Court electronic file oriented case information automatic extraction method

Abstract

The invention discloses a court electronic file-oriented case information automatic extraction method, which comprises the steps of creating a case information extraction framework, performing statistical analysis on case elements in various case electronic file files, and constructing a basic case body; customizing a basic case information extraction strategy according to the case ontology and the layout characteristics of the electronic file; constructing a case information extraction framework based on a cross-domain information extraction tool TZIE of XML; building an NER model and a training and multi-granularity semantic unit combination mode based on multi-granularity semantic legal documents, employing a Word2vec Skip-gram model and an LDA topic model, obtaining a judicial domain word vector model and a judicial domain word vector model through training respectively, combining domain entity recognition of a BiLSTM-Attention CRF model, adding an Attentio mechanism in a referencemodel BiLSTM-CRF, predicting the probability that each semantic unit belongs to different labels, assisting an optimized model training mode, taking a manual annotation corpus training model as a main task, and taking manual and automatic annotation corpus training models as auxiliary tasks.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence text information extraction, in particular to an automatic extraction method for court electronic case information. Background technique [0002] At present, courts at all levels have accumulated large-scale electronic case files in the process of informatization. For the unstructured text information recorded in massive electronic files, there is an urgent need to use more efficient methods for structured and knowledge-based processing, which provides the depth of court electronic files. Application lays the groundwork. [0003] Information extraction is the process of obtaining structured data from unstructured text. Structured data mainly includes four data types: entity, relationship, attribute, and event. For a court case file, the case information is composed of The semantic network formed by the above four elements, the process of extracting information from electronic file...

AI Technical Summary

Problems solved by technology

[0003] Information extraction is the process of obtaining structured data from unstructured text. Structured data mainly includes four data types: entity, relationship, attribute, and event. For a court case file, the case information is composed of The semantic network formed by the above four elements, the process of extracting information from electronic files is also the process of constructing case knowledge ontology. This process usually includes four links: symbolic annotation, syntactic analysis, extraction, alignment and merging. The research results of foreign text information extraction methods are mainly divided into three categories: method based on rule template; method based on statistical machine learning; Analysis and other natural language processing technologies, writing rule templates, and extracting matching pattern information. Later, researchers used unsupervised methods to automatically generate new rules to expand the template library, and made some progress. The writing of rules is extremely dependent on domain knowledge, which makes rule-based The template method has a high extraction accuracy for specific corpus, but it is often unable to be transplanted across domains and has poor generalization ability

[0004] The method based on statistical machine learning is to solve the problem of information extraction as a sequence labeling problem. It is mainly divided into traditional machine learning methods and deep learning methods. Currently, the commonly used traditional machine learning methods are: hidden Markov model (Hidden Markov model) Model, HMM), maximum entropy Markov model (Maximum Entropy Markov Model, MEMM), and conditional random field model (Conditional Random Field Model, CRF), these three models are all through the known observation sequence X, to solve The optimal labeling sequence Y, the difference is that the assumptions and constraints of the three dependencies are reduced in turn, and CRF is more widely used to solve the sequence of On the issue of labeling, CRF training needs to set feature values ​​based on experience, and it completely relies on the training lexicon for entity judgment. Compared with methods based on statistical machine learning, the deep neural network model has the advantages of better generalization and less dependence on artificial features, and has been widely used in named entity recognition in various fields, but for legal document named entity recognition The main problem is that the length difference between different named entities is large, which makes the semantic granularity of language representation have a greater impact on the model training effect, resulting in a large difference in the recognition performance of different named entities. In the dossier documents of different types of cases, naming The contextual characteristics of entities are significantly different, resulting in poor robustness of the model applied to different types of case files. Training deep learning models requires a large amount of labeled corpus, but the current legal document annotation corpus is insufficient

Images