CN109241528A - A kind of measurement of penalty prediction of result method, apparatus, equipment and storage medium - Google Patents
A kind of measurement of penalty prediction of result method, apparatus, equipment and storage medium Download PDFInfo
- Publication number
- CN109241528A CN109241528A CN201810971990.1A CN201810971990A CN109241528A CN 109241528 A CN109241528 A CN 109241528A CN 201810971990 A CN201810971990 A CN 201810971990A CN 109241528 A CN109241528 A CN 109241528A
- Authority
- CN
- China
- Prior art keywords
- measurement
- penalty
- court verdict
- label
- specified
- 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
- 238000005259 measurement Methods 0.000 title claims abstract description 375
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 63
- 239000000284 extract Substances 0.000 claims abstract description 18
- 239000013598 vector Substances 0.000 claims description 68
- 238000000605 extraction Methods 0.000 claims description 65
- 230000008569 process Effects 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 10
- 235000013399 edible fruits Nutrition 0.000 claims description 9
- 238000011282 treatment Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 14
- 238000004891 communication Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 4
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000008878 coupling Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 241000132179 Eurotium medium Species 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biophysics (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Molecular Biology (AREA)
- Tourism & Hospitality (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Technology Law (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application provides a kind of measurement of penalty prediction of result method, apparatus, equipment and storage medium, method includes: the non-court verdict for the specified case for obtaining specified charge as the non-court verdict of target;The specified corresponding measurement of penalty element of measurement of penalty element label is obtained from the non-court verdict of target as target measurement of penalty element;Target measurement of penalty element is inputted to the measurement of penalty prediction of result model pre-established, obtains the measurement of penalty result of the specified case of measurement of penalty prediction of result model output;Measurement of penalty prediction of result model is trained to obtain using measurement of penalty element extracting from the training court verdict of specified charge, corresponding with specified measurement of penalty element label as training sample using court verdict element extract from training court verdict, corresponding with specified court verdict element label as sample label.Measurement of penalty prediction of result method, apparatus, equipment and storage medium provided by the present application can automatic Prediction go out more accurately the measurement of penalty as a result, the measurement of penalty result for judge reference, to assist judge to make decisions.
Description
Technical field
This application involves field of artificial intelligence more particularly to a kind of measurement of penalty prediction of result method, apparatus, equipment and deposit
Storage media.
Background technique
In recent years, with the rapid development of big data and artificial intelligence technology, carry out indirect labor using machine and have become
The hot spot direction of all trades and professions.
Judicial information construction is the important directions that Chinese justice realizes modernization, and judicial relevant work is also by traditional
One collimation method official is directed to different merits, inspection information and juristic writing, and it is complete under the auxiliary of machine gradually to develop into a collimation method official
At affairs such as such as court's trial record, merit analyses, thus there is the construction system of " wisdom law court ".
The overall architecture of wisdom law court includes that artificial intelligence technology and big data technology are used for the service society public, service
Four case trial, service enforcement of the judgment, service judicial administration aspects.However, this application scenarios is tried for service case,
There presently does not exist the schemes for capableing of the automatic measurement of penalty.
Summary of the invention
In view of this, this application provides a kind of measurement of penalty prediction of result method, apparatus, equipment and storage mediums, to base
Go out measurement of penalty result in case correlation document automatic Prediction for judge's reference, its technical solution is as follows:
A kind of measurement of penalty prediction of result method, comprising:
The non-court verdict for obtaining the specified case of specified charge, as the non-court verdict of target;
Measurement of penalty element corresponding with specified measurement of penalty element label is obtained from the non-court verdict of the target, as the target measurement of penalty
Element;
The target measurement of penalty element is inputted to the measurement of penalty prediction of result model pre-established, obtains the measurement of penalty prediction of result
The measurement of penalty result of the specified case of model output;
Wherein, the measurement of penalty prediction of result model with it is being extracted from the training court verdict of the specified charge, with it is described
The specified corresponding measurement of penalty element of measurement of penalty element label is training sample, with it is being extracted from the trained court verdict, sentence with specifying
Certainly the corresponding court verdict element of result element label is that sample label is trained to obtain.
Wherein, the extraction from the training court verdict of the specified charge is corresponding with the specified measurement of penalty element label
The process of measurement of penalty element and court verdict element corresponding with the specified court verdict element label includes:
Using the court verdict element extraction model pre-established, never extracts in mark court verdict and wanted with the specified measurement of penalty
The corresponding measurement of penalty element of plain label and court verdict element corresponding with the specified court verdict element label;
Wherein, the court verdict element extraction model is to be labeled with specified measurement of penalty element label and specified court verdict element
The training court verdict of label is trained to obtain.
Wherein, the court verdict element extraction model that the utilization pre-establishes, from it is described do not mark in court verdict extract with
The corresponding measurement of penalty element of the specified measurement of penalty element label and judgement corresponding with the specified court verdict element label are tied
Fruit element, comprising:
By the semantic vector determining module in the court verdict element extraction model, the court verdict that do not mark is carried out
Word segmentation processing determines the corresponding semantic vector of each word that word segmentation processing obtains;
Pass through the element label determining module and the corresponding semanteme of each word in the court verdict element extraction model
Vector determines the element label of word corresponding with each semantic vector;
The identical continuous multiple words of element label are merged, the content after merging is as the corresponding element of element label.
Wherein, described that measurement of penalty element corresponding with specified measurement of penalty element label, packet are obtained from the non-court verdict of the target
It includes:
Using the non-court verdict element extraction model pre-established, extracts from the non-court verdict of the target and specified with described
The corresponding measurement of penalty element of measurement of penalty element label;
Wherein, the non-court verdict element extraction model is using the non-court verdict of training as training sample, based on described specified
Measurement of penalty element label is that sample label is trained to obtain to the annotation results that the non-court verdict of the training is labeled.
Wherein, the annotation results packet that the non-court verdict is labeled based on the specified measurement of penalty element label
It includes:
In the non-court verdict of training the initial position of measurement of penalty element corresponding with the specified amount type element label and rise
Beginning position score, end position and end position score, and, measurement of penalty element is empty score.
Wherein, the non-court verdict element extraction model that the utilization pre-establishes, is extracted from the non-court verdict of the target
Measurement of penalty element corresponding with the specified measurement of penalty element label, comprising:
By the non-court verdict element extraction model determine in the non-court verdict of the target with the specified measurement of penalty element
The information of the corresponding measurement of penalty element of label includes the target as target measurement of penalty element information, the target measurement of penalty element information
The initial position of measurement of penalty element corresponding with the specified measurement of penalty element label and initial position score, stop bits in non-court verdict
Setting with end position score and measurement of penalty element is empty score;
By the target measurement of penalty element information, determine in the non-court verdict of the target with the specified measurement of penalty element label
Corresponding measurement of penalty element.
Wherein, described determined in the non-court verdict of the target by the non-court verdict element extraction model is specified with described
The information of the corresponding measurement of penalty element of measurement of penalty element label is as target measurement of penalty element information, comprising:
By the non-court verdict of the target and problem label text, the non-court verdict element extraction model is inputted, obtains institute
The target answer information for stating non-court verdict element extraction model output, as the target measurement of penalty element information;
Wherein, problem label text is the text comprising given problem label, and the given problem label is by the finger
Quantitative punishment element label is converted to the label after problem form;The target answer information includes and the given problem label pair
The initial position and initial position score of the answer answered, end position and end position score, and, answer is empty score.
Wherein, described by the target measurement of penalty element information, determine in the non-court verdict of the target with the specified amount
The corresponding measurement of penalty element of punishment element label, comprising:
If there is the target problem label with multiple answers in the given problem label, from the target problem mark
The answer that removal in corresponding multiple answers is unsatisfactory for preset condition is signed, remaining answer is obtained;
If the residue answer be it is multiple, will multiple remaining answers progress duplicate removal merging treatments, at duplicate removal merging
The answer obtained after reason is as the target measurement of penalty element.
Wherein, the removal from the target problem label corresponding multiple answers is unsatisfactory for the answer of preset condition,
Obtain remaining answer, comprising:
Determine that default answer forms the first candidate answers set from the corresponding multiple answers of the target problem label,
Wherein, the score of the default answer is above the score of other answers, and the score of each answer passes through the starting of the answer
Position score and end position score determine;
Candidate answers by score in the first candidate answers set lower than first threshold remove, and obtain the second candidate and answer
Case set, wherein the first threshold is that the corresponding answer of the target problem label is empty score;
Candidate answers removal by score in the second candidate answers set lower than the non-top score of second threshold, is obtained
Obtain the remaining answer, wherein the score of candidate answers of the second threshold based on the top score is set.
It is wherein, described that the answer for being unsatisfactory for preset condition is filtered out from the corresponding multiple answers of the target problem label,
Obtain remaining answer, further includes:
If the score of each candidate answers is below the first threshold in the first candidate answers set, it is determined that institute
The corresponding answer of target problem label is stated as sky.
A kind of measurement of penalty prediction of result device, comprising: non-court verdict obtains module, measurement of penalty element determining module and measurement of penalty result
Prediction module;
The non-court verdict obtains module, and the non-court verdict of the specified case for obtaining specified charge is non-as target
Court verdict;
The measurement of penalty element determining module, for being obtained and specified measurement of penalty element label pair from the non-court verdict of the target
The measurement of penalty element answered, as target measurement of penalty element;
The measurement of penalty prediction of result module, for the target measurement of penalty element to be inputted the measurement of penalty prediction of result pre-established
Model obtains the measurement of penalty result of the specified case of the measurement of penalty prediction of result model output;
Wherein, the measurement of penalty prediction of result model with it is being extracted from the training court verdict of the specified charge, with it is described
The specified corresponding measurement of penalty element of measurement of penalty element label is training sample, with it is being extracted from the trained court verdict, sentence with specifying
Certainly the corresponding court verdict element of result element label is that sample label is trained to obtain.
A kind of measurement of penalty prediction of result equipment, comprising: memory and processor;
The memory, for storing program;
The processor, for executing described program, described program is specifically used for:
The non-court verdict for obtaining the specified case of specified charge, as the non-court verdict of target;
Measurement of penalty element corresponding with specified measurement of penalty element label is obtained from the non-court verdict of the target, as the target measurement of penalty
Element;
The target measurement of penalty element is inputted to the measurement of penalty prediction of result model pre-established, obtains the measurement of penalty prediction of result
The measurement of penalty result of the specified case of model output;
Wherein, the measurement of penalty prediction of result model with it is being extracted from the training court verdict of the specified charge, with it is described
The specified corresponding measurement of penalty element of measurement of penalty element label is training sample, with it is being extracted from the trained court verdict, sentence with specifying
Certainly the corresponding court verdict element of result element label is that sample label is trained to obtain.
A kind of readable storage medium storing program for executing, is stored thereon with computer program, which is characterized in that the computer program is processed
When device executes, each step of above-mentioned measurement of penalty prediction of result method is realized.
Via above scheme it is found that measurement of penalty prediction of result method, apparatus, equipment and storage medium provided by the present application, first
The non-court verdict of target for first obtaining the specified case of specified charge, then obtains and specified measurement of penalty element from the non-court verdict of target
Target measurement of penalty element is finally inputted the measurement of penalty prediction of result model pre-established by the corresponding target measurement of penalty element of label, is obtained
The measurement of penalty result of measurement of penalty prediction of result model output.It can be seen that measurement of penalty prediction of result method, apparatus provided by the present application, setting
Standby and storage medium can utilize the measurement of penalty prediction of result model automatic Prediction pre-established based on the non-court verdict of specified case
Out specify case the measurement of penalty as a result, the measurement of penalty result for judge reference, to assist judge to make decisions specified case, thus
It can be improved the trial efficiency of case.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of invention for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is the flow diagram of measurement of penalty prediction of result method provided by the embodiments of the present application;
The court verdict element extraction model that Fig. 2 utilization provided by the embodiments of the present application pre-establishes, never mark court verdict
Middle extraction and the corresponding measurement of penalty element of specified measurement of penalty element label and judgement corresponding with specified court verdict element label are tied
The flow diagram of the realization process of fruit element;
Fig. 3 is an exemplary schematic diagram of the topological structure of court verdict element extraction model provided by the embodiments of the present application;
Fig. 4 be it is provided by the embodiments of the present application is extracted from target non-court verdict using non-court verdict element extraction model and
The flow diagram of the realization process of the specified corresponding measurement of penalty element of measurement of penalty element label;
Fig. 5 is an exemplary signal of the topological structure of non-court verdict element extraction model provided by the embodiments of the present application
Figure;
Fig. 6 is that the removal provided by the embodiments of the present application from the corresponding multiple answers of target problem label is unsatisfactory for default item
The answer of part obtains the flow diagram of the realization process of remaining answer;
Fig. 7 is the structural schematic diagram of measurement of penalty prediction of result device provided by the embodiments of the present application;
Fig. 8 is the structural schematic diagram of measurement of penalty prediction of result equipment provided by the embodiments of the present application.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Inventor has found during realizing the invention: some measurement of penalty prediction of result sides exist in the prior art
Case, these measurement of penalty prediction of result schemes are most are as follows: using sides such as regular expression, context-free grammar, name Entity recognitions
Formula extracts the plot information of legal documents, carries out structuring to the information of extraction later, the spy of data is described with mathematical model
Point and rule calculate the mathematical model being most consistent with it or algorithm, and artificial intelligence is by simulation algorithm, according to plot derivation amount
Punishment result.However, existing measurement of penalty prediction of result scheme is substantially based on traditional machine learning method, need to design more
Regular (regular expression), feature (context-free question and answer, name Entity recognition) etc., need a large amount of manpower interventions, and due to
The description of law correlation document is complicated, and case is difficult for this scheme to be generalized to more case by auxiliary measurement of penalty effect compares by numerous
Difference.
In order to overcome the problems, such as that existing scheme exists, the embodiment of the present application provides a kind of measurement of penalty prediction of result method, asks
Refering to fig. 1, the flow diagram of the measurement of penalty prediction of result method is shown, may include:
Step S101: the non-court verdict of the specified case of specified charge is obtained, as the non-court verdict of target.
Wherein, specified charge can be any one of the charge of legal provisions, such as robbery crime, larceny, crime of fraud, move
With fund crime, crime of forcible seizure, offence of extortion by blackmail etc., i.e., the measurement of penalty prediction of result method scope of application provided by the present application is wider.
Wherein, non-court verdict can with but be not limited to interrogation record, court's trial notes, indictment etc..
It should be noted that it may also be more parts that the non-court verdict of specified case, which may be portion, for example, may be directed to
Different objects does interrogation record, can so obtain more parts of interrogation records.
Step S102: measurement of penalty element corresponding with specified measurement of penalty element label is obtained from the non-court verdict of target, as mesh
Scalar punishment element.
It should be noted that specified measurement of penalty element label is set based on specified charge, measurement of penalty element label pair is specified
The measurement of penalty element answered refers to content corresponding with specified measurement of penalty element label in the non-court verdict of target.
Target measurement of penalty element: being inputted the measurement of penalty prediction of result model pre-established by step S103, and it is pre- to obtain measurement of penalty result
Survey the measurement of penalty result of the specified case of model output.
Wherein, measurement of penalty prediction of result model is with extract from the training court verdict of specified charge and specified measurement of penalty element
The corresponding measurement of penalty element of label is training sample, with the extract and specified court verdict element label pair from training court verdict
The court verdict element answered is that sample label is trained to obtain.
It should be noted that the format write of court verdict generally standardizes very much, and in court verdict information specifically include that by
People's essential information, measurement of penalty related episodes and the measurement of penalty are accused as a result, therefore, it can be based on the content setting amount of the court verdict of specified charge
Punishment element label and court verdict element label.
By taking specified charge is larceny as an example, can set measurement of penalty element label include: whether burglary, whether steal,
Whether have previous conviction, whether disabled person, the theft amount of money, plead guilty show repentance attitude it is whether good, it is whether honest, whether confess one's crime, can set
Surely specified court verdict element label includes principal penalty, accessary penalty etc..
Furthermore, it is contemplated that target measurement of penalty element be discrete data, measurement of penalty prediction of result model can with but be not limited to SVM mould
Type, Logic Regression Models, deep neural network model etc..
Measurement of penalty prediction of result method provided by the embodiments of the present application, the target for obtaining the specified case of specified charge first are non-
Then court verdict obtains target measurement of penalty element corresponding with specified measurement of penalty element label, finally by mesh from the non-court verdict of target
Scalar punishment element inputs the measurement of penalty prediction of result model pre-established, obtains the specified case of measurement of penalty prediction of result model output
Measurement of penalty result.Measurement of penalty prediction of result method provided by the embodiments of the present application can be based on the non-court verdict of specified case, using preparatory
The measurement of penalty prediction of result model automatic Prediction of foundation go out the measurement of penalty as a result, and the measurement of penalty result that predicts it is more accurate, the measurement of penalty knot
Fruit has preferable auxiliary measurement of penalty effect for judge's reference.
In measurement of penalty prediction of result method provided by the above embodiment, for predicting the measurement of penalty prediction of result mould of measurement of penalty result
The training sample of type be extracted from the training court verdict of specified charge, corresponding with the specified measurement of penalty element label measurement of penalty wants
Element, sample label are court verdict element extract from training court verdict, corresponding with specified court verdict element label.With
Under to extracted from training court verdict corresponding with specified measurement of penalty element label measurement of penalty element and with specified court verdict element
The realization process of the corresponding court verdict element of label is illustrated.
In one possible implementation, using the court verdict element extraction model pre-established, never mark is sentenced
It certainly extracts and the corresponding measurement of penalty element of specified measurement of penalty element label and corresponding with specified court verdict element label sentences in book
Certainly result element.Wherein, court verdict element extraction model is to be labeled with specified measurement of penalty element label and specified court verdict element
The training court verdict of label is trained to obtain.
It should be noted that since the format write of court verdict compares specification, mould is extracted in training court verdict element
When type, it is only necessary to mark a small amount of court verdict, for example 500~1000 parts of court verdicts can be marked, be sentenced with what these had been marked
Certainly book training court verdict element extraction model.That is, first with being labeled with specified measurement of penalty element label on a small quantity and specified court verdict is wanted
The court verdict training court verdict element extraction model of plain label, is then never marked with the court verdict element extraction model that training obtains
It infuses and extracts corresponding with specified measurement of penalty element label measurement of penalty element and corresponding with specified court verdict element label in court verdict
Court verdict element, the data of extraction are for training measurement of penalty prediction of result model.
In one possible implementation, training court verdict can be labeled based on BIOES label system,
In BIOES label system, indicate that label originates word with B, label starting word here refers to the starting of the corresponding element of label
Word indicates label medium term, the i.e. medium term of the corresponding element of label with I, indicates label closing with E, i.e. label is corresponding
The closing of element indicates single label word, i.e. only one word of the corresponding element of label with S, and being indicated with O have to plain word.Show
Example property, for " whether confessing ", relevant label have B_ whether honest, I_ whether honest, E_ whether honest, S_ whether
It is honest.
Further, Fig. 2 and Fig. 3 are please referred to, Fig. 2 shows using the court verdict element extraction model pre-established, from
Do not mark extracted in court verdict corresponding with specified measurement of penalty element label measurement of penalty element and with specified court verdict element label
The flow diagram of the realization process of corresponding court verdict element, Fig. 3 show the topology knot of court verdict element extraction model
The exemplary schematic diagram of the one of structure, the process that element is never extracted in mark court verdict may include:
Step S201: by the semantic vector determining module (301 in Fig. 3) in court verdict element extraction model, to not
It marks court verdict and carries out word segmentation processing, determine the corresponding semantic vector of each word that word segmentation processing obtains.
In one possible implementation, determine that the process for the corresponding semantic vector of each word that word segmentation processing obtains can
To include: that each word is executed: firstly, the name entity of the part of speech of identification word and word;Then, respectively by word, word part of speech,
The name entity of word is converted to vector, for example, can by wordembedding model by word, the part of speech of word, word name entity
Be converted to vector, obtain the corresponding vector of word, word the corresponding vector sum word of part of speech the corresponding vector of name entity;Finally,
The corresponding vector of name entity of the corresponding vector of word, the corresponding vector of part of speech of word, word is spliced (by Fig. 3
Three vectors are stitched together by vector splicing module concat), semantic vector of the vector obtained after splicing as word.
In order to keep the semanteme of the corresponding semantic vector of each word richer, in alternatively possible implementation, for
For each word, in addition to needing to obtain the corresponding vector of part of speech of the corresponding vector of word, word, the corresponding vector of name entity of word
Outside, it is also necessary to each word in word is converted into vector, the corresponding vector of each word is spliced, feature extraction is then passed through
Module (such as convolutional neural networks CNN) extracts feature vector from spliced vector, finally by the corresponding vector of word, the word of word
The corresponding vector of property, the corresponding vector of name entity of word and the feature vector of said extracted are spliced (by Fig. 3
Four vectors are stitched together by vector splicing module concat), semantic vector of the vector obtained after splicing as word.
Step S202: by the element label determining module (302 in Fig. 3) in court verdict element extraction model and each
The corresponding semantic vector of word determines the element label of word corresponding with each semantic vector.
Specifically, the corresponding semantic vector of each word is inputted into element label determining module, obtains element label and determines mould
The element label of block output, corresponding with the semantic vector of input word, in this way, just not marked each word in court verdict
Element label.In addition, element label determining module in court verdict element extraction model can with but be not limited to two-way
BiLSTM (i.e. two-way LSTM).
Step S203: the identical continuous multiple words of element label are merged, the content after merging is as the element label pair
The element answered.
Illustratively, it is assumed that the corresponding element label of word a is " whether B_ confesses ", the corresponding element label of word b is that " I_ is
It is no honest ", the corresponding element label of word c be " whether I_ confesses ", the corresponding element label of word d is " whether E_ confesses ", due to
Whether word a, word b, word c and the corresponding element label of word d are " confessing ", therefore, word a, word b, word c and word d are merged into one
It rises and is used as element label " whether confessing " corresponding element.
In another embodiment of the application, obtained and the specified measurement of penalty in the non-court verdict of slave target in above-described embodiment
The corresponding measurement of penalty element of element label is introduced.
In one possible implementation, the specified corresponding measurement of penalty of measurement of penalty element label is obtained from the non-court verdict of target
The process of element may include: using the non-court verdict element extraction model that pre-establishes, extracted from the non-court verdict of target with
The specified corresponding measurement of penalty element of measurement of penalty element label.
Wherein, non-court verdict element extraction model is using the non-court verdict of training as training sample, based on specified measurement of penalty element
Label is that sample label is trained to obtain to the annotation results that the non-court verdict of training is labeled.
In one possible implementation, the mark non-court verdict of training being labeled based on specified measurement of penalty element label
Infusing result may include: the initial position of measurement of penalty element corresponding with specified amount type element label and starting in trained non-court verdict
Position score, end position and end position score.
In view of some or certain specified measurement of penalty element labels may not occur in the non-court verdict of training, correspondingly,
Also there would not be measurement of penalty element corresponding with some or certain specified measurement of penalty element labels in non-court verdict, this is based on, another
In a kind of possible implementation, the annotation results that are labeled based on specified measurement of penalty element label to the non-court verdict of training can be with
Include: the initial position of measurement of penalty element corresponding with specified amount type element label and initial position score in trained non-court verdict,
End position and end position score, and, measurement of penalty element is empty score.
Referring to Fig. 4, show using non-court verdict mark based on above-mentioned second of notation methods it is trained obtain it is non-
Court verdict element extraction model extracts the realization of the specified corresponding measurement of penalty element of measurement of penalty element label from the non-court verdict of target
The flow diagram of journey may include:
Step S401: by non-court verdict element extraction model, determine in the non-court verdict of target with specified measurement of penalty element mark
The information for signing corresponding measurement of penalty element, as target measurement of penalty element information.
Wherein, target measurement of penalty element information include in the non-court verdict of target the measurement of penalty corresponding with specified measurement of penalty element label want
The initial position of element and initial position score, end position and end position score and measurement of penalty element are empty score.
In one possible implementation, non-court verdict element extraction model can understand model using reading, the mould
The input of type includes non-court verdict and specified measurement of penalty element label, in a kind of preferred implementation, reads reason to be promoted
The effect for solving model, can be converted to problem form for specified measurement of penalty element label based on the semanteme of specified measurement of penalty element label, than
Such as, measurement of penalty element label is specified are as follows: " whether pleading guilty " can then be converted are as follows: " behavior for you, you are to deny
Crime ", after converting to specified measurement of penalty element label, obtains given problem label.It should be noted that in a kind of possible reality
In existing mode, for each specified element label, it can be converted into a problem, use the problem to read as label training and manage
Solve model;In order to further enhance the effect for understanding model is read, in alternatively possible implementation, for each specified
Element label can be converted into multiple problems based on its semanteme, be read with the training of multiple problems and understand model.
Then by non-court verdict element extraction model, determine corresponding with specified measurement of penalty element label in the non-court verdict of target
The information of measurement of penalty element, the process as target measurement of penalty element information may include: by the non-court verdict of target and problem label text
The non-court verdict element extraction model (reading understands model) of this input, the target for obtaining non-court verdict element extraction model output are answered
Case information, as target measurement of penalty element information.Wherein, problem label text is the text comprising given problem label, and target is answered
Case information includes initial position and the initial position score, end position and end position of answer corresponding with given problem label
Score and answer are empty score.
Referring to Fig. 5, showing an exemplary schematic diagram of the topological structure of non-court verdict element extraction model, Fig. 5 shows
The input of non-court verdict element extraction model out is non-court verdict and problem label text, Concat 502a in Fig. 5 and
Concat 502b is for splicing the vector of input, and specifically, the input of Concat 502a is carried out to non-court verdict
The vector for the corresponding vector of each word and CNN the 501a output that word segmentation processing obtains, the input of CNN501a are equivalent
The spliced vector of each corresponding vector progress of word, the vector of Concat 502a output is as each word pair in non-court verdict
The object vector answered.Similarly, the input of Concat 502b is to carry out each word that word segmentation processing obtains to question text to correspond to
Vector and CNN 501b output vector, the input of CNN 501b is that the corresponding vector of each word of equivalent is spliced
Vector afterwards, the vector of Concat 502b output is as the corresponding object vector of word each in question text.It will be in non-court verdict
The corresponding object vector of each word inputs BiLSTM (two-way LSTM) 503a, obtains in non-court verdict that each word is corresponding, has
The corresponding object vector of word each in question text is similarly inputted BiLSTM503b, is asked by the vector of contextual information
It inscribes in text that each word is corresponding, vector with contextual information, there is context by word each in non-court verdict is corresponding
The corresponding vector with contextual information of each word inputs Attention 504 in the vector sum question text of information,
The vector of Attention504 output is that each base is exported in the expression of question text, Attention504 in non-court verdict
Vector inputs BiLSTM505, and BiLSTM505 exports each word in non-court verdict to be terminated as the score of answer starting word, answer
The score of word and the word are not belonging to the score of answer.
Step S402: by target measurement of penalty element information, determine in the non-court verdict of target with specified measurement of penalty element label pair
The measurement of penalty element answered.
It should be noted that sometimes, in fact it could happen that some given problem label has the case where multiple answers, when
Occur such case when, can multiple answers corresponding to given problem label handle.Specifically, by with multiple answers
Given problem label is as target problem label, and removal is unsatisfactory for preset condition from the corresponding multiple answers of target problem label
Answer, obtain remaining answer;If remaining answer be it is multiple, multiple remaining answers are subjected to duplicate removal merging treatments, go to be overlapped
And the answer obtained after handling is as measurement of penalty element corresponding to specified measurement of penalty element label corresponding with target problem label.
Further, it is unsatisfactory for presetting referring to Fig. 6, showing the removal from the corresponding multiple answers of target problem label
The answer of condition, obtains the flow diagram of the realization process of remaining answer, which may include:
Step S601: determine that default answer forms the first candidate answers from the corresponding multiple answers of target problem label
Set.
Wherein, the score of each answer is determined by the initial position score and end position score of the answer.Specifically,
It, can be using the product of the corresponding initial position score of the answer and end position score as the answer for each answer
Score.
Wherein, the score for presetting an answer is above the score of other answers.It in one possible implementation, can be first
The answer of highest scoring is filtered out, the initial position score of the answer filtered out and end position then must be split 0, then
The initial position score of the answer filtered out and end position must be split 0, with such by the answer for screening highest scoring again
It pushes away, filters out default answer.It, can be by the sequence of score from high to low to target problem in alternatively possible implementation
The corresponding multiple answers of label are ranked up, it is assumed that default is N (such as 5), then the answer composition first of N before ranking is candidate
Answer set.
It should be noted that if the quantity of the corresponding answer of target problem label is greater than default, S601 is thened follow the steps,
If the quantity of the corresponding answer of target problem label is less than or equal to default, by direct group of the corresponding answer of target problem label
At the first candidate answers set, step S402 is then executed.
Step S602: the candidate answers by score in the first candidate answers set lower than first threshold remove, and obtain second
Candidate answers set.
Wherein, it is empty score that first threshold, which is the corresponding answer of target problem label,.
It illustratively, include 5 answers in the first candidate collection, wherein be below target there are two the score of answer and ask
Inscribing the corresponding answer of label is empty score, then removes the two answers from the first candidate answers set.
It should be noted that if the score of each candidate answers is below target problem label in the first candidate answers set
Corresponding answer is empty score, then it is assumed that answer not corresponding with the target problem label in the non-court verdict of target.
Step S603: the candidate answers by score in the second candidate answers set lower than the non-top score of second threshold are gone
It removes, obtains remaining answer.
Wherein, the score setting of candidate answers of the second threshold based on top score.Illustratively, second threshold can be set
It is the 1/10 of the score of the candidate answers of top score, i.e., if the score of some candidate answers in the second candidate answers set
Lower than the 1/10 of the score of the candidate answers of top score, then the candidate answers are removed from the second candidate answers set.
After obtaining remaining answer, if remaining answer have it is multiple, it is understood that there may be multiple residue answers, which exist, is overlapped content
Situation, at this time, it may be necessary to multiple remaining answers are carried out duplicate removal merging treatments, the answer obtained after duplicate removal merging treatment as with mesh
The corresponding target answer of mark problem label, the measurement of penalty corresponding to specified measurement of penalty element label as corresponding with target problem label
Element.
Illustratively, the corresponding answer of target problem label has multiple, from the corresponding multiple answers of target problem label
Removal is unsatisfactory for the answer of preset condition, and the remaining answer of acquisition includes 3, and 3 answers are respectively as follows: (1) younger brother: XXX, and 37
It in year, is engaged in agriculture at ancestral home, younger sister: XXX, 35 years old;(2) younger sister: XXX, 35 years old, present ancestral home trade;(3) son: XXX, 12 years old, from
Former wife is returned to bring up after marriage, due to being overlapped content between first answer and second answer: younger sister: XXX 35 years old, then needs
The coincidence content is removed, removal obtains after being overlapped content: (1) younger brother: XXX 37 years old, is engaged in agriculture at ancestral home;(2) younger sister: XXX,
35 years old, present ancestral home trade;(3) son: XXX 12 years old, returns former wife to bring up after divorce.
Measurement of penalty prediction of result method provided by the embodiments of the present application, using be labeled on a small quantity specified measurement of penalty element label and
The court verdict training court verdict element extraction model of specified court verdict element label, the court verdict element then obtained with training
Extraction model, which never marks, to be extracted measurement of penalty element corresponding with specified measurement of penalty element label and ties with specified judgement in court verdict
The corresponding court verdict element of fruit element label, then never marks the number extracted in court verdict with court verdict element extraction model
According to training measurement of penalty prediction of result model, finally extracted and specified measurement of penalty element label pair from the non-court verdict of target of specified case
The target measurement of penalty element answered, the measurement of penalty prediction of result model that the input training of target measurement of penalty element is obtained, to obtain measurement of penalty knot
The measurement of penalty result of the output of fruit prediction model, specified case.Measurement of penalty prediction of result method provided by the present application is based on legal documents
Internal characteristics and the legal documents that manually mark just can reach preferable auxiliary measurement of penalty effect on a small quantity, overcome existing prediction side
The problem that case manpower intervention is excessive and auxiliary measurement of penalty effect is poor.
Corresponding with above-mentioned measurement of penalty prediction of result method, the embodiment of the present application also provides a kind of measurement of penalty prediction of result dresses
It sets, the apparatus may include: non-court verdict obtains module 701, measurement of penalty element determining module 702 and measurement of penalty prediction of result module
703。
Non- court verdict obtains module 701, the non-court verdict of the specified case for obtaining specified charge, sentences as target is non-
Certainly book.
Measurement of penalty element determining module 702, it is corresponding with specified measurement of penalty element label for being obtained from the non-court verdict of target
Measurement of penalty element, as target measurement of penalty element.
Measurement of penalty prediction of result module 703, for target measurement of penalty element to be inputted the measurement of penalty prediction of result model pre-established,
Obtain the measurement of penalty result of the specified case of measurement of penalty prediction of result model output.
Wherein, measurement of penalty prediction of result model is with extract from the training court verdict of specified charge and specified measurement of penalty element
The corresponding measurement of penalty element of label is training sample, with the extract and specified court verdict element label pair from training court verdict
The court verdict element answered is that sample label is trained to obtain.
Measurement of penalty prediction of result device provided by the embodiments of the present application, the target for obtaining the specified case of specified charge first are non-
Then court verdict obtains target measurement of penalty element corresponding with specified measurement of penalty element label, finally by mesh from the non-court verdict of target
Scalar punishment element inputs the measurement of penalty prediction of result model pre-established, obtains the specified case of measurement of penalty prediction of result model output
Measurement of penalty result.Measurement of penalty prediction of result device provided by the embodiments of the present application can be based on the non-court verdict of specified case, using preparatory
The measurement of penalty prediction of result model automatic Prediction of foundation go out the measurement of penalty as a result, and the measurement of penalty result that predicts it is more accurate, the measurement of penalty knot
Fruit has preferable auxiliary measurement of penalty effect for judge's reference.
Measurement of penalty prediction of result device provided by the above embodiment can also include: court verdict element abstraction module.
Court verdict element abstraction module, for utilizing the court verdict element extraction model pre-established, never mark judgement
Extracted in book measurement of penalty element corresponding with the specified measurement of penalty element label and with the specified court verdict element label pair
The court verdict element answered.
Wherein, the court verdict element extraction model is to be labeled with specified measurement of penalty element label and specified court verdict element
The training court verdict of label is trained to obtain.
Further, above-mentioned court verdict element abstraction module is specifically used for passing through the court verdict element extraction model
In semantic vector determining module, to it is described do not mark court verdict carry out word segmentation processing, determine each word that word segmentation processing obtains
Corresponding semantic vector;Pass through the element label determining module and each word correspondence in the court verdict element extraction model
Semantic vector, determine the element label of corresponding with each semantic vector word;The identical continuous multiple words of element label are closed
And the content after merging is as the corresponding element of element label.
In one possible implementation, measurement of penalty element determining module 702 provided by the above embodiment is specifically used for benefit
With the non-court verdict element extraction model pre-established, extracted and the specified measurement of penalty element mark from the non-court verdict of the target
Sign corresponding measurement of penalty element.
Wherein, the non-court verdict element extraction model is using the non-court verdict of training as training sample, based on described specified
Measurement of penalty element label is that sample label is trained to obtain to the annotation results that the non-court verdict of the training is labeled.
The non-court verdict is labeled based on the specified measurement of penalty element label in one possible implementation
Annotation results include: measurement of penalty element corresponding with the specified amount type element label in the non-court verdict of the training start bit
Set with initial position score, end position and end position score, and, measurement of penalty element is empty score.
Then measurement of penalty element determining module 702 includes: that measurement of penalty element information determines that submodule and measurement of penalty element determine submodule.
Measurement of penalty element information determines submodule, for determining that the target is non-by the non-court verdict element extraction model
The information of measurement of penalty element corresponding with the specified measurement of penalty element label is as target measurement of penalty element information in court verdict.
The target measurement of penalty element information includes: corresponding with the specified measurement of penalty element label in the non-court verdict of the target
Measurement of penalty element initial position and initial position score, end position and end position score and measurement of penalty element be empty
Score.
Measurement of penalty element determines submodule, for determining the non-court verdict of the target by the target measurement of penalty element information
In measurement of penalty element corresponding with the specified measurement of penalty element label.
Further, measurement of penalty element information determines submodule, is specifically used for the non-court verdict of the target and problem label
Text inputs the non-court verdict element extraction model, obtains the target answer of the non-court verdict element extraction model output
Information, as the target measurement of penalty element information.
Wherein, problem label text is the text comprising given problem label, and the given problem label is by the finger
Quantitative punishment element label is converted to the label after problem form;The target answer information includes and the given problem label pair
The initial position and initial position score of the answer answered, end position and end position score, and, answer is empty score.
Further, measurement of penalty element determines submodule, if being specifically used for existing in the given problem label with multiple
The target problem label of answer, then removal is unsatisfactory for answering for preset condition from the target problem label corresponding multiple answers
Case obtains remaining answer;If the residue answer be it is multiple, will multiple remaining answers progress duplicate removal merging treatments, go
The answer obtained after weight merging treatment is as the target measurement of penalty element.
Further, measurement of penalty element determines that submodule is removed from the corresponding multiple answers of the target problem label
It is unsatisfactory for the answer of preset condition, when obtaining remaining answer, is specifically used for from the corresponding multiple answers of the target problem label
Default answer of middle determination forms the first candidate answers set, wherein the score of the default answer is above other answers
Score, the score of each answer determined by the initial position score and end position score of the answer;Described first is waited
It selects score in answer set to remove lower than the candidate answers of first threshold, obtains the second candidate answers set, wherein described first
Threshold value is that the corresponding answer of the target problem label is empty score;By score in the second candidate answers set lower than the
The candidate answers of the non-top score of two threshold values remove, and obtain the remaining answer, wherein the second threshold be based on it is described most
The score of the candidate answers of high score is set.
Measurement of penalty element determines submodule, is also used to when the score of each candidate answers in the first candidate answers set is equal
When lower than the first threshold, determine the corresponding answer of the target problem label for sky.
The embodiment of the present application also provides a kind of measurement of penalty prediction of result equipment, referring to Fig. 7, it is pre- to show the measurement of penalty result
The structural schematic diagram of measurement equipment may include: memory 801 and processor 802.
Memory 801, for storing program;
Processor 802, for executing described program, described program is specifically used for:
The non-court verdict for obtaining specified charge, as the non-court verdict of target;
The specified corresponding measurement of penalty element of measurement of penalty element label is obtained from the non-court verdict of the target, is wanted as the target measurement of penalty
Element;
The target measurement of penalty element is inputted to the measurement of penalty prediction of result model pre-established, obtains the measurement of penalty prediction of result
The measurement of penalty result of model output;
Wherein, the measurement of penalty prediction of result model with it is being extracted from the training court verdict of the specified charge, with it is described
The specified corresponding measurement of penalty element of measurement of penalty element label is training sample, with it is being extracted from the trained court verdict, sentence with specifying
Certainly the corresponding court verdict element of result element label is that sample label is trained to obtain.
Measurement of penalty prediction of result equipment can also include: bus, communication interface 803, input equipment 804 and output equipment 805.
Processor 802, memory 801, communication interface 803, input equipment 804 and output equipment 805 are mutual by bus
Connection.Wherein:
Bus may include an access, transmit information between computer system all parts.
Processor 1102 can be general processor, such as general central processor (CPU), microprocessor etc., can also be with
It is application-specific integrated circuit (application-specific integrated circuit, ASIC), or one or more
For controlling the integrated circuit of the present invention program program execution.It can also be digital signal processor (DSP), specific integrated circuit
(ASIC), ready-made programmable gate array (FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components.
Processor 802 may include primary processor, may also include baseband chip, modem etc..
The program for executing technical solution of the present invention is preserved in memory 801, can also preserve operating system and other
Key business.Specifically, program may include program code, and program code includes computer operation instruction.More specifically, it stores
Device 801 may include read-only memory (read-only memory, ROM), the other types that can store static information and instruction
Static storage device, random access memory (random access memory, RAM), can store information and instruction its
The dynamic memory of his type, magnetic disk storage, flash etc..
Input equipment 804 may include the device for receiving the data and information of user's input, such as camera, light pen, touch
Screen etc..
Output equipment 805 may include allowing output information to the device, such as display screen, loudspeaker etc. of user.
Communication interface 803 may include using the device of any transceiver one kind, so as to logical with other equipment or communication network
Letter, such as Ethernet, wireless access network (RAN), WLAN (WLAN) etc..
Processor 802 executes the program stored in memory 801, and calls other equipment, can be used for realizing this hair
Each step of measurement of penalty prediction of result method provided by bright embodiment.
The embodiment of the present application also provides a kind of readable storage medium storing program for executing, are stored thereon with computer program, the computer
When program is executed by processor, each step of measurement of penalty prediction of result method provided by the above embodiment is realized.
Each embodiment in this specification is described in a progressive manner, the highlights of each of the examples are with other
The difference of embodiment, the same or similar parts in each embodiment may refer to each other.
In several embodiments provided herein, it should be understood that disclosed method, apparatus and equipment, it can be with
It realizes by another way.For example, the apparatus embodiments described above are merely exemplary, for example, the unit
It divides, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components
It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or
The mutual coupling, direct-coupling or communication connection discussed can be by some communication interfaces, between device or unit
Coupling or communication connection are connect, can be electrical property, mechanical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
In network unit.It can select some or all of unit therein according to the actual needs to realize the mesh of this embodiment scheme
's.In addition, the functional units in various embodiments of the present invention may be integrated into one processing unit, it is also possible to each
Unit physically exists alone, and can also be integrated in one unit with two or more units.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest scope of cause.
Claims (13)
1. a kind of measurement of penalty prediction of result method characterized by comprising
The non-court verdict for obtaining the specified case of specified charge, as the non-court verdict of target;
Measurement of penalty element corresponding with specified measurement of penalty element label is obtained from the non-court verdict of the target, is wanted as the target measurement of penalty
Element;
The target measurement of penalty element is inputted to the measurement of penalty prediction of result model pre-established, obtains the measurement of penalty prediction of result model
The measurement of penalty result of the specified case of output;
Wherein, the measurement of penalty prediction of result model from the training court verdict of the specified charge with extracting and described specified
The corresponding measurement of penalty element of measurement of penalty element label is training sample, with judgement knot extract from the trained court verdict and specified
The corresponding court verdict element of fruit element label is that sample label is trained to obtain.
2. measurement of penalty prediction of result method according to claim 1, which is characterized in that the training from the specified charge
Extracted in court verdict measurement of penalty element corresponding with the specified measurement of penalty element label and with the specified court verdict element mark
The process for signing corresponding court verdict element includes:
Using the court verdict element extraction model pre-established, never extracted and the specified measurement of penalty element mark in mark court verdict
Sign corresponding measurement of penalty element and court verdict element corresponding with the specified court verdict element label;
Wherein, the court verdict element extraction model is to be labeled with specified measurement of penalty element label and specified court verdict element label
Training court verdict be trained to obtain.
3. measurement of penalty prediction of result method according to claim 2, which is characterized in that the court verdict that the utilization pre-establishes
Element extraction model, from it is described do not mark extracted in court verdict measurement of penalty element corresponding with the specified measurement of penalty element label and
Court verdict element corresponding with the specified court verdict element label, comprising:
By the semantic vector determining module in the court verdict element extraction model, the court verdict that do not mark is segmented
Processing, determines the corresponding semantic vector of each word that word segmentation processing obtains;
By the element label determining module and the corresponding semantic vector of each word in the court verdict element extraction model,
Determine the element label of word corresponding with each semantic vector;
The identical continuous multiple words of element label are merged, the content after merging is as the corresponding element of element label.
4. measurement of penalty prediction of result method according to claim 1, which is characterized in that described from the non-court verdict of the target
Obtain measurement of penalty element corresponding with specified measurement of penalty element label, comprising:
Using the non-court verdict element extraction model pre-established, extracted and the specified measurement of penalty from the non-court verdict of the target
The corresponding measurement of penalty element of element label;
Wherein, the non-court verdict element extraction model is using the non-court verdict of training as training sample, to be based on the specified measurement of penalty
Element label is that sample label is trained to obtain to the annotation results that the non-court verdict of the training is labeled.
5. measurement of penalty prediction of result method according to claim 4, which is characterized in that described to be based on the specified measurement of penalty element
The annotation results that label is labeled the non-court verdict include:
The initial position of measurement of penalty element corresponding with the specified amount type element label and start bit in the non-court verdict of training
Score, end position and end position score are set, and, measurement of penalty element is empty score.
6. measurement of penalty prediction of result method according to claim 5, which is characterized in that the non-judgement that the utilization pre-establishes
Book element extraction model extracts measurement of penalty element corresponding with the specified measurement of penalty element label from the non-court verdict of the target,
Include:
By the non-court verdict element extraction model determine in the non-court verdict of the target with the specified measurement of penalty element label
As target measurement of penalty element information, the target measurement of penalty element information includes that the target is non-to be sentenced the information of corresponding measurement of penalty element
Certainly the initial position of measurement of penalty element corresponding with the specified measurement of penalty element label and initial position score in book, end position and
End position score and measurement of penalty element are empty score;
By the target measurement of penalty element information, determine corresponding with the specified measurement of penalty element label in the non-court verdict of the target
Measurement of penalty element.
7. measurement of penalty prediction of result method according to claim 6, which is characterized in that described to pass through the non-court verdict element
Extraction model determines the information conduct of measurement of penalty element corresponding with the specified measurement of penalty element label in the non-court verdict of the target
Target measurement of penalty element information, comprising:
By the non-court verdict of the target and problem label text, the non-court verdict element extraction model is inputted, is obtained described non-
The target answer information of court verdict element extraction model output, as the target measurement of penalty element information;
Wherein, problem label text is the text comprising given problem label, and the given problem label is by the specified amount
Punishment element label is converted to the label after problem form;The target answer information includes corresponding with the given problem label
The initial position and initial position score of answer, end position and end position score, and, answer is empty score.
8. measurement of penalty prediction of result method according to claim 7, which is characterized in that described to pass through the target measurement of penalty element
Information determines measurement of penalty element corresponding with the specified measurement of penalty element label in the non-court verdict of the target, comprising:
If there is the target problem label with multiple answers in the given problem label, from the target problem label pair
Removal is unsatisfactory for the answer of preset condition in the multiple answers answered, and obtains remaining answer;
If the residue answer be it is multiple, will multiple remaining answers progress duplicate removal merging treatments, after duplicate removal merging treatment
Obtained answer is as the target measurement of penalty element.
9. measurement of penalty prediction of result method according to claim 7, which is characterized in that described from the target problem label pair
Removal is unsatisfactory for the answer of preset condition in the multiple answers answered, and obtains remaining answer, comprising:
Determine that default answer forms the first candidate answers set from the corresponding multiple answers of the target problem label,
In, the score of the default answer is above the score of other answers, and the score of each answer passes through the start bit of the answer
It sets score and end position score determines;
Candidate answers by score in the first candidate answers set lower than first threshold remove, and obtain the second candidate answers collection
It closes, wherein the first threshold is that the corresponding answer of the target problem label is empty score;
Candidate answers removal by score in the second candidate answers set lower than the non-top score of second threshold, obtains institute
State remaining answer, wherein the score of candidate answers of the second threshold based on the top score is set.
10. measurement of penalty prediction of result method according to claim 9, which is characterized in that described from the target problem label
The answer for being unsatisfactory for preset condition is filtered out in corresponding multiple answers, obtains remaining answer, further includes:
If the score of each candidate answers is below the first threshold in the first candidate answers set, it is determined that the mesh
The corresponding answer of mark problem label is sky.
11. a kind of measurement of penalty prediction of result device characterized by comprising non-court verdict obtains module, measurement of penalty element determining module
With measurement of penalty prediction of result module;
The non-court verdict obtains module, the non-court verdict of the specified case for obtaining specified charge, as the non-judgement of target
Book;
The measurement of penalty element determining module, it is corresponding with specified measurement of penalty element label for being obtained from the non-court verdict of the target
Measurement of penalty element, as target measurement of penalty element;
The measurement of penalty prediction of result module, for the target measurement of penalty element to be inputted the measurement of penalty prediction of result mould pre-established
Type obtains the measurement of penalty result of the specified case of the measurement of penalty prediction of result model output;
Wherein, the measurement of penalty prediction of result model from the training court verdict of the specified charge with extracting and described specified
The corresponding measurement of penalty element of measurement of penalty element label is training sample, with judgement knot extract from the trained court verdict and specified
The corresponding court verdict element of fruit element label is that sample label is trained to obtain.
12. a kind of measurement of penalty prediction of result equipment characterized by comprising memory and processor;
The memory, for storing program;
The processor, for executing described program, described program is specifically used for:
The non-court verdict for obtaining the specified case of specified charge, as the non-court verdict of target;
Measurement of penalty element corresponding with specified measurement of penalty element label is obtained from the non-court verdict of the target, is wanted as the target measurement of penalty
Element;
The target measurement of penalty element is inputted to the measurement of penalty prediction of result model pre-established, obtains the measurement of penalty prediction of result model
The measurement of penalty result of the specified case of output;
Wherein, the measurement of penalty prediction of result model from the training court verdict of the specified charge with extracting and described specified
The corresponding measurement of penalty element of measurement of penalty element label is training sample, with judgement knot extract from the trained court verdict and specified
The corresponding court verdict element of fruit element label is that sample label is trained to obtain.
13. a kind of readable storage medium storing program for executing, is stored thereon with computer program, which is characterized in that the computer program is processed
When device executes, each step of the measurement of penalty prediction of result method as described in any one of claims 1 to 10 is realized.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810971990.1A CN109241528B (en) | 2018-08-24 | 2018-08-24 | Criminal investigation result prediction method, device, equipment and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810971990.1A CN109241528B (en) | 2018-08-24 | 2018-08-24 | Criminal investigation result prediction method, device, equipment and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109241528A true CN109241528A (en) | 2019-01-18 |
CN109241528B CN109241528B (en) | 2023-09-01 |
Family
ID=65069041
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810971990.1A Active CN109241528B (en) | 2018-08-24 | 2018-08-24 | Criminal investigation result prediction method, device, equipment and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109241528B (en) |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109949185A (en) * | 2019-03-15 | 2019-06-28 | 南京邮电大学 | Judicial case judgement system and method based on Event Tree Analysis |
CN110210031A (en) * | 2019-05-31 | 2019-09-06 | 吉林中科结诚科技有限公司 | A kind of merit intelligent identification Method and system |
CN110334217A (en) * | 2019-05-10 | 2019-10-15 | 科大讯飞股份有限公司 | A kind of element abstracting method, device, equipment and storage medium |
CN110489546A (en) * | 2019-07-11 | 2019-11-22 | 深圳追一科技有限公司 | Case load penalizes determination method, apparatus, computer equipment and the storage medium of index |
CN110610005A (en) * | 2019-09-16 | 2019-12-24 | 哈尔滨工业大学 | Stealing crime auxiliary criminal investigation method based on deep learning |
CN110738039A (en) * | 2019-09-03 | 2020-01-31 | 平安科技(深圳)有限公司 | Prompting method, device, storage medium and server for case auxiliary information |
CN111259673A (en) * | 2020-01-13 | 2020-06-09 | 山东财经大学 | Feedback sequence multi-task learning-based law decision prediction method and system |
CN111325387A (en) * | 2020-02-13 | 2020-06-23 | 清华大学 | Interpretable law automatic decision prediction method and device |
CN111783472A (en) * | 2020-06-30 | 2020-10-16 | 鼎富智能科技有限公司 | Judgment book content extraction method and related device |
CN112559754A (en) * | 2019-09-25 | 2021-03-26 | 北京国双科技有限公司 | Judgment result processing method and device |
CN112579732A (en) * | 2019-09-30 | 2021-03-30 | 北京国双科技有限公司 | Sentencing prediction method and device |
CN112818996A (en) * | 2021-01-29 | 2021-05-18 | 青岛海尔科技有限公司 | Instruction identification method and device, storage medium and electronic equipment |
CN113408263A (en) * | 2020-03-16 | 2021-09-17 | 北京国双科技有限公司 | Criminal period prediction method and device, storage medium and electronic device |
CN116823541A (en) * | 2023-08-29 | 2023-09-29 | 山东大学 | Criminal investigation calculation method and system based on nonlinear model |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN107562938A (en) * | 2017-09-21 | 2018-01-09 | 重庆工商大学 | A kind of law court intelligently tries method |
WO2018023981A1 (en) * | 2016-08-03 | 2018-02-08 | 平安科技(深圳)有限公司 | Public opinion analysis method, device, apparatus and computer readable storage medium |
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 |
CN108153732A (en) * | 2017-12-25 | 2018-06-12 | 科大讯飞股份有限公司 | The checking method and device of a kind of hearing record |
WO2018113498A1 (en) * | 2016-12-23 | 2018-06-28 | 北京国双科技有限公司 | Method and apparatus for retrieving legal knowledge |
-
2018
- 2018-08-24 CN CN201810971990.1A patent/CN109241528B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2018023981A1 (en) * | 2016-08-03 | 2018-02-08 | 平安科技(深圳)有限公司 | Public opinion analysis method, device, apparatus and computer readable storage medium |
WO2018113498A1 (en) * | 2016-12-23 | 2018-06-28 | 北京国双科技有限公司 | Method and apparatus for retrieving legal knowledge |
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 |
CN107562938A (en) * | 2017-09-21 | 2018-01-09 | 重庆工商大学 | A kind of law court intelligently tries method |
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 |
CN108153732A (en) * | 2017-12-25 | 2018-06-12 | 科大讯飞股份有限公司 | The checking method and device of a kind of hearing record |
Non-Patent Citations (2)
Title |
---|
佘贵清等: "审判案例自动抽取与标注模型研究", 《现代图书情报技术》 * |
张德: "自然语言处理技术在司法过程中的应用研究", 《信息与电脑(理论版)》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109949185A (en) * | 2019-03-15 | 2019-06-28 | 南京邮电大学 | Judicial case judgement system and method based on Event Tree Analysis |
CN110334217A (en) * | 2019-05-10 | 2019-10-15 | 科大讯飞股份有限公司 | A kind of element abstracting method, device, equipment and storage medium |
CN110334217B (en) * | 2019-05-10 | 2021-10-08 | 科大讯飞股份有限公司 | Element extraction method, device, equipment and storage medium |
CN110210031A (en) * | 2019-05-31 | 2019-09-06 | 吉林中科结诚科技有限公司 | A kind of merit intelligent identification Method and system |
CN110489546A (en) * | 2019-07-11 | 2019-11-22 | 深圳追一科技有限公司 | Case load penalizes determination method, apparatus, computer equipment and the storage medium of index |
CN110738039A (en) * | 2019-09-03 | 2020-01-31 | 平安科技(深圳)有限公司 | Prompting method, device, storage medium and server for case auxiliary information |
CN110610005A (en) * | 2019-09-16 | 2019-12-24 | 哈尔滨工业大学 | Stealing crime auxiliary criminal investigation method based on deep learning |
CN112559754A (en) * | 2019-09-25 | 2021-03-26 | 北京国双科技有限公司 | Judgment result processing method and device |
WO2021057202A1 (en) * | 2019-09-25 | 2021-04-01 | 北京国双科技有限公司 | Method and apparatus for processing judgement result |
CN112579732A (en) * | 2019-09-30 | 2021-03-30 | 北京国双科技有限公司 | Sentencing prediction method and device |
CN111259673A (en) * | 2020-01-13 | 2020-06-09 | 山东财经大学 | Feedback sequence multi-task learning-based law decision prediction method and system |
CN111259673B (en) * | 2020-01-13 | 2023-05-09 | 山东财经大学 | Legal decision prediction method and system based on feedback sequence multitask learning |
CN111325387A (en) * | 2020-02-13 | 2020-06-23 | 清华大学 | Interpretable law automatic decision prediction method and device |
CN113408263A (en) * | 2020-03-16 | 2021-09-17 | 北京国双科技有限公司 | Criminal period prediction method and device, storage medium and electronic device |
CN111783472A (en) * | 2020-06-30 | 2020-10-16 | 鼎富智能科技有限公司 | Judgment book content extraction method and related device |
CN112818996A (en) * | 2021-01-29 | 2021-05-18 | 青岛海尔科技有限公司 | Instruction identification method and device, storage medium and electronic equipment |
CN116823541A (en) * | 2023-08-29 | 2023-09-29 | 山东大学 | Criminal investigation calculation method and system based on nonlinear model |
Also Published As
Publication number | Publication date |
---|---|
CN109241528B (en) | 2023-09-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109241528A (en) | A kind of measurement of penalty prediction of result method, apparatus, equipment and storage medium | |
JP6894534B2 (en) | Information processing method and terminal, computer storage medium | |
CN110032623B (en) | Method and device for matching question of user with title of knowledge point | |
CN109271493A (en) | A kind of language text processing method, device and storage medium | |
CN108319720A (en) | Man-machine interaction method, device based on artificial intelligence and computer equipment | |
CN113742488B (en) | Embedded knowledge graph completion method and device based on multitask learning | |
CN109271624A (en) | A kind of target word determines method, apparatus and storage medium | |
WO2020063524A1 (en) | Method and system for determining legal instrument | |
CN113220862A (en) | Standard question recognition method and device, computer equipment and storage medium | |
CN114818718B (en) | Contract text recognition method and device | |
CN111475607A (en) | Web data clustering method based on Mashup service function characteristic representation and density peak detection | |
CN112307048A (en) | Semantic matching model training method, matching device, equipment and storage medium | |
CN110209772B (en) | Text processing method, device and equipment and readable storage medium | |
CN112632248A (en) | Question answering method, device, computer equipment and storage medium | |
CN108197660A (en) | Multi-model Feature fusion/system, computer readable storage medium and equipment | |
CN113204643B (en) | Entity alignment method, device, equipment and medium | |
CN109065015B (en) | Data acquisition method, device and equipment and readable storage medium | |
CN115438158A (en) | Intelligent dialogue method, device, equipment and storage medium | |
CN111125379B (en) | Knowledge base expansion method and device, electronic equipment and storage medium | |
CN111680514B (en) | Information processing and model training method, device, equipment and storage medium | |
CN115759048A (en) | Script text processing method and device | |
CN113051869B (en) | Method and system for realizing identification of text difference content by combining semantic recognition | |
CN113158688B (en) | Domain knowledge base construction method, device, equipment and storage medium | |
CN114780700A (en) | Intelligent question-answering method, device, equipment and medium based on machine reading understanding | |
CN109033078A (en) | The recognition methods of sentence classification and device, storage medium, processor |
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 |