CN109241528B - Criminal investigation result prediction method, device, equipment and storage medium - Google Patents
Criminal investigation result prediction method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN109241528B CN109241528B CN201810971990.1A CN201810971990A CN109241528B CN 109241528 B CN109241528 B CN 109241528B CN 201810971990 A CN201810971990 A CN 201810971990A CN 109241528 B CN109241528 B CN 109241528B
- Authority
- CN
- China
- Prior art keywords
- criminal
- label
- book
- sentencing
- target
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000011840 criminal investigation Methods 0.000 title claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 93
- 238000000605 extraction Methods 0.000 claims description 68
- 239000013598 vector Substances 0.000 claims description 66
- 206010014405 Electrocution Diseases 0.000 claims description 18
- 238000012545 processing Methods 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 2
- 238000002372 labelling Methods 0.000 description 15
- 238000010586 diagram Methods 0.000 description 8
- 238000004891 communication Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 6
- 238000013473 artificial intelligence Methods 0.000 description 4
- 230000002457 bidirectional effect Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000013178 mathematical model Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 230000001276 controlling effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000010801 machine learning Methods 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
- 238000011022 operating instruction Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 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
The application provides a method, a device, equipment and a storage medium for predicting a sentencing result, wherein the method comprises the following steps: acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book; acquiring a criminal element corresponding to a designated criminal element label from a target non-judgment book as a target criminal element; inputting a target criminal investigation element into a pre-established criminal investigation result prediction model to obtain a criminal investigation result of a specified case output by the criminal investigation result prediction model; the criminal result prediction model is obtained by taking criminal elements which are extracted from a training judgment book with appointed criminal names and correspond to appointed criminal element labels as training samples and taking judgment result elements which are extracted from the training judgment book and correspond to the appointed judgment result element labels as sample labels. The criminal result prediction method, the device, the equipment and the storage medium provided by the application can automatically predict the relatively accurate criminal result, and the criminal result can be used for reference by a judge to assist the judge.
Description
Technical Field
The application relates to the technical field of artificial intelligence, in particular to a method, a device, equipment and a storage medium for predicting a criminal result.
Background
In recent years, with rapid development of big data and artificial intelligence technology, the use of machines to assist in the manual work has become a hotspot direction for various industries.
The judicial informatization construction is an important direction for realizing modernization of China judicial, and the judicial related work is also referred to data and legal documents by the traditional first-line judges aiming at different cases, and gradually evolved into a first-line judges to finish matters such as court trial records, case analysis and the like with the aid of a machine, so that a construction system of a smart court appears.
The whole architecture of the intelligent court comprises four aspects of using artificial intelligence technology and big data technology for serving the public of society, judging service cases, executing service judgment and managing service judicial. However, for the application scenario of service case trial, no solution capable of automatic criminal investigation exists at present.
Disclosure of Invention
In view of the above, the application provides a method, a device, equipment and a storage medium for predicting a criminal result, which are used for automatically predicting the criminal result for reference by a judge based on a case related document, and the technical scheme is as follows:
a method of sentencing results prediction comprising:
acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book;
Acquiring a criminal element corresponding to a specified criminal element label from the target non-judgment book, and taking the criminal element as a target criminal element;
inputting the target criminal investigation element into a pre-established criminal investigation result prediction model to obtain the criminal investigation result of the specified case output by the criminal investigation result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
The process of extracting the electrocuting element corresponding to the appointed electrocuting element label and the judging result element corresponding to the appointed judging result element label from the training judging book with the appointed crime name comprises the following steps:
extracting the sentencing element corresponding to the appointed sentencing element label and the judgment result element corresponding to the appointed judgment result element label from the unlabeled judgment book by using a pre-established judgment book element extraction model;
the judgment book element extraction model is obtained by training a training judgment book marked with a designated criminal element label and a designated judgment result element label.
The method for extracting the electrocuting element corresponding to the appointed electrocuting element label and the judging result element corresponding to the appointed judging result element label from the unlabeled judging book by utilizing a pre-established judging book element extraction model comprises the following steps:
the semantic vector determining module in the judgment book element extraction model is used for carrying out word segmentation on the unlabeled judgment books, and determining semantic vectors corresponding to each word obtained through word segmentation;
determining element labels of words corresponding to each semantic vector through an element label determining module in the judgment book element extraction model and the semantic vector corresponding to each word;
and merging the continuous words with the same element label, wherein the merged content is taken as the element corresponding to the element label.
The acquiring the electrocuting element corresponding to the appointed electrocuting element label from the target non-judgment book comprises the following steps:
extracting the sentencing element corresponding to the appointed sentencing element label from the target non-judgment book by using a pre-established non-judgment book element extraction model;
the non-decision book element extraction model is obtained by training a training non-decision book as a training sample and training a labeling result of labeling the training non-decision book based on the appointed sentencing element label as a sample label.
The labeling result of labeling the non-decision book based on the appointed criminal element label comprises:
and the starting position score, the ending position and the ending position score of the sentencing element corresponding to the appointed amount element label in the training non-judgment book, and the score that the sentencing element is empty.
The method for extracting the electrocuting element corresponding to the appointed electrocuting element label from the target non-decision book by utilizing a pre-established non-decision book element extraction model comprises the following steps:
determining information of the sentencing element corresponding to the appointed sentencing element label in the target non-judgment book as target sentencing element information through the non-judgment book element extraction model, wherein the target sentencing element information comprises a starting position score, an ending position score and an ending position score of the sentencing element corresponding to the appointed sentencing element label in the target non-judgment book, and a score that the sentencing element is empty;
and determining the criminal element corresponding to the appointed criminal element label in the target non-judgment book through the target criminal element information.
The determining, by the non-decision element extraction model, information of an electrocuting element corresponding to the specified electrocuting element tag in the target non-decision as target electrocuting element information includes:
Inputting the target non-decision book and the question label text into the non-decision book element extraction model to obtain target answer information output by the non-decision book element extraction model as the target sentencing element information;
the text of the question label is a text containing a specified question label, and the specified question label is a label obtained by converting the specified criminal element label into a question form; the target answer information includes a starting position and a starting position score, an ending position and an ending position score of an answer corresponding to the specified question label, and a score that the answer is null.
The determining, by the target electrocuting element information, electrocuting elements in the target non-decision book corresponding to the specified electrocuting element tag includes:
if the target question label with a plurality of answers exists in the specified question label, removing the answers which do not meet the preset condition from the plurality of answers corresponding to the target question label, and obtaining the rest answers;
if the number of the residual answers is multiple, performing de-coincidence processing on the multiple residual answers, and taking the answers obtained after de-coincidence processing as the target criminal investigation elements.
The step of removing the answers which do not meet the preset condition from the answers corresponding to the target question label to obtain the remaining answers includes:
determining a preset number of answers from a plurality of answers corresponding to the target question label to form a first candidate answer set, wherein the scores of the preset number of answers are higher than the scores of other answers, and the score of each answer is determined through the starting position score and the ending position score of the answer;
removing candidate answers with scores lower than a first threshold value in the first candidate answer set to obtain a second candidate answer set, wherein the first threshold value is a score that an answer corresponding to the target question label is empty;
and removing the non-highest scoring candidate answers with scores lower than a second threshold value in the second candidate answer set to obtain the residual answers, wherein the second threshold value is set based on the scores of the highest scoring candidate answers.
The filtering, from a plurality of answers corresponding to the target question label, the answer that does not meet the preset condition, to obtain the remaining answers, further includes:
and if the scores of the candidate answers in the first candidate answer set are lower than the first threshold value, determining that the answer corresponding to the target question label is empty.
A sentencing result prediction device comprising: the system comprises a non-decision acquisition module, a criminal element determination module and a criminal result prediction module;
the non-judgment book acquisition module is used for acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book;
the criminal element determining module is used for acquiring the criminal element corresponding to the appointed criminal element label from the target non-judgment book and taking the criminal element as a target criminal element;
the crime result prediction module is used for inputting the target crime element into a pre-established crime result prediction model to obtain the crime result of the specified case output by the crime result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
A sentencing result prediction apparatus comprising: a memory and a processor;
the memory is used for storing programs;
the processor is configured to execute the program, where the program is specifically configured to:
Acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book;
acquiring a criminal element corresponding to a specified criminal element label from the target non-judgment book, and taking the criminal element as a target criminal element;
inputting the target criminal investigation element into a pre-established criminal investigation result prediction model to obtain the criminal investigation result of the specified case output by the criminal investigation result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
A readable storage medium having stored thereon a computer program, characterized in that the computer program, when executed by a processor, implements the steps of the above-mentioned sentencing result prediction method.
According to the crime result prediction method, the crime result prediction device, the crime result prediction equipment and the crime result storage medium, a target non-judgment book of a specified case with a specified crime name is firstly obtained, then a target crime element corresponding to a specified crime element label is obtained from the target non-judgment book, and finally the target crime element is input into a pre-established crime result prediction model to obtain a crime result output by the crime result prediction model. Therefore, the criminal result prediction method, the criminal result prediction device, the criminal result prediction equipment and the storage medium can automatically predict the criminal result of the appointed case by utilizing the pre-established criminal result prediction model based on the non-judgment book of the appointed case, and the criminal result can be used for reference by a judge to assist the judge to judge the appointed case, so that the judging efficiency of the case can be improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only embodiments of the present application, and that other drawings can be obtained according to the provided drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of a method for predicting a criminal result according to an embodiment of the present application;
fig. 2 is a schematic flow chart of a realization process of extracting an sentencing element corresponding to a designated sentencing element label and a decision result element corresponding to the designated decision result element label from an unlabeled decision book by using a pre-established decision book element extraction model according to an embodiment of the present application;
fig. 3 is a schematic diagram of an example of a topology structure of a decision element extraction model according to an embodiment of the present application;
fig. 4 is a schematic flow chart of an implementation process of extracting an electrocution element corresponding to a specified electrocution element label from a target non-decision book by using a non-decision book element extraction model according to an embodiment of the present application;
FIG. 5 is a schematic diagram of an example of a topology of a non-decision element extraction model according to an embodiment of the present application;
fig. 6 is a flowchart illustrating an implementation process of removing answers that do not meet a preset condition from a plurality of answers corresponding to a target question label to obtain remaining answers according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of a criminal result prediction device provided by the embodiment of the application;
fig. 8 is a schematic structural diagram of a criminal result prediction device provided by the embodiment of the application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
The inventor finds in the process of realizing the application: there are some sentency outcome prediction schemes in the prior art, most of which are: extracting scenario information of legal documents by adopting regular expressions, context-free grammar, named entity recognition and other modes, structuring the extracted information, describing the characteristics and rules of data by using mathematical models, calculating the mathematical model or algorithm most conforming to the mathematical model or algorithm, and deducing a sentencing result according to the scenario by artificial intelligence through a simulation algorithm. However, the existing solution for predicting the sentencing results is basically based on the traditional machine learning method, more rules (regular expressions), features (context-free questions and answers, named entity recognition) and the like are required to be designed, a great deal of manual intervention is required, and due to the fact that the related documents of law are described in a complex manner, the solution is difficult to popularize into more solutions, and the auxiliary sentencing effect is poor.
In order to overcome the problems of the existing solutions, the embodiment of the present application provides a method for predicting a sentencing result, please refer to fig. 1, which shows a flow chart of the method for predicting a sentencing result, and may include:
step S101: and acquiring a non-judgment book of the appointed case of the appointed crime name as a target non-judgment book.
The appointed crime name can be any crime name regulated by law, such as robbing crimes, theft crimes, fraud crimes, money stealing crimes, robbing crimes, frisby crimes and the like, namely the crime result prediction method provided by the application has a wider application range.
The non-decision book can be, but not limited to, an inquiry, a court trial, a prosecution, etc.
It should be noted that the non-decision book of the specified case may be one or multiple, for example, the query records may be made for different objects, so that multiple query records may be obtained.
Step S102: and acquiring the sentencing element corresponding to the appointed sentencing element label from the target non-judgment book as a target sentencing element.
It should be noted that the specified criminal element label is set based on the specified criminal name, and the specified criminal element corresponding to the specified criminal element label refers to the content corresponding to the specified criminal element label in the target non-judgment book.
Step S103: and inputting the target criminal investigation element into a pre-established criminal investigation result prediction model to obtain the criminal investigation result of the specified case output by the criminal investigation result prediction model.
The crime result prediction model is obtained by taking criminal elements which are extracted from a training judgment book with appointed criminal names and correspond to appointed criminal element labels as training samples and taking judgment result elements which are extracted from the training judgment book and correspond to the appointed judgment result element labels as sample labels.
It should be noted that, the writing format of the decision book is generally an unconventional model, and the information in the decision book mainly includes: the basic information of the reported person, the criminal related scenario and the criminal result, therefore, the criminal element label and the judgment result element label can be set based on the content of the judgment book of the appointed criminal name.
Taking the designated crime as a theft crime as an example, the criminal element label may be set to include: whether burglary, theft, forecourt, disabled person, theft amount, congregation attitude, tank, self-first, etc., can be set, and the specified decision element labels comprise principal crimes, additional crimes, etc.
In addition, considering that the target crime element is discrete data, the crime result prediction model may be, but not limited to, an SVM model, a logistic regression model, a deep neural network model, and the like.
According to the crime result prediction method provided by the embodiment of the application, firstly, the target non-judgment book of the appointed case with the appointed crime name is obtained, then, the target crime element corresponding to the appointed crime element label is obtained from the target non-judgment book, and finally, the target crime element is input into the pre-established crime result prediction model, and the crime result of the appointed case output by the crime result prediction model is obtained. The criminal result prediction method provided by the embodiment of the application can be based on the non-judgment book of the appointed case, automatically predicts the criminal result by utilizing the pre-established criminal result prediction model, and the predicted criminal result is more accurate, can be used for reference by a legal officer, and has a better auxiliary criminal effect.
In the method for predicting the sentencing result provided in the foregoing embodiment, the training sample of the sentencing result prediction model for predicting the sentencing result is a sentencing element corresponding to a specified sentencing element label extracted from a training decision book with a specified crime name, and the sample label is a decision result element corresponding to a specified decision result element label extracted from the training decision book. The following describes the implementation process of extracting the electrocution element corresponding to the appointed electrocution element label and the judgment result element corresponding to the appointed judgment result element label from the training judgment book.
In one possible implementation, a pre-established decision element extraction model may be utilized to extract the sentency element corresponding to the specified sentency element tag and the decision element corresponding to the specified decision element tag from the unlabeled decision. The judgment book element extraction model is obtained by training a training judgment book marked with a designated criminal element label and a designated judgment result element label.
Because the writing format of the decision book is more standard, when the decision book element extraction model is trained, only a small number of decision books are marked, for example, 500-1000 decision books can be marked, and the marked decision books are used for training the decision book element extraction model. Firstly training a decision element extraction model by using a small number of decision books marked with specified crime element labels and specified decision element labels, and then extracting the crime elements corresponding to the specified crime element labels and the decision element corresponding to the specified decision element labels from unlabeled decision books by using the decision element extraction model obtained by training, wherein the extracted data are used for training a crime result prediction model.
In one possible implementation, the training decision book may be labeled based on a BIOES label system, where B represents a label start word, where the label start word refers to a start word of an element corresponding to a label, I represents a label intermediate word, i.e., an intermediate word of an element corresponding to the label, E represents a label end word, i.e., an end word of an element corresponding to the label, S represents a single label word, i.e., only one element corresponding to the label, and O represents a non-element word. Illustratively, for "whether or not tank" the relevant tags are B_whether or not tank, I_whether or not tank, E_whether or not tank, S_whether or not tank.
Further, referring to fig. 2 and 3, fig. 2 is a flow chart illustrating a process of extracting a sentencing element corresponding to a specified sentencing element tag and a decision result element corresponding to a specified decision result element tag from an unlabeled decision using a pre-established decision element extraction model, and fig. 3 is a schematic diagram illustrating an example of a topology of the decision element extraction model, and the process of extracting elements from an unlabeled decision may include:
step S201: and (3) performing word segmentation on the unlabeled decision book through a semantic vector determination module (301 in fig. 3) in the decision book element extraction model, and determining a semantic vector corresponding to each word obtained through the word segmentation.
In one possible implementation manner, the process of determining the semantic vector corresponding to each word obtained by word segmentation processing may include: for each word, perform: firstly, identifying the part of speech of a word and the named entity of the word; then, converting the word, the part of speech of the word and the named entity of the word into vectors respectively, for example, converting the word, the part of speech of the word and the named entity of the word into vectors through a wordbelting model to obtain vectors corresponding to the word, vectors corresponding to the part of speech of the word and vectors corresponding to the named entity of the word; finally, the vectors corresponding to the words, the vectors corresponding to the parts of speech of the words and the vectors corresponding to the named entities of the words are spliced (three vectors are spliced together through a vector splicing module concat in fig. 3), and the spliced vectors are used as semantic vectors of the words.
In order to enrich the semantics of the semantic vector corresponding to each word, in another possible implementation manner, for each word, in addition to the vector corresponding to the word, the vector corresponding to the part of speech of the word, and the vector corresponding to the named entity of the word, each word in the word needs to be converted into a vector, the vectors corresponding to each word are spliced, then the feature vector is extracted from the spliced vectors through a feature extraction module (such as a convolutional neural network CNN), finally the vector corresponding to the word, the vector corresponding to the part of speech of the word, the vector corresponding to the named entity of the word, and the extracted feature vector are spliced (the four vectors are spliced together through a vector splicing module concat in fig. 3), and the spliced vector is used as the semantic vector of the word.
Step S202: the element tag of the word corresponding to each semantic vector is determined by the element tag determination module (302 in fig. 3) in the decision element extraction model and the semantic vector corresponding to each word.
Specifically, the semantic vector corresponding to each word is input into the element tag determining module, and the element tag of the word corresponding to the input semantic vector output by the element tag determining module is obtained, so that the element tag of each word in the unlabeled judgment book is obtained. In addition, the element label determining module in the decision element extraction model may be, but not limited to, a bidirectional BiLSTM (i.e., bidirectional LSTM).
Step S203: and merging the continuous words with the same element label, wherein the merged content is taken as the element corresponding to the element label.
For example, it is assumed that the element tag corresponding to the word a is "whether or not tank", the element tag corresponding to the word B is "whether or not tank", the element tag corresponding to the word c is "whether or not tank", the element tag corresponding to the word d is "whether or not tank", and the element tags corresponding to the word a, the word B, the word c, and the word d are all "whether or not tank", so that the word a, the word B, the word c, and the word d are combined together as the element tag "whether or not tank".
In another embodiment of the present application, the above embodiment is described in which the electrocution element corresponding to the specified electrocution element tag is obtained from the target non-decision book.
In one possible implementation, the process of acquiring the electrocution element corresponding to the specified electrocution element tag from the target non-decision book may include: and extracting the sentencing element corresponding to the appointed sentencing element label from the target non-judgment book by using a pre-established non-judgment book element extraction model.
The non-decision book element extraction model is obtained by training a training non-decision book serving as a training sample and a labeling result of labeling the training non-decision book based on a designated sentencing element label serving as a sample label.
In one possible implementation, the labeling result for labeling the training non-decision book based on the specified electrocution element label may include: training starting position and starting position score, ending position and ending position score of the sentencing element corresponding to the appointed quantity element label in the non-judgment book.
Considering that a specific electrocution element tag or tags may not appear in the training non-decision book, and accordingly, there will not be electrocution elements corresponding to the specific electrocution element tag or tags in the non-decision book, in another possible implementation manner, the labeling result of labeling the training non-decision book based on the specific electrocution element tag may include: training starting position and starting position scores, ending position and ending position scores of the sentencing elements corresponding to the appointed amount element labels in the non-judgment books, and scoring that the sentencing elements are empty.
Referring to fig. 4, a flow diagram of an implementation process of extracting a electrocuting element corresponding to a specified electrocuting element label from a target electrocuting book by using a model for extracting a electrocuting element obtained by training a exclusionary book based on the second labeling mode is shown, which may include:
step S401: and determining the information of the sentencing element corresponding to the appointed sentencing element label in the target non-judgment book as target sentencing element information through a non-judgment book element extraction model.
The target criminal element information comprises starting positions and starting position scores, ending positions and ending position scores of the criminal elements corresponding to the appointed criminal element labels in the target non-judgment book, and scores that the criminal elements are empty.
In one possible implementation, the non-decision element extraction model may employ a reading understanding model, and the input of the model includes a non-decision and a specified sentency element label, and in a preferred implementation, in order to promote the effect of reading the understanding model, the specified sentency element label may be converted into a problem form based on the semantics of the specified sentency element label, for example, the specified sentency element label is: "whether crime is accepted" it can be converted into: "you do your job, if you are criminal", after converting the specified criminal element label, get the specified question label. It should be noted that, in one possible implementation, for each specified element tag, it may be converted into a question, and the question is used as a tag to train a reading understanding model; to further enhance the effectiveness of the reading understanding model, in another possible implementation, for each specified element tag, it may be converted into a plurality of questions based on its semantics, with which the reading understanding model is trained.
The process of determining the information of the sentencing element corresponding to the appointed sentencing element label in the target non-decision book through the non-decision book element extraction model as the target sentencing element information can comprise: and inputting the target non-decision book and the question label text into a non-decision book element extraction model (reading understanding model) to obtain target answer information output by the non-decision book element extraction model, wherein the target answer information is used as target criminal investigation element information. The question label text is a text containing a specified question label, and the target answer information comprises a starting position and a starting position score, an ending position and an ending position score of an answer corresponding to the specified question label, and a score that the answer is null.
Referring to fig. 5, a schematic diagram of an example of a topological structure of a non-decision element extraction model is shown, input of the non-decision element extraction model shown in fig. 5 is a non-decision and a problem tag text, the Concat 502a and the Concat 502b in fig. 5 are used for splicing input vectors, specifically, the input of the Concat 502a is a vector corresponding to each word obtained by word segmentation of the non-decision element and a vector output by the CNN501a, the input of the CNN501a is a vector obtained by splicing vectors corresponding to each word of the corresponding word, and the vector output by the Concat 502a is used as a target vector corresponding to each word in the non-decision element. Similarly, the input of the Concat 502b is a vector corresponding to each word obtained by word segmentation of the question text and a vector output by the CNN501 b, the input of the CNN501 b is a vector obtained by splicing vectors corresponding to respective words of the corresponding word, and the vector output by the Concat 502b is used as a target vector corresponding to each word in the question text. The method comprises the steps of inputting a target vector corresponding to each word in a non-decision book into BiLSTM (bidirectional LSTM) 503a to obtain a vector corresponding to each word in the non-decision book and having context information, inputting a target vector corresponding to each word in a question text into BiLSTM503b to obtain a vector corresponding to each word in the question text and having context information, inputting a vector corresponding to each word in the non-decision book and having context information into an attribute 504, wherein the vector output by the attribute 504 is a representation of each word in the non-decision book based on the question text, inputting a score of each word in the non-decision book as an answer starting word, a score of an answer ending word and a score of the word not belonging to the answer by the BiLSTM 505.
Step S402: and determining the criminal element corresponding to the appointed criminal element label in the target non-judgment book through the target criminal element information.
It should be noted that, at some time, a case may occur in which a certain specified question label has a plurality of answers, and when this occurs, a plurality of answers corresponding to the specified question label may be processed. Specifically, taking a designated question label with a plurality of answers as a target question label, and removing the answers which do not meet the preset condition from the plurality of answers corresponding to the target question label to obtain residual answers; if the number of the residual answers is multiple, performing de-coincidence processing on the multiple residual answers, and taking the answers obtained after the de-coincidence processing as the sentencing elements corresponding to the appointed sentencing element labels corresponding to the target question labels.
Further, referring to fig. 6, a flowchart illustrating an implementation process of removing answers that do not meet a preset condition from a plurality of answers corresponding to a target question label to obtain remaining answers is shown, where the implementation process may include:
step S601: and determining a preset answer from a plurality of answers corresponding to the target question label to form a first candidate answer set.
Wherein the score of each answer is determined by the starting location score and the ending location score of the answer. Specifically, for each answer, the product of the starting position score and the ending position score corresponding to the answer may be used as the score of the answer.
Wherein, the score of the preset answers is higher than the scores of other answers. In one possible implementation, the answer with the highest score may be screened out first, then the starting position score and the ending position score of the screened out answer are set to 0, then the answer with the highest score is screened out, the starting position score and the ending position score of the screened out answer are set to 0, and accordingly the preset answers are screened out. In another possible implementation, the answers corresponding to the target question label may be ranked in order of high-to-low score, and the top-ranked N answers are combined into the first candidate answer set, assuming that the preset number is N (e.g., 5).
It should be noted that, if the number of answers corresponding to the target question label is greater than the preset number, step S601 is executed, and if the number of answers corresponding to the target question label is less than or equal to the preset number, the answers corresponding to the target question label are directly formed into the first candidate answer set, and then step S402 is executed.
Step S602: and removing the candidate answers with scores lower than a first threshold value in the first candidate answer set to obtain a second candidate answer set.
The first threshold is a score that an answer corresponding to the target question label is null.
Illustratively, the first candidate set includes 5 answers, where the scores of two answers are lower than the score that the answer corresponding to the target question label is null, and the two answers are removed from the first candidate answer set.
It should be noted that, if the score of each candidate answer in the first candidate answer set is lower than the score that the answer corresponding to the target question label is null, the answer corresponding to the target question label is not considered to be in the target non-decision book.
Step S603: and removing the candidate answers with the scores lower than the second threshold value from the second candidate answer set, and obtaining the residual answers.
Wherein the second threshold is set based on the score of the highest scoring candidate answer. For example, the second threshold may be set to 1/10 of the score of the highest scoring candidate answer, i.e., if the score of a certain candidate answer in the second set of candidate answers is lower than 1/10 of the score of the highest scoring candidate answer, that candidate answer is removed from the second set of candidate answers.
After obtaining the residual answers, if there are multiple residual answers, and there may be a situation that the multiple residual answers have overlapping contents, at this time, the multiple residual answers need to be de-overlapped and processed, and the answer obtained after the de-overlapping and combining processing is used as the target answer corresponding to the target question label, that is, the sentencing element corresponding to the designated sentencing element label corresponding to the target question label.
For example, there are a plurality of answers corresponding to the target question label, the answers that do not meet the preset condition are removed from the plurality of answers corresponding to the target question label, the obtained remaining answers include 3, and the 3 answers are respectively: (1) sibling: XXX,37 years old, farmers in home, sister: XXX,35 years old; (2) sister: XXX,35 years old, now home manager; (3) son: XXX,12 years old, after divorce, the wife is cared before coming back, since there is coincidence between the first answer and the second answer: sister: XXX, age 35, then the overlapping content needs to be removed, and the overlapping content is removed to obtain: (1) sibling: XXX,37 years old, in home service farmers; (2) sister: XXX,35 years old, now home manager; (3) son: XXX,12 years old, is cared for by the wife before the marriage.
According to the method for predicting the sentencing results, provided by the embodiment of the application, a small amount of judgment book training judgment book element extraction models marked with the appointed sentencing element labels and the appointed judgment result element labels can be utilized, then the judgment book element extraction models obtained through training are used for extracting sentencing elements corresponding to the appointed sentencing element labels and judgment result elements corresponding to the appointed judgment result element labels from unlabeled judgment books, then the sentencing result prediction models are trained by data extracted from unlabeled judgment books through the judgment book element extraction models, finally target sentencing elements corresponding to the appointed sentencing element labels are extracted from target non-judgment books of the appointed cases, and the target sentencing elements are input into the sentencing result prediction models obtained through training, so that the sentencing results of the appointed cases output by the sentencing result prediction models are obtained. According to the method for predicting the crime result, a good auxiliary crime effect can be achieved based on the internal characteristics of the legal documents and a small amount of manually marked legal documents, and the problems that the existing prediction scheme is too much in manual intervention and poor in auxiliary crime effect are solved.
Corresponding to the above-mentioned method for predicting the sentencing results, the embodiment of the present application further provides a device for predicting the sentencing results, which may include: a non-decision acquisition module 701, a sentencing element determination module 702 and a sentencing result prediction module 703.
The non-decision acquiring module 701 is configured to acquire a non-decision book of a specified case of a specified crime name as a target non-decision book.
The electrocuting element determining module 702 is configured to obtain, from a target non-decision book, an electrocuting element corresponding to a specified electrocuting element tag as a target electrocuting element.
The crime result prediction module 703 is configured to input a target crime element into a pre-established crime result prediction model, and obtain a crime result of a specific case output by the crime result prediction model.
The crime result prediction model is obtained by taking criminal elements which are extracted from a training judgment book with appointed criminal names and correspond to appointed criminal element labels as training samples and taking judgment result elements which are extracted from the training judgment book and correspond to the appointed judgment result element labels as sample labels.
According to the crime result prediction device provided by the embodiment of the application, firstly, the target non-judgment book of the appointed case with the appointed crime name is obtained, then, the target crime element corresponding to the appointed crime element label is obtained from the target non-judgment book, and finally, the target crime element is input into the pre-established crime result prediction model, so that the crime result of the appointed case output by the crime result prediction model is obtained. The crime result prediction device provided by the embodiment of the application can automatically predict the crime result by utilizing the pre-established crime result prediction model based on the non-judgment book of the appointed case, and the predicted crime result is more accurate, can be used for reference by a legal officer, and has a better auxiliary crime effect.
The crime result prediction device provided in the above embodiment may further include: and the judgment book element extraction module.
And the judgment book element extraction module is used for extracting the sentencing element corresponding to the appointed sentencing element label and the judgment result element corresponding to the appointed judgment result element label from the unlabeled judgment book by utilizing a pre-established judgment book element extraction model.
The judgment book element extraction model is obtained by training a training judgment book marked with a designated criminal element label and a designated judgment result element label.
Further, the above-mentioned decision book element extraction module is specifically configured to perform word segmentation on the unlabeled decision book by using a semantic vector determination module in the decision book element extraction model, so as to determine a semantic vector corresponding to each word obtained by the word segmentation; determining element labels of words corresponding to each semantic vector through an element label determining module in the judgment book element extraction model and the semantic vector corresponding to each word; and merging the continuous words with the same element label, wherein the merged content is taken as the element corresponding to the element label.
In a possible implementation manner, the electrocution element determining module 702 provided in the foregoing embodiment is specifically configured to extract, from the target non-decision book, an electrocution element corresponding to the specified electrocution element label by using a pre-established non-decision book element extraction model.
The non-decision book element extraction model is obtained by training a training non-decision book as a training sample and training a labeling result of labeling the training non-decision book based on the appointed sentencing element label as a sample label.
In one possible implementation, the labeling result of labeling the non-decision book based on the specified criminal element label includes: and the starting position score, the ending position and the ending position score of the sentencing element corresponding to the appointed amount element label in the training non-judgment book, and the score that the sentencing element is empty.
The sentencing element determination module 702 includes: the criminal element information determination submodule and the criminal element determination submodule.
And the electrocuting element information determining submodule is used for determining information of electrocuting elements corresponding to the appointed electrocuting element labels in the target non-decision book as target electrocuting element information through the non-decision book element extraction model.
The target criminal element information includes: starting position and starting position scores, ending position and ending position scores of the sentencing elements corresponding to the appointed sentencing element labels in the target non-judgment book, and scores that the sentencing elements are empty.
And the criminal element determination submodule is used for determining the criminal element corresponding to the appointed criminal element label in the target non-decision book through the target criminal element information.
Further, the electrocuting element information determining submodule is specifically configured to input the target non-decision book and the question label text into the non-decision book element extraction model, and obtain target answer information output by the non-decision book element extraction model as the target electrocuting element information.
The text of the question label is a text containing a specified question label, and the specified question label is a label obtained by converting the specified criminal element label into a question form; the target answer information includes a starting position and a starting position score, an ending position and an ending position score of an answer corresponding to the specified question label, and a score that the answer is null.
Further, the electrocuting element determining submodule is specifically configured to remove an answer that does not meet a preset condition from a plurality of answers corresponding to the target question label if the target question label with the plurality of answers exists in the specified question label, so as to obtain a remaining answer; if the number of the residual answers is multiple, performing de-coincidence processing on the multiple residual answers, and taking the answers obtained after de-coincidence processing as the target criminal investigation elements.
Further, the criminal element determining submodule is specifically configured to determine that a preset number of answers form a first candidate answer set from a plurality of answers corresponding to the target question label when the answers which do not meet a preset condition are removed from the plurality of answers corresponding to the target question label to obtain remaining answers, wherein the score of the preset number of answers is higher than the scores of other answers, and the score of each answer is determined by the starting position score and the ending position score of the answer; removing candidate answers with scores lower than a first threshold value in the first candidate answer set to obtain a second candidate answer set, wherein the first threshold value is a score that an answer corresponding to the target question label is empty; and removing the non-highest scoring candidate answers with scores lower than a second threshold value in the second candidate answer set to obtain the residual answers, wherein the second threshold value is set based on the scores of the highest scoring candidate answers.
And the criminal element determination submodule is further used for determining that the answer corresponding to the target question label is empty when the score of each candidate answer in the first candidate answer set is lower than the first threshold value.
The embodiment of the application also provides a crime result prediction device, referring to fig. 7, which shows a schematic structural diagram of the crime result prediction device, and the crime result prediction device may include: a memory 801 and a processor 802.
A memory 801 for storing a program;
a processor 802, configured to execute the program, where the program is specifically configured to:
acquiring a non-judgment book of a designated crime name as a target non-judgment book;
acquiring a specified criminal element corresponding to the criminal element label from the target non-judgment book, and taking the specified criminal element label as a target criminal element;
inputting the target criminal investigation element into a pre-established criminal investigation result prediction model to obtain a criminal investigation result output by the criminal investigation result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
The sentencing result prediction apparatus may further include: a bus, a communication interface 803, an input device 804, and an output device 805.
The processor 802, the memory 801, the communication interface 803, the input device 804, and the output device 805 are connected to each other by a bus. Wherein:
A bus may comprise a path that communicates information between components of a computer system.
The processor 1102 may be a general-purpose processor such as a general-purpose Central Processing Unit (CPU), microprocessor, etc., or may be an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of programs in accordance with aspects of the present invention. But may also be a Digital Signal Processor (DSP), application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components.
The processor 802 may include a main processor, and may also include a baseband chip, a modem, and the like.
The memory 801 stores programs for implementing the technical scheme of the present invention, and may also store an operating system and other critical services. In particular, the program may include program code including computer-operating instructions. More specifically, the memory 801 may include read-only memory (ROM), other types of static storage devices that may store static information and instructions, random access memory (random access memory, RAM), other types of dynamic storage devices that may store information and instructions, disk storage, flash, and the like.
The input device 804 may include means for receiving data and information entered by a user, such as a camera, light pen, touch screen, etc.
The output device 805 may include means, such as a display screen, speakers, etc., that allow information to be output to a user.
The communication interface 803 may include devices using any transceiver or the like to communicate with other devices or communication networks, such as ethernet, radio Access Network (RAN), wireless Local Area Network (WLAN), etc.
The processor 802 executes programs stored in the memory 801 and invokes other devices that may be used to implement the various steps of the electrocution result prediction method provided by embodiments of the present application.
The embodiment of the application also provides a readable storage medium, on which a computer program is stored, which when being executed by a processor, implements the steps of the electrocuting result prediction method provided by the above embodiment.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other.
In the several embodiments provided in the present application, it should be understood that the disclosed method, apparatus and device may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be through some communication interface, device or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment. In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (11)
1. A method of predicting a sentencing result, comprising:
acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book;
acquiring a criminal element corresponding to a designated criminal element label from the target non-judgment book by using a pre-established non-judgment book element extraction model as a target criminal element; the non-decision book element extraction model takes a training non-decision book as a training sample, and takes a starting position and a starting position score, an ending position and an ending position score of a sentencing element corresponding to the appointed quantity element label in the training non-decision book, and a score of the sentencing element being empty as a sample label for training;
Inputting the target criminal investigation element into a pre-established criminal investigation result prediction model to obtain the criminal investigation result of the specified case output by the criminal investigation result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
2. The method for predicting the sentencing results according to claim 1, wherein the process of extracting the sentencing element corresponding to the designated sentencing element tag and the judgment result element corresponding to the designated judgment result element tag from the training judgment book of the designated crime name comprises:
extracting the sentencing element corresponding to the appointed sentencing element label and the judgment result element corresponding to the appointed judgment result element label from the unlabeled judgment book by using a pre-established judgment book element extraction model;
the judgment book element extraction model is obtained by training a training judgment book marked with a designated criminal element label and a designated judgment result element label.
3. The method for predicting the sentencing results according to claim 2, wherein the extracting the sentencing element corresponding to the specified sentencing element label and the decision result element corresponding to the specified decision result element label from the unlabeled decision book by using a pre-established decision element extraction model includes:
the semantic vector determining module in the judgment book element extraction model is used for carrying out word segmentation on the unlabeled judgment books, and determining semantic vectors corresponding to each word obtained through word segmentation;
determining element labels of words corresponding to each semantic vector through an element label determining module in the judgment book element extraction model and the semantic vector corresponding to each word;
and merging the continuous words with the same element label, wherein the merged content is taken as the element corresponding to the element label.
4. The method for predicting the sentencing results according to claim 1, wherein the extracting the sentencing element corresponding to the specified sentencing element tag from the target non-decision book using a pre-established non-decision book element extraction model includes:
determining information of the sentencing element corresponding to the appointed sentencing element label in the target non-judgment book as target sentencing element information through the non-judgment book element extraction model, wherein the target sentencing element information comprises a starting position score, an ending position score and an ending position score of the sentencing element corresponding to the appointed sentencing element label in the target non-judgment book, and a score that the sentencing element is empty;
And determining the criminal element corresponding to the appointed criminal element label in the target non-judgment book through the target criminal element information.
5. The method for predicting the sentencing results according to claim 4, wherein the determining, by the non-decision element extraction model, information of the sentencing element corresponding to the specified sentencing element tag in the target non-decision book as target sentencing element information includes:
inputting the target non-decision book and the question label text into the non-decision book element extraction model to obtain target answer information output by the non-decision book element extraction model as the target sentencing element information;
the text of the question label is a text containing a specified question label, and the specified question label is a label obtained by converting the specified criminal element label into a question form; the target answer information includes a starting position and a starting position score, an ending position and an ending position score of an answer corresponding to the specified question label, and a score that the answer is null.
6. The method for predicting the sentencing results according to claim 5, wherein the determining, by the target sentencing element information, the sentencing element corresponding to the specified sentencing element tag in the target non-decision book includes:
If the target question label with a plurality of answers exists in the specified question label, removing the answers which do not meet the preset condition from the plurality of answers corresponding to the target question label, and obtaining the rest answers;
if the number of the residual answers is multiple, performing de-coincidence processing on the multiple residual answers, and taking the answers obtained after de-coincidence processing as the target criminal investigation elements.
7. The method for predicting the sentencing results according to claim 5, wherein the step of removing the answers which do not meet the preset condition from the plurality of answers corresponding to the target question label to obtain the remaining answers includes:
determining a preset number of answers from a plurality of answers corresponding to the target question label to form a first candidate answer set, wherein the scores of the preset number of answers are higher than the scores of other answers, and the score of each answer is determined through the starting position score and the ending position score of the answer;
removing candidate answers with scores lower than a first threshold value in the first candidate answer set to obtain a second candidate answer set, wherein the first threshold value is a score that an answer corresponding to the target question label is empty;
And removing the non-highest scoring candidate answers with scores lower than a second threshold value in the second candidate answer set to obtain the residual answers, wherein the second threshold value is set based on the scores of the highest scoring candidate answers.
8. The method for predicting the sentencing results according to claim 7, wherein the filtering the answers which do not meet the preset condition from the plurality of answers corresponding to the target question label to obtain the remaining answers further comprises:
and if the scores of the candidate answers in the first candidate answer set are lower than the first threshold value, determining that the answer corresponding to the target question label is empty.
9. A sentencing result prediction device, comprising: the system comprises a non-decision acquisition module, a criminal element determination module and a criminal result prediction module;
the non-judgment book acquisition module is used for acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book;
the sentencing element determining module is used for acquiring the sentencing element corresponding to the appointed sentencing element label from the target non-decision book by utilizing a pre-established non-decision book element extraction model, and taking the sentencing element as a target sentencing element; the non-decision book element extraction model takes a training non-decision book as a training sample, and takes a starting position and a starting position score, an ending position and an ending position score of a sentencing element corresponding to the appointed quantity element label in the training non-decision book, and a score of the sentencing element being empty as a sample label for training;
The crime result prediction module is used for inputting the target crime element into a pre-established crime result prediction model to obtain the crime result of the specified case output by the crime result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
10. A device for predicting a sentencing result, comprising: a memory and a processor;
the memory is used for storing programs;
the processor is configured to execute the program, where the program is specifically configured to:
acquiring a non-judgment book of a designated case of a designated crime name as a target non-judgment book;
acquiring a criminal element corresponding to a designated criminal element label from the target non-judgment book by using a pre-established non-judgment book element extraction model as a target criminal element; the non-decision book element extraction model takes a training non-decision book as a training sample, and takes a starting position and a starting position score, an ending position and an ending position score of a sentencing element corresponding to the appointed quantity element label in the training non-decision book, and a score of the sentencing element being empty as a sample label for training;
Inputting the target criminal investigation element into a pre-established criminal investigation result prediction model to obtain the criminal investigation result of the specified case output by the criminal investigation result prediction model;
the criminal result prediction model is obtained by training a criminal element which is extracted from a training judgment book with a specified criminal name and corresponds to a specified criminal element label as a training sample, and a judgment result element which is extracted from the training judgment book and corresponds to the specified judgment result element label as a sample label.
11. A readable storage medium having stored thereon a computer program, which when executed by a processor, implements the steps of the electrocution result prediction method according to any of claims 1 to 8.
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 CN109241528A (en) | 2019-01-18 |
CN109241528B true 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) |
Families Citing this family (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 |
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 |
CN110738039B (en) * | 2019-09-03 | 2023-04-07 | 平安科技(深圳)有限公司 | Case auxiliary information prompting method and device, storage medium and server |
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 |
CN112579732A (en) * | 2019-09-30 | 2021-03-30 | 北京国双科技有限公司 | Sentencing prediction method and device |
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 |
CN111325387B (en) * | 2020-02-13 | 2023-08-18 | 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 |
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 (1)
Title |
---|
审判案例自动抽取与标注模型研究;佘贵清等;《现代图书情报技术》;20130625(第06期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN109241528A (en) | 2019-01-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109241528B (en) | Criminal investigation result prediction method, device, equipment and storage medium | |
CN108399228B (en) | Article classification method and device, computer equipment and storage medium | |
CN111078837B (en) | Intelligent question-answering information processing method, electronic equipment and computer readable storage medium | |
CN112667794A (en) | Intelligent question-answer matching method and system based on twin network BERT model | |
CN110334217B (en) | Element extraction method, device, equipment and storage medium | |
CN110399473B (en) | Method and device for determining answers to user questions | |
CN112270188A (en) | Questioning type analysis path recommendation method, system and storage medium | |
CN117520503A (en) | Financial customer service dialogue generation method, device, equipment and medium based on LLM model | |
CN110969005B (en) | Method and device for determining similarity between entity corpora | |
CN114491079A (en) | Knowledge graph construction and query method, device, equipment and medium | |
CN113051384B (en) | User portrait extraction method based on dialogue and related device | |
CN110765276A (en) | Entity alignment method and device in knowledge graph | |
CN114037007A (en) | Data set construction method and device, computer equipment and storage medium | |
CN112712056A (en) | Video semantic analysis method and device, storage medium and electronic equipment | |
US11475529B2 (en) | Systems and methods for identifying and linking events in structured proceedings | |
CN116090450A (en) | Text processing method and computing device | |
CN115358817A (en) | Intelligent product recommendation method, device, equipment and medium based on social data | |
CN114398482A (en) | Dictionary construction method and device, electronic equipment and storage medium | |
CN114840642A (en) | Event extraction method, device, equipment and storage medium | |
CN111597453B (en) | User image drawing method, device, computer equipment and computer readable storage medium | |
CN114529191A (en) | Method and apparatus for risk identification | |
CN113935326A (en) | Knowledge extraction method, device, equipment and storage medium | |
CN113704422A (en) | Text recommendation method and device, computer equipment and storage medium | |
CN111309773A (en) | Vehicle information query method, device and system and storage medium | |
CN111382247A (en) | Content pushing optimization method, content pushing optimization device and electronic equipment |
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 |