test set is used for finally evaluating the model performance.
(5) And (5) training the model.
The specific topology of the deep learning extraction model is shown in fig. 3, and the model includes: input layer, hidden layer, output layer, wherein:
an input layer: the input is a two-dimensional matrix of word vectors of training text, for example, if the maximum length of the text is defined as 500 words, and the dimension of each word vector is set to 200, then the input should be a two-dimensional matrix of 500 × 200.
Order to
Representing a k-dimensional word vector corresponding to the ith word in a sentence, a sentence of length n is represented as
Wherein
Is the concatenation operator.
Hiding the layer: the method is mainly used for abstracting the text input vector matrix to obtain deeper text information. The conventional Convolutional Neural Network (CNN) or long-short term memory network (LSTM) can be used as the layer, and the Convolutional Neural Network (CNN) is taken as an example, the convolutional neural network is a deep feedforward artificial neural network and achieves remarkable results in the aspects of computer vision and speech recognition. Two classification tasks that treat episode extraction as different names of guilties use CNN to extract features for classification. And performing convolution operation on the word vectors by using convolution kernels with different sizes, performing maximum pooling, and splicing the obtained features together to obtain the final features.
In particular, for CNN, a convolution kernel
Generating features for h words, e.g. from the word xi:i+h-1Window generation feature ci：
ci＝f(w·xi:i+h-1+b)
Wherein
Is a bias term, f is a non-linear function like a linear rectification function (relu (x) max (0, x)), for a sentence { x1:h,x2:h+1,…xn-h+1:nGeneration of feature maps by convolution with a convolution kernel
c＝[c1,c2,…cn-h+1]
Wherein
Each feature map is considered to capture the most important features and may process sentences of variable length. Maximum pooling of features
An output layer: and passing the obtained feature vector through one or more full connection layers and an activation function layer. And activating a function through Sigmoid:
Sigmoid(x)＝σ(x)＝(1+e-x)-1
and obtaining a text feature prediction based classification result.
The linear layer input is c and the output is y, which can be expressed as y ═ cAT+ b. Where b is a deviation term.
The model is lightweight, does not occupy excessive time cost, and has good robustness.
(6) Experimental setup
The experiment uses the jieba word segmentation component for Chinese word segmentation and uses the 200-dimensional pre-training word vector of Tencent AI laboratories.
The experimental hidden layer used CNN, which used convolution kernels with window sizes of 1, 2, 3, and 4, with 64 convolution kernels per convolution kernel. The output layer adopts a structure of two linear layers, and the characteristic size is SfThe size of the hidden layer is ShThe number of labels is Sl. First through Sf×ShAnd then through the Tanh function:
Tanh(x)＝(ex-e-x)/(ex+e-x)
finally, pass through Sh×SlThe linear layer of (2). Wherein the hidden layer size Sh256. And the linear layer sets dropout with a probability of 0.5.
The training batch size is set to 100, the maximum epoch of the training is set to 50, the training is stopped using the early-stop mechanism, and the training is stopped when f1score is no longer elevated on the development set after 10 epochs, where f1 score:
f1score＝2*(precision*recall)/(precision+recall)
the learning rate for training was set to 0.001. Training uses an Adam optimizer and BCELoss as a loss function, which can be expressed as ln＝-wn[yn·logxn+(1-yn)·log(1-xn)]Wherein y isnIs true value label, xnIs output by the network. To ensure xnFor numbers between 0 and 1, Sigmoid activation function is used.
According to the invention, by integrating keyword matching and regular expression matching extraction scenarios and deep learning extraction scenarios, not only can explicit scenarios with strong interpretability be extracted, but also some implicit scenarios with not so strong interpretability can be extracted. The invention pre-constructs a deep learning extraction model and an episode base help system to extract episodes.
Compared with the prior art, the technical scheme provided by the invention integrates the advantages of the keyword matching and regular expression matching and the deep learning plot extraction method. Not only can explicit plots with strong interpretability be extracted, but also some implicit plots with not so strong interpretability can be extracted. Meanwhile, different deep neural networks are used for different criminal names, so that the problem of low analysis accuracy of low-frequency criminal names is solved. From the experimental results, the extraction effect is good after the method is adopted. The requirement of plot extraction is basically met. The invention is simple to realize, effectively extracts the plot and meets the application requirement.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (7)
1. A case scenario extraction method based on machine learning and rule matching is characterized by comprising the following steps:
the keyword matching and regularization matching method comprises the following steps:
extracting description sentences containing specified keywords or conforming to the regular expression in the paragraphs of the referee document as features;
searching an episode corresponding to the characteristics in a pre-constructed episode library;
the deep learning process comprises the following steps:
performing word segmentation processing on the text to obtain a word sequence after word segmentation;
vectorizing the word sequence after word segmentation to obtain a text vector of the text to be extracted;
and inputting the text vector of the text to be extracted into a pre-constructed deep learning extraction model, and obtaining a result according to the output of the extraction model.
2. The case scenario extraction method based on machine learning and rule matching as claimed in claim 1, further comprising: an episode base is constructed in advance.
3. The case scenario extraction method based on machine learning and rule matching as claimed in claim 1, further comprising: a deep learning extraction model is constructed in advance:
collecting official documents aiming at different crimes;
cleaning data of the referee document, and dividing and extracting a part of case fact description according to keywords;
manually calibrating the plot corresponding to the case fact;
and (5) training the model.
4. The case scenario extraction method based on machine learning and rule matching as claimed in claim 1, wherein pre-constructing a scenario library comprises:
(1) determining general plots and exclusive plots of various criminal names;
(2) making regular expressions and matching rules for the determined universal plots and the special plots of the various criminal names;
(3) and testing the various criminal names by using massive actual cases, and modifying the regular expression and the matching rule according to the test result.
5. The case scenario extraction method based on machine learning and rule matching as claimed in claim 3, wherein the building of deep learning extraction model further comprises:
the method comprises the steps of dividing a referee document according to the names of the crimes, dividing the referee document with each name of the crimes into a training set, a testing set and a development set according to a certain proportion, wherein the training set is used for training a model, the development set is used for adjusting model parameters, and the testing set is used for finally evaluating the performance of the model.
6. The case scenario extraction method based on machine learning and rule matching as claimed in claim 3, wherein training the model comprises: an input layer, a hidden layer, and an output layer, wherein:
an input layer: inputting a word vector two-dimensional matrix of a training text;
order to
The representation corresponds to a sentenceThe k-dimensional word vector of the ith word and the sentence with the length of n are expressed as
Wherein
Is a concatenation operator;
hiding the layer: the system comprises a convolutional neural network, a word vector, a maximum pool and a vector matrix, wherein the convolutional neural network is used for abstracting a text input vector matrix to obtain deeper text information, extracting features for classification by using a binary classification task with plot extraction regarded as different criminals, performing convolution operation on the word vector by using convolution kernels with different sizes, and splicing the obtained features together to obtain final features through maximum pool pooling;
an output layer: and (3) passing the obtained feature vector through one or more full-connection layers and activation function layers, and then passing through a Sigmoid activation function to obtain a text-based feature prediction classification result.
7. The case scenario extraction method based on machine learning and rule matching as claimed in claim 3, further comprising: setting an experiment, wherein the experiment uses a jieba word segmentation component to perform Chinese word segmentation and uses a pre-training word vector of an Tencent AI laboratory;
the experimental hidden layer adopts a convolutional neural network, the convolutional neural network uses convolutional kernels with the window sizes of 1, 2, 3 and 4, each convolutional kernel has 64, the output layer adopts a structure of two linear layers, and the characteristic size is SfThe size of the hidden layer is ShThe number of labels is SlFirst by Sf×ShAnd then through the Tanh function:
Tanh(x)＝(ex-e-x)/(ex+e-x)
by Sh×SlIn which the layer size S is hiddenh256 and linear layer dropout, probability 0.5;
the learning rate for the training was set to 0.001, the training used Adam optimizer and BCELoss as loss function and Sigmoid activation function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010920756.3A CN112100212A (en) | 2020-09-04 | 2020-09-04 | Case scenario extraction method based on machine learning and rule matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010920756.3A CN112100212A (en) | 2020-09-04 | 2020-09-04 | Case scenario extraction method based on machine learning and rule matching |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112100212A true CN112100212A (en) | 2020-12-18 |
Family
ID=73757730
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010920756.3A Pending CN112100212A (en) | 2020-09-04 | 2020-09-04 | Case scenario extraction method based on machine learning and rule matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112100212A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112784578A (en) * | 2021-03-16 | 2021-05-11 | 北京华宇元典信息服务有限公司 | Legal element extraction method and device and electronic equipment |
| CN113378563A (en) * | 2021-02-05 | 2021-09-10 | 中国司法大数据研究院有限公司 | Case feature extraction method and device based on genetic variation, semi-supervision and reinforcement learning |
| CN114611486A (en) * | 2022-03-09 | 2022-06-10 | 上海弘玑信息技术有限公司 | Information extraction engine generation method and device and electronic equipment |
| CN115687632A (en) * | 2022-08-25 | 2023-02-03 | 中国司法大数据研究院有限公司 | Criminal measuring plot decomposition analysis method and system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107894981A (en) * | 2017-12-13 | 2018-04-10 | 武汉烽火普天信息技术有限公司 | A kind of automatic abstracting method of case semantic feature |
| CN107918921A (en) * | 2017-11-21 | 2018-04-17 | 南京擎盾信息科技有限公司 | Criminal case court verdict measure and system |
| CN108009284A (en) * | 2017-12-22 | 2018-05-08 | 重庆邮电大学 | Using the Law Text sorting technique of semi-supervised convolutional neural networks |
| CN110032721A (en) * | 2018-01-11 | 2019-07-19 | 北京国双科技有限公司 | A kind of judgement document's method for pushing and device |
| CN110276068A (en) * | 2019-05-08 | 2019-09-24 | 清华大学 | Law merit analysis method and device |
| CN110991694A (en) * | 2019-10-30 | 2020-04-10 | 南京大学 | Sentencing prediction method based on deep learning |
| CN111104798A (en) * | 2018-10-27 | 2020-05-05 | 北京智慧正安科技有限公司 | Analysis method, system and computer readable storage medium for criminal plot in legal document |
| CN111198953A (en) * | 2018-11-16 | 2020-05-26 | 北京智慧正安科技有限公司 | Case text information based method and system for recommending cases and computer readable storage medium |
| CN111476027A (en) * | 2020-04-07 | 2020-07-31 | 南京森林警察学院 | Big data based anti-smuggling case information extraction method |
-
2020
- 2020-09-04 CN CN202010920756.3A patent/CN112100212A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107918921A (en) * | 2017-11-21 | 2018-04-17 | 南京擎盾信息科技有限公司 | Criminal case court verdict measure and system |
| CN107894981A (en) * | 2017-12-13 | 2018-04-10 | 武汉烽火普天信息技术有限公司 | A kind of automatic abstracting method of case semantic feature |
| CN108009284A (en) * | 2017-12-22 | 2018-05-08 | 重庆邮电大学 | Using the Law Text sorting technique of semi-supervised convolutional neural networks |
| CN110032721A (en) * | 2018-01-11 | 2019-07-19 | 北京国双科技有限公司 | A kind of judgement document's method for pushing and device |
| CN111104798A (en) * | 2018-10-27 | 2020-05-05 | 北京智慧正安科技有限公司 | Analysis method, system and computer readable storage medium for criminal plot in legal document |
| CN111198953A (en) * | 2018-11-16 | 2020-05-26 | 北京智慧正安科技有限公司 | Case text information based method and system for recommending cases and computer readable storage medium |
| CN110276068A (en) * | 2019-05-08 | 2019-09-24 | 清华大学 | Law merit analysis method and device |
| CN110991694A (en) * | 2019-10-30 | 2020-04-10 | 南京大学 | Sentencing prediction method based on deep learning |
| CN111476027A (en) * | 2020-04-07 | 2020-07-31 | 南京森林警察学院 | Big data based anti-smuggling case information extraction method |
Non-Patent Citations (2)
| Title |
|---|
| 林志宏 等: "基于卷积神经网络的公安案件文本语义特征提取方法研究" * |
| 陈慧炜: "公安领域案件文本信息抽取研究综述" * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113378563A (en) * | 2021-02-05 | 2021-09-10 | 中国司法大数据研究院有限公司 | Case feature extraction method and device based on genetic variation, semi-supervision and reinforcement learning |
| CN113378563B (en) * | 2021-02-05 | 2022-05-17 | 中国司法大数据研究院有限公司 | Case feature extraction method and device based on genetic variation and semi-supervision |
| CN112784578A (en) * | 2021-03-16 | 2021-05-11 | 北京华宇元典信息服务有限公司 | Legal element extraction method and device and electronic equipment |
| CN114611486A (en) * | 2022-03-09 | 2022-06-10 | 上海弘玑信息技术有限公司 | Information extraction engine generation method and device and electronic equipment |
| CN114611486B (en) * | 2022-03-09 | 2022-12-16 | 上海弘玑信息技术有限公司 | Method and device for generating information extraction engine and electronic equipment |
| CN115687632A (en) * | 2022-08-25 | 2023-02-03 | 中国司法大数据研究院有限公司 | Criminal measuring plot decomposition analysis method and system |
| CN115687632B (en) * | 2022-08-25 | 2024-04-09 | 中国司法大数据研究院有限公司 | Criminal investigation plot decomposition analysis method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113011533B (en) | Text classification method, apparatus, computer device and storage medium | |
| CN112231447B (en) | Method and system for extracting Chinese document events | |
| CN110362819B (en) | Text emotion analysis method based on convolutional neural network | |
| CN111414461B (en) | Intelligent question-answering method and system fusing knowledge base and user modeling | |
| CN109902175A (en) | A kind of file classification method and categorizing system based on neural network structure model | |
| CN108595708A (en) | A kind of exception information file classification method of knowledge based collection of illustrative plates | |
| CN110287323B (en) | Target-oriented emotion classification method | |
| CN112100212A (en) | Case scenario extraction method based on machine learning and rule matching | |
| CN111159485B (en) | Tail entity linking method, device, server and storage medium | |
| CN110598005A (en) | Public safety event-oriented multi-source heterogeneous data knowledge graph construction method | |
| CN111966812B (en) | Automatic question answering method based on dynamic word vector and storage medium | |
| CN103473327A (en) | Image retrieval method and image retrieval system | |
| CN113505200B (en) | Sentence-level Chinese event detection method combined with document key information | |
| CN110502742B (en) | Complex entity extraction method, device, medium and system | |
| CN111191442A (en) | Similar problem generation method, device, equipment and medium | |
| CN111581364B (en) | Chinese intelligent question-answer short text similarity calculation method oriented to medical field | |
| CN112256939A (en) | Text entity relation extraction method for chemical field | |
| CN113220844A (en) | Remote supervision relation extraction method based on entity characteristics | |
| CN114020906A (en) | Chinese medical text information matching method and system based on twin neural network | |
| CN112988970A (en) | Text matching algorithm serving intelligent question-answering system | |
| CN114417851A (en) | Emotion analysis method based on keyword weighted information | |
| CN110659392B (en) | Retrieval method and device, and storage medium | |
| CN112329441A (en) | Legal document reading model and construction method | |
| CN114881043A (en) | Deep learning model-based legal document semantic similarity evaluation method and system | |
| CN114925702A (en) | Text similarity recognition method and device, electronic equipment and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201218 |