CN115878815B - Legal document judgment result prediction method, legal document judgment result prediction device and storage medium - Google Patents

Abstract

The invention relates to the technical field of data processing, in particular to a method, a device and a storage medium for predicting a judgment result of a legal document, wherein the method comprises the following steps: acquiring legal judgment documents of massive legal judgment examples; carrying out legal element extraction and relation determination between legal elements on each legal judgment document to obtain a judgment event chain and element relation from original complaint type, judgment reason to judgment result; according to the relation between the decision event chains and the elements corresponding to each legal decision document, constructing a knowledge map of the legal decision document by taking a legal decision example as the center; receiving an input legal fact, wherein the legal fact comprises a legal appeal; and predicting and outputting legal judgment results corresponding to the legal facts by using the knowledge graph and the legal retrieval model. According to the technical scheme, complex redundancy work of aligning and mapping elements of complex redundancy to an inference function is omitted, and good effects are achieved on document retrieval based on excellent feature table answers.

Description

Legal document judgment result prediction method, legal document judgment result prediction device and storage medium

Technical Field

The present invention relates to the field of data processing technologies, and in particular, to a method and apparatus for predicting a decision result of a legal document, and a storage medium.

Background

In recent years, big data and artificial intelligence algorithms are increasingly gaining high importance and promotion in various industries and fields. Many countries, including China, are raising artificial intelligence to national strategic levels. In the judicial field, the promotion of big data technology benefits, and every level of judicial institutions in China enter a construction period of 'intelligent court' with the provision of intelligent judicial services as a core.

However, the methods proposed for specific tasks in the current intelligent judicial research field still have fundamental impediments to their practical application. One is that the mainstream method at the present stage is mostly based on models such as machine learning, neural networks and the like, and the defects of black boxes of the models cause the general lack of interpretability of research processes and results, so that the credibility and usability of the models are greatly reduced. Secondly, models relying on large-scale data training generally lack an inference mechanism, and machine intelligence generally refers to that the intelligent agent can learn, perceive, understand and work like a human, wherein understanding human cognition is one of the essential conditions for realizing the intelligence, knowledge reasoning is an important means for human cognition, and most of methods based on statistical models today cannot utilize knowledge reasoning to obtain results, namely the models do not have an inference mechanism.

Judicial decision reasoning is a method for obtaining decisions of court trial cases and is also an important means for proving judicial decision validity, so that the judicial decision reasoning is not only a legal thinking method, but also a practical rationality or practical reasoning process for solving the problems of the judges. Theoretically, judicial decisions should be the logical result of judicial reasoning. In the law society, any case judgment should provide a certain reason or basis, and judicial reasoning can provide validity proof for judicial judgment, because the primary role of legal reasoning is to provide validity reason for conclusion, and meanwhile, a strict logic judicial reasoning forms strong reason or basis.

The existing reasoning logic mainly has the following defects:

1) There are a number of legal elements that generally need to be aligned to standard categories, for example, child career decisions in divorce decisions 'child less than 2 years' need to be aligned to a number of reasons such as "child |age| less than 2 years", etc., which is not an easy matter of knowledge architecture construction and batch reasoning if conditions are missing.

2) Generally, the existing reasoning mode is to construct a decision function, basically based on rule reasoning, and does not have the diversity, fault tolerance and early warning functions of big data.

3) The existing element extraction analysis model has overlarge dependence on large-batch data, and has no reliable robustness and small sample learning capability.

Disclosure of Invention

In order to overcome the problems in the related art, the invention provides a method, a device and a storage medium for predicting the judgment result of a legal document, thereby reducing the labor cost for acquiring data, improving the field suitability and better adapting to the application scene of fine granularity.

According to a first aspect of an embodiment of the present invention, there is provided a method for predicting a decision result of a legal document, the method including:

acquiring legal judgment documents of massive legal judgment examples;

carrying out legal element extraction and relation determination among legal elements on each legal judgment document to obtain a judgment event chain from original complaint type, judgment reason to judgment result and element relation, wherein the legal elements comprise original complaint type, judgment reason and judgment result, and the element relation comprises causal relation and corresponding relation;

according to the relation between the decision event chain and the element corresponding to each legal decision document, constructing a knowledge graph of the legal decision document by taking a legal decision example as a center, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

Receiving an input legal fact, wherein the legal fact comprises a legal appeal;

and predicting and outputting legal judgment results corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

In one embodiment, preferably, the performing, for each legal decision document, the extracting of legal elements and the determining of the relationship between legal elements includes:

performing legal element recognition on each legal decision document by using an element category classification model to obtain an element category classification recognition result;

and splicing the element category classification recognition result and the characteristics of the legal judgment document, adding case-by-case category information, inputting the case-by-case information into a reading understanding model, enabling the reading understanding model to determine the content and the position of legal elements according to the element category classification recognition result, and outputting a judgment event chain and element relation from original notice requirement type, judgment reason to judgment result.

In one embodiment, preferably, the method further comprises:

and constructing an index set according to the level from top to bottom of the knowledge graph by taking Milvus as a characteristic memory, coding by using a transform law coding model to obtain law judgment document characteristics, document vector characteristics, a judgment event chain and event chain vectors, and storing the law judgment document characteristics, the judgment event chain and the event chain vectors in correspondence with a law judgment document ID.

In one embodiment, preferably, predicting and outputting the legal decision result corresponding to the legal facts by using the knowledge graph and the legal retrieval model includes:

performing intention analysis on the legal facts to determine target case category and target appeal category corresponding to the legal facts;

searching in the knowledge graph according to the target case by category and the target appeal category to determine at least one corresponding legal judgment instance;

performing feature coding on the legal facts and the judgment reasons of at least one corresponding legal judgment example by using a transformer legal coding model to obtain corresponding legal fact feature vectors and judgment reason feature vectors of the legal judgment example;

and calculating the similarity of the legal fact feature vector and the judgment reason feature vector of the legal judgment example, and determining a legal judgment result corresponding to the legal fact according to the similarity.

In one embodiment, preferably, the method further comprises:

receiving an input search command for target legal facts;

according to the search command, performing feature coding on the target legal facts by using a transformation law coding model to obtain coded features;

And carrying out similarity matching on the coded features and the stored legal judgment document features and judgment event chains so as to retrieve the judgment document corresponding to the target legal facts.

In one embodiment, preferably, the method further comprises:

receiving an input similar document retrieval command;

determining a decision event chain corresponding to the current legal document according to the similar document retrieval command;

performing feature coding on a judgment event chain of the current legal document by using a transformer legal coding model to obtain coded event chain features;

and carrying out similarity matching on the encoded event chain features, the stored legal decision document features and the decision event chain so as to retrieve the legal decision document similar to the current legal document.

In one embodiment, preferably, the method further comprises:

and learning and training the transducer model according to legal judgment documents of massive legal judgment examples to obtain the transducer legal coding model.

According to a second aspect of the embodiments of the present invention, there is provided a device for predicting a decision result of a legal document, the device comprising:

the acquisition module is used for acquiring legal judgment documents of massive legal judgment examples;

The first determining module is used for extracting legal elements and determining relations among the legal elements for each legal decision document so as to obtain a decision event chain from an original notice type, a decision reason to a decision result and element relations, wherein the legal elements comprise the original notice type, the decision reason and the decision result, and the element relations comprise causal relations and corresponding relations;

the construction module is used for constructing a knowledge graph of the legal decision document by taking the legal decision instance as a center according to the decision event chain and element relation corresponding to each legal decision document, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

a receiving module for receiving an input legal fact, wherein the legal fact comprises a legal appeal;

and the prediction module is used for predicting and outputting a legal judgment result corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

In one embodiment, preferably, the first determining module is configured to:

performing legal element recognition on each legal decision document by using an element category classification model to obtain an element category classification recognition result;

And splicing the element category classification recognition result and the characteristics of the legal judgment document, adding case-by-case category information, inputting the case-by-case information into a reading understanding model, enabling the reading understanding model to determine the content and the position of legal elements according to the element category classification recognition result, and outputting a judgment event chain and element relation from original notice requirement type, judgment reason to judgment result.

In one embodiment, preferably, the apparatus further comprises:

and the storage module is used for taking Milvus as a characteristic storage, constructing an index set according to the level from top to bottom of the knowledge graph, obtaining legal judgment document characteristics, document vector characteristics, judgment event chains and event chain vectors by utilizing a transform legal coding model code, and storing the legal judgment document characteristics, the document vector characteristics, the judgment event chains and the event chain vectors in correspondence with the legal judgment document ID.

In one embodiment, preferably, the prediction module is configured to:

performing intention analysis on the legal facts to determine target case category and target appeal category corresponding to the legal facts;

searching in the knowledge graph according to the target case by category and the target appeal category to determine at least one corresponding legal judgment instance;

Performing feature coding on the legal facts and the judgment reasons of at least one corresponding legal judgment example by using a transformer legal coding model to obtain corresponding legal fact feature vectors and judgment reason feature vectors of the legal judgment example;

and calculating the similarity of the legal fact feature vector and the judgment reason feature vector of the legal judgment example, and determining a legal judgment result corresponding to the legal fact according to the similarity.

In one embodiment, preferably, the apparatus further comprises:

the first retrieval module is used for receiving an input retrieval command for the target legal facts;

the first coding module is used for carrying out feature coding on the target legal facts by utilizing a transformation law coding model according to the search command to obtain coded features;

and the first matching module is used for matching the similarity between the coded features and the stored legal judgment document features and judgment event chains so as to retrieve the judgment document corresponding to the target legal facts.

In one embodiment, preferably, the apparatus further comprises:

the second retrieval module is used for receiving an input similar document retrieval command;

The second determining module is used for determining a decision event chain corresponding to the current legal document according to the similar document retrieval command;

the second coding module is used for carrying out feature coding on the judgment event chain of the current legal document by using a transformation law coding model to obtain the coded event chain features;

and the second matching module is used for matching the similarity between the encoded event chain features and the stored legal judgment document features and judgment event chains so as to retrieve the legal judgment document similar to the current legal document.

In one embodiment, preferably, the apparatus further comprises:

and the training module is used for learning and training the transducer model according to legal judgment documents of massive legal judgment examples so as to obtain the transducer legal coding model.

According to a third aspect of the embodiments of the present invention, there is provided a device for predicting a decision result of a legal document, the device comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

acquiring legal judgment documents of massive legal judgment examples;

carrying out legal element extraction and relation determination among legal elements on each legal judgment document to obtain a judgment event chain from original complaint type, judgment reason to judgment result and element relation, wherein the legal elements comprise original complaint type, judgment reason and judgment result, and the element relation comprises causal relation and corresponding relation;

According to the relation between the decision event chain and the element corresponding to each legal decision document, constructing a knowledge graph of the legal decision document by taking a legal decision example as a center, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

receiving an input legal fact, wherein the legal fact comprises a legal appeal;

and predicting and outputting legal judgment results corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

According to a fourth aspect of embodiments of the present invention, there is provided a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method as in any of the embodiments of the second aspect.

The technical scheme provided by the embodiment of the invention can comprise the following beneficial effects:

in the embodiment of the invention, a chain structure causal relation is constructed, deeper analysis and table answer are constructed and one-to-one association is constructed from the case to the requirement category to the judgment reason to the judgment result, a legal judgment knowledge map is constructed, multiple rounds of reasoning functions are simplified into semantic understanding tasks through customizing the table answer of a semantic model, complex redundancy work of mapping complex redundancy elements to the reasoning functions in an aligned manner is omitted, and a good effect is obtained on document retrieval based on excellent feature table answer.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention as claimed.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.

Fig. 1 is a flow chart illustrating a method of predicting a decision outcome of a legal document in accordance with an exemplary embodiment.

Fig. 2 is a schematic diagram illustrating the composition of a legal decision document according to an exemplary embodiment.

Fig. 3 is a schematic diagram of a decision event chain shown in accordance with an exemplary embodiment.

Fig. 4 is a schematic diagram of a knowledge graph, shown according to an exemplary embodiment.

Fig. 5 is a flowchart illustrating step S102 in a method for predicting a decision outcome of a legal document according to an exemplary embodiment.

FIG. 6A is a schematic diagram of a specific model structure and process of an element category classification model and a reading understanding model, according to an example embodiment.

FIG. 6B is a schematic diagram illustrating tag naming according to an exemplary embodiment.

FIG. 7 is a feature storage schematic diagram shown in accordance with an exemplary embodiment.

Fig. 8 is a flowchart illustrating step S105 in a method for predicting a decision result of a legal document according to an exemplary embodiment.

Fig. 9 is a flow chart illustrating another method of predicting a decision outcome of a legal document in accordance with an exemplary embodiment.

Fig. 10 is a flowchart illustrating a decision outcome prediction method of yet another legal document according to an exemplary embodiment.

Fig. 11 is a block diagram illustrating a decision result prediction apparatus of a legal document according to an exemplary embodiment.

Detailed Description

Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, the same numbers in different drawings refer to the same or similar elements, unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the invention. Rather, they are merely examples of apparatus and methods consistent with aspects of the invention as detailed in the accompanying claims.

Fig. 1 is a flow chart illustrating a method of predicting a decision outcome of a legal document in accordance with an exemplary embodiment.

As shown in fig. 1, according to a first aspect of the embodiment of the present invention, there is provided a method for predicting a decision result of a legal document, the method including:

step S101, obtaining legal judgment documents of massive legal judgment examples;

The legal decision document mainly has principal information, a trial process, a prosecution title, a warned title, a court finding, 6 main parts considered by the court and three view modeling (a prosecution view, a warned view, a court view), as shown in fig. 2, wherein a legal instrument represents the whole decision document, party information represents principal information of the whole decision document, a three process represents the trial process of the whole decision document, a plaintiff claim represents the prosecution title of the whole decision document, defendant defense represents the warned title of the whole decision document, court investigation represents the court finding of the whole decision document, and a curt belies represents the court of the whole decision document.

There are also main views, where playfield view represents the view of the original notice, dependent view represents the view of the notice, and court view represents the court view.

Step S102, carrying out legal element extraction and relationship determination among legal elements on each legal decision document to obtain a decision event chain from original notice requirement type, decision reason to decision result and element relationship, wherein the legal elements comprise original notice requirement type, decision reason and decision result, and the element relationship comprises causal relationship and corresponding relationship;

Since the invention mainly builds causal relation of decisions, namely, court theory, the court view part is mainly refined here, wherein the judgment can be modeled as original complaint type (judgment type), judgment reason (condition) and judgment result:

the content of the paragraph is considered by the court, the paragraph is resolved and split into a plurality of elements, wherein the class is the original complaint type, the reason represents the judgment reason (condition), and the res represents the judgment result.

The decision causal relation extraction of legal documents can be regarded as element structured extraction of the court theory part, and the element extraction and causal relation correspondence are mainly carried out for the court theory part with paragraphs or spreads, wherein the decision causal relation extraction comprises the following steps: the correspondence relationship of the original complaint type (class), the decision reason (reason), and the decision result (res) is shown in fig. 3, and a relationship of (1, n, 1) one-to-many 1 is shown. Finally, < class, r1, coast > (where r1 represents the correspondence), and < coast, r2, res > (where r2 represents the causal relationship) are represented in a chain structure.

The conventional method for relation extraction is generally defined as two tasks, namely text extraction and relation prediction (graph neural network), however, due to the problem of data cost in the real generation environment, for various reasons such as sparse data of decision types, data quantity of the overall energy supply model, complexity of relation labeling and the like, the learning efficiency of the conventional relation analysis model cannot reach the expected level. Therefore, the invention provides a method based on data association and progressive relation extraction, which is used for analyzing a chain structure from a appeal type to a judgment reason (condition) and finally to a judgment result.

Step S103, constructing a knowledge graph of the legal decision document by taking the legal decision instance as a center according to the decision event chain and element relation corresponding to each legal decision document, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

the knowledge graph is shown in fig. 4, and the root node is the core node of the whole graph; the subordinate structure is case category; the case is represented by a category lower level as a appeal category, including the core common appeal of the case, and is illustrated here by taking the divorce decision as an example: divorce decision- > tending right decision |divorce property decision. A many-to-one or one-to-one relationship of decision reason meason and decision result res is built centered on each decision instance example, with each sub-graph representing one instance of the appeal class. The causal relation of the decisions is continuously accumulated to meet the reasoning requirement of the decisions.

Step S104, receiving an input legal fact, wherein the legal fact comprises legal requirements;

step S105, predicting and outputting a legal judgment result corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

In the embodiment of the invention, a chain structure causal relation is constructed, deeper analysis and table answer are constructed and one-to-one association is constructed from the case to the requirement category to the judgment reason to the judgment result, a legal judgment knowledge map is constructed, multiple rounds of reasoning functions are simplified into semantic understanding tasks through customizing the table answer of a semantic model, complex redundancy work of mapping complex redundancy elements to the reasoning functions in an aligned manner is omitted, and a good effect is obtained on document retrieval based on excellent feature table answer.

Fig. 5 is a flowchart illustrating step S102 in a method for predicting a decision outcome of a legal document according to an exemplary embodiment.

As shown in fig. 5, in one embodiment, preferably, step S102 includes:

step S501, performing legal element recognition on each legal decision document by using an element class classification model to obtain an element class classification recognition result;

step S502, the element category classification recognition result and the characteristics of the legal judgment document are spliced, and the case-by-case category information is added to be input into a reading understanding model, so that the reading understanding model determines the content and the position of legal elements according to the element category classification recognition result, and a judgment event chain and element relation from original notice requirement type, judgment reason to judgment result are output.

The key problem of small information amount of a small sample data set is overcome through high-efficiency learning efficiency of classification prediction. Then, the result of the classification prediction part can clearly represent element category information existing in the sentence, and the reading understanding model (mrc) reduces the learning cost of the reading understanding model (mrc) based on a mechanism of Prompt learning and greatly improves the efficiency of the reading understanding model (mrc) during prediction.

The specific model structure and the processing procedure of the element category classification model and the reading and understanding model are shown in fig. 6A, and the key problem of less information content of the small sample data set is overcome through the efficient learning efficiency of classification prediction. Then, the result of the classification prediction part can clearly represent element category information existing in the sentence, and the reading understanding model (mrc) reduces the learning cost of the reading understanding model (mrc) based on a mechanism of Prompt learning and greatly improves the efficiency of the reading understanding model (mrc) during prediction. Where input (sentence) represents sentence-level text input; the classer represents an element category classification model contained in the sentence; mrc input represents that the read and understand input concatenation contains the result res1query of the classifier as the overall feature of mrc model of the text sentence of the read and understand question concatenation input (sentence); mrc a reading understanding model; output represents the model output including the location categories in the overall chain structure, such as {..} complaint types {..} cause of decision, {..} result of decision and specific element information; c represents classification of the case to provide preconditions for model learning so as to improve learning efficiency; element category task the element category data included in the phrase; mrc task means reading understanding task for extracting detailed elements.

Specifically, as shown in fig. 6A, the element category classification model includes:

input (sentence) layer: the input layer is an original text input by the model, the legal decision text is segmented into single small sentences according to symbols, and then the small sentences in one section are input into the model in batches for analysis of the model. The model learning efficiency is improved by the category (C).

A classer layer: the element category classifier mainly analyzes element categories contained in sentences, a feature representation layer is also called an embedding layer, and the neural network is utilized to represent the categories of the input texts. Typical characteristic representation means are convolutional neural networks, long and short memory machines and a transfomer series model, and the invention adopts a lawformer model.

As shown in fig. 6B, element categories may be labeled in the form of labels. And then, the relationship labels are converted into a chain structure in a label naming mode, so that the low-efficiency relationship learning model is conveniently converted into an efficient classification learning model.

The reading and understanding model comprises:

src input layer: reading and understanding input mainly comprises a result of three-party information classification model identification, an original legal text and a prompting case-by-case type (C), splicing the result of information classification model identification and the original legal text characteristics, and then adding the information of the case-by-case type (C) after splicing the characteristics.

mrc layer: the details and positions of the elements are analyzed mainly according to the progressive information. The main technical means are named identification (ner) and fragment classification (span_class), and the invention adopts a reading and understanding model (mrc) to achieve the control purpose of independently extracting a certain class of elements, greatly improve the efficiency and reduce the calculated amount.

output layer: outputting the model results to form a chain structure comprising: {..} Requirements type {..} cause of decision {.} decision results and specific element information.

In one embodiment, preferably, the method further comprises:

as shown in fig. 7, the Milvus is used as a feature memory, an index set is constructed according to the level from top to bottom of the knowledge graph, legal judgment document features, document vector features, judgment event chains and event chain vectors are obtained by using a transform legal coding model for coding, and the legal judgment document features, the document vector features, the judgment event chains and the event chain vectors are stored correspondingly with legal judgment document IDs.

As shown in fig. 8, in one embodiment, preferably, step S105 includes:

step S801, carrying out intention analysis on the legal facts to determine a target case category and a target appeal category corresponding to the legal facts;

step S802, searching in the knowledge graph according to the target case by category and the target appeal category to determine at least one corresponding legal judgment instance;

Step S803, performing feature coding on the legal facts and the decision reason of the at least one corresponding legal decision instance by using a transform law coding model to obtain corresponding legal fact feature vectors and decision reason feature vectors of the legal decision instance;

step S804, calculating the similarity of the legal fact feature vector and the judgment reason feature vector of the legal judgment example, and determining the legal judgment result corresponding to the legal fact according to the similarity.

The invention converts the reasoning problem of legal judgment into the legal text retrieval problem. Traditional text matching processing objects are character level or word level features, and models are utilized to learn the character or word features of the text. The method can not completely learn the semantics of the text, so that the recall capability of legal text retrieval is not strong or the judgment events are not matched, and the characteristic representation of the text is the premise of improving legal text retrieval.

The invention carries out the vertical field characteristic enhancement of the legal field based on a transducer model, and is mainly divided into two aspects: 1. large amount of legal field text 2 and decision book scene learning: legal professional field data such as principal information, fact description, court opinion and decision results. The understanding capability and the characteristic retrieval capability of an algorithm to legal texts are improved on the basis of the original transformation model, and more importantly, the model can better adapt to the characteristic scenes such as principal information, fact description, court opinion, judgment result and the like in judgment and obtain good effects.

After receiving legal facts, text preprocessing can be performed first, and conventional text denoising means can be performed, wherein the conventional text denoising means comprise: removing special symbols, removing redundant blank, converting the traditional text into simplified text, and the like, and then carrying out intention analysis on legal text to obtain case classification by classification and requirement classification. And according to the analysis result, the classification of the requirements is carried out from the upper level of the map to the instance layer. According to the feature table then using the custom transducer law retrieval model, there are mainly two sources of data:

A. judging reason data: the decision reason is directly converted to a transducer fixed length feature, which is then converted if multiple reasons splice the reason into a segment.

B. Judging document data: and analyzing the court theory part of the judgment text, analyzing the judgment text into a appeal category, a judgment reason and a judgment result, and then answering the judgment result into fixed-length characteristics through a transducer table.

Based on the judgment reason data, converting the judgment reason data into a transducer table to be answered into fixed length characteristics, reasoning the most likely result according to the vector calculation module, and discarding the poor result according to the threshold value, thereby achieving the function of causal reasoning. The proportion of different judgment results in similar judgment conditions can be counted based on a statistical reasoning method so as to achieve the purpose of analysis and early warning.

The judging result mainly comprises the following steps:

(1) Whether or not the appeal is supported (trade contract dispute-term validity: commodity house trade contract relationship is valid and legal, and whether or not trade contract dispute-claim is valid: claim is valid).

(2) Judging the original notice or the notice (divorce judgment-tending right: judging the original notice).

As shown in fig. 9, in one embodiment, preferably, the method further comprises:

step S901, receiving an input search command for a target legal fact;

step S902, according to the search command, performing feature coding on the target legal facts by using a transducer legal coding model to obtain coded features;

and step S903, performing similarity matching on the coded features and the stored legal judgment document features and judgment event chains to retrieve the judgment document corresponding to the target legal facts.

As shown in fig. 10, in one embodiment, preferably, the method further comprises:

step S1001, receiving an input similar document retrieval command;

step S1002, determining a decision event chain corresponding to the current legal document according to the similar document retrieval command;

step S1003, performing feature coding on a decision event chain of the current legal document by using a transform law coding model to obtain coded event chain features;

And step S1004, performing similarity matching on the encoded event chain features and the stored legal judgment document features and judgment event chains to retrieve a legal judgment document similar to the current legal document.

The method comprises the steps of centering a core paragraph of a decision document (considered by a hospital), constructing decision document core logic, namely decision logic, constructing the decision logic into a chain structure, encoding the chain structure into core characteristics based on a customized legal form transducer, and carrying out vector calculation, wherein the main characteristics are hierarchical analysis category, document whole characteristics and core decision logic characteristics, and combining calculation scores, so that similar documents are retrieved.

In one embodiment, preferably, the method further comprises:

and learning and training the transducer model according to legal judgment documents of massive legal judgment examples to obtain the transducer legal coding model.

Fig. 11 is a block diagram illustrating a decision result prediction apparatus of a legal document according to an exemplary embodiment.

As shown in fig. 11, according to a second aspect of the embodiment of the present invention, there is provided a decision result prediction apparatus for legal documents, the apparatus including:

an obtaining module 1101, configured to obtain legal decision documents of a huge amount of legal decision examples;

A first determining module 1102, configured to perform legal element extraction and relationship determination between legal elements for each legal decision document, so as to obtain a decision event chain from an original notice type, a decision reason to a decision result, and an element relationship, where the legal elements include the original notice type, the decision reason and the decision result, and the element relationship includes a causal relationship and a corresponding relationship;

a construction module 1103, configured to construct a knowledge graph of each legal decision document with a legal decision instance as a center according to a decision event chain and element relation corresponding to the legal decision document, where the knowledge graph sequentially includes from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

a receiving module 1104 for receiving an input legal fact, wherein the legal fact includes a legal appeal;

and a prediction module 1105, configured to predict and output a legal decision result corresponding to the legal fact by using the knowledge graph and the legal retrieval model.

In one embodiment, preferably, the first determining module is configured to:

Performing legal element recognition on each legal decision document by using an element category classification model to obtain an element category classification recognition result;

and splicing the element category classification recognition result and the characteristics of the legal judgment document, adding case-by-case category information, inputting the case-by-case information into a reading understanding model, enabling the reading understanding model to determine the content and the position of legal elements according to the element category classification recognition result, and outputting a judgment event chain and element relation from original notice requirement type, judgment reason to judgment result.

In one embodiment, preferably, the apparatus further comprises:

and the storage module is used for taking Milvus as a characteristic storage, constructing an index set according to the level from top to bottom of the knowledge graph, obtaining legal judgment document characteristics, document vector characteristics, judgment event chains and event chain vectors by utilizing a transform legal coding model code, and storing the legal judgment document characteristics, the document vector characteristics, the judgment event chains and the event chain vectors in correspondence with the legal judgment document ID.

In one embodiment, preferably, the prediction module is configured to:

performing intention analysis on the legal facts to determine target case category and target appeal category corresponding to the legal facts;

searching in the knowledge graph according to the target case by category and the target appeal category to determine at least one corresponding legal judgment instance;

Performing feature coding on the legal facts and the judgment reasons of at least one corresponding legal judgment example by using a transformer legal coding model to obtain corresponding legal fact feature vectors and judgment reason feature vectors of the legal judgment example;

calculating the similarity of the legal fact feature vector and the judgment reason feature vector of the legal judgment instance, and determining legal judgment results corresponding to the legal facts according to the similarity.

In one embodiment, preferably, the apparatus further comprises:

the first retrieval module is used for receiving an input retrieval command for the target legal facts;

the first coding module is used for carrying out feature coding on the target legal facts by utilizing a transformation law coding model according to the search command to obtain coded features;

and the first matching module is used for matching the similarity between the coded features and the stored legal judgment document features and judgment event chains so as to retrieve the judgment document corresponding to the target legal facts.

In one embodiment, preferably, the apparatus further comprises:

the second retrieval module is used for receiving an input similar document retrieval command;

The second determining module is used for determining a decision event chain corresponding to the current legal document according to the similar document retrieval command;

the second coding module is used for carrying out feature coding on the judgment event chain of the current legal document by using a transformation law coding model to obtain the coded event chain features;

and the second matching module is used for matching the similarity between the encoded event chain features and the stored legal judgment document features and judgment event chains so as to retrieve the legal judgment document similar to the current legal document.

In one embodiment, preferably, the apparatus further comprises:

and the training module is used for learning and training the transducer model according to legal judgment documents of massive legal judgment examples so as to obtain the transducer legal coding model.

According to a third aspect of the embodiments of the present invention, there is provided a device for predicting a decision result of a legal document, the device comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

acquiring legal judgment documents of massive legal judgment examples;

carrying out legal element extraction and relation determination among legal elements on each legal judgment document to obtain a judgment event chain from original complaint type, judgment reason to judgment result and element relation, wherein the legal elements comprise original complaint type, judgment reason and judgment result, and the element relation comprises causal relation and corresponding relation;

According to the relation between the decision event chain and the element corresponding to each legal decision document, constructing a knowledge graph of the legal decision document by taking a legal decision example as a center, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

receiving an input legal fact, wherein the legal fact comprises a legal appeal;

and predicting and outputting legal judgment results corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

According to a fourth aspect of embodiments of the present invention, there is provided a computer readable storage medium having stored thereon computer instructions which, when executed by a processor, implement the steps of the method as in any of the embodiments of the second aspect.

It is further understood that the term "plurality" in this disclosure means two or more, and other adjectives are similar thereto. "and/or", describes an association relationship of an association object, and indicates that there may be three relationships, for example, a and/or B, and may indicate: a exists alone, A and B exist together, and B exists alone. The character "/" generally indicates that the context-dependent object is an "or" relationship. The singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.

It is further understood that the terms "first," "second," and the like are used to describe various information, but such information should not be limited to these terms. These terms are only used to distinguish one type of information from another and do not denote a particular order or importance. Indeed, the expressions "first", "second", etc. may be used entirely interchangeably. For example, first information may also be referred to as second information, and similarly, second information may also be referred to as first information, without departing from the scope of the invention.

It will further be appreciated that although operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results. In certain circumstances, multitasking and parallel processing may be advantageous.

Other embodiments of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the invention following, in general, the principles of the invention and including such departures from the present disclosure as come within known or customary practice within the art to which the invention pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.

It is to be understood that the invention is not limited to the precise arrangements and instrumentalities shown in the drawings, which have been described above, and that various modifications and changes may be effected without departing from the scope thereof. The scope of the invention is limited only by the appended claims.

Claims (8)

1. A method for predicting a decision result of a legal document, the method comprising:

step S101, obtaining legal judgment documents of massive legal judgment examples;

step S102, carrying out legal element extraction and relationship determination among legal elements on each legal decision document to obtain a decision event chain from original notice requirement type, decision reason to decision result and element relationship, wherein the legal elements comprise original notice requirement type, decision reason and decision result, and the element relationship comprises causal relationship and corresponding relationship;

constructing a judgment causal relationship, namely, a court theory, wherein the judgment causal relationship extraction of the legal documents is element structural extraction of a court theory part, and element extraction and causal relationship correspondence are carried out on the court theory part forming paragraphs or spreads;

step S103, constructing a knowledge graph of the legal decision document by taking the legal decision instance as a center according to the decision event chain and element relation corresponding to each legal decision document, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

Step S104, receiving an input legal fact, wherein the legal fact comprises legal requirements;

step S105, predicting and outputting a legal judgment result corresponding to the legal facts by using the knowledge graph and the legal retrieval model;

in the step S102, performing the legal element extraction and the relationship determination between the legal elements for each legal decision document includes:

step S501, performing legal element recognition on each legal decision document by using an element class classification model to obtain an element class classification recognition result;

step S502, the element category classification recognition result and the characteristics of the legal judgment document are spliced, and the category information is added to be input into a reading understanding model, so that the reading understanding model determines the content and the position of legal elements according to the element category classification recognition result, and a judgment event chain and element relation from original notice requirement type, judgment reason to judgment result are output;

in the step S105, predicting and outputting a legal decision result corresponding to the legal facts by using the knowledge graph and the legal retrieval model, including:

step S801, carrying out intention analysis on the legal facts to determine a target case category and a target appeal category corresponding to the legal facts;

Step S802, searching in the knowledge graph according to the target case by category and the target appeal category to determine at least one corresponding legal judgment instance;

step S803, performing feature coding on the legal facts and the decision reason of the at least one corresponding legal decision instance by using a transform law coding model to obtain corresponding legal fact feature vectors and decision reason feature vectors of the legal decision instance;

step S804, calculating the similarity of the legal fact feature vector and the judgment reason feature vector of the legal judgment example, and determining the legal judgment result corresponding to the legal fact according to the similarity.

2. The method according to claim 1, wherein the method further comprises:

and constructing an index set according to the level from top to bottom of the knowledge graph by taking Milvus as a characteristic memory, coding by using a transform law coding model to obtain law judgment document characteristics, document vector characteristics, a judgment event chain and event chain vectors, and storing the law judgment document characteristics, the judgment event chain and the event chain vectors in correspondence with a law judgment document ID.

3. The method according to claim 2, wherein the method further comprises:

receiving an input search command for target legal facts;

According to the search command, performing feature coding on the target legal facts by using a transformation law coding model to obtain coded features;

and carrying out similarity matching on the coded features and the stored legal judgment document features and judgment event chains so as to retrieve the judgment document corresponding to the target legal facts.

4. The method according to claim 2, wherein the method further comprises:

receiving an input similar document retrieval command;

determining a decision event chain corresponding to the current legal document according to the similar document retrieval command;

performing feature coding on a judgment event chain of the current legal document by using a transformer legal coding model to obtain coded event chain features;

and carrying out similarity matching on the encoded event chain features, the stored legal decision document features and the decision event chain so as to retrieve the legal decision document similar to the current legal document.

5. The method according to claim 2, wherein the method further comprises:

and learning and training the transducer model according to legal judgment documents of massive legal judgment examples to obtain the transducer legal coding model.

6. A legal document decision result prediction device using the legal document decision result prediction method according to any one of claims 1 to 5, characterized in that the device comprises:

the acquisition module is used for acquiring legal judgment documents of massive legal judgment examples;

the first determining module is used for extracting legal elements and determining relations among the legal elements for each legal decision document so as to obtain a decision event chain from an original notice type, a decision reason to a decision result and element relations, wherein the legal elements comprise the original notice type, the decision reason and the decision result, and the element relations comprise causal relations and corresponding relations;

the construction module is used for constructing a knowledge graph of the legal decision document by taking the legal decision instance as a center according to the decision event chain and element relation corresponding to each legal decision document, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

a receiving module for receiving an input legal fact, wherein the legal fact comprises a legal appeal;

And the prediction module is used for predicting and outputting a legal judgment result corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

7. A legal document decision result prediction device using the legal document decision result prediction method according to any one of claims 1 to 5, characterized in that the device comprises:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to:

acquiring legal judgment documents of massive legal judgment examples;

carrying out legal element extraction and relation determination among legal elements on each legal judgment document to obtain a judgment event chain from original complaint type, judgment reason to judgment result and element relation, wherein the legal elements comprise original complaint type, judgment reason and judgment result, and the element relation comprises causal relation and corresponding relation;

according to the relation between the decision event chain and the element corresponding to each legal decision document, constructing a knowledge graph of the legal decision document by taking a legal decision example as a center, wherein the knowledge graph sequentially comprises from top to bottom: the method comprises the steps of a root node, a case by category, a appeal category, a legal judgment example, a judgment reason and a judgment result, wherein the knowledge graph also comprises a label of element relation;

Receiving an input legal fact, wherein the legal fact comprises a legal appeal;

and predicting and outputting legal judgment results corresponding to the legal facts by using the knowledge graph and the legal retrieval model.

8. A computer readable storage medium having stored thereon computer instructions, which when executed by a processor, implement the steps of the method of any of claims 1-5.