Method and device for constructing legal knowledge graph based on joint entity and relation extraction

A method for constructing a legal knowledge graph based on joint entity and relation extraction, comprising the following steps:

• constructing a triple data set;
• splitting a sentence of a legal text into short sentences;
• completing default subjects in the short sentences using zero anaphora resolution and dependency parsing;
• extracting triples from the short sentences and constructing the triple data set;
• designing a model architecture and training a model;
• wherein the model architecture comprises an encoding layer, a head entity extraction layer and a relation-tail entity extraction layer;
• and
• wherein the encoding layer uses a Bert pre-training model;
• two BiLSTMs are used as binary classifiers in the head entity extraction layer, and encoding of a text is used as an input of the classifiers;
• in output information, an entity start position corresponding to a first BiLSTM binary classifier is output as 1, with outputs of other positions being all 0, and an entity end position corresponding to a second BiLSTM binary classifier is output as 1, with outputs of other positions being all 0;
• the relation-tail entity extraction layer combines encoding information of head entities with the encoding of the sentence as an input, and for each head entity, finds tail entities that may exist under each relation, and finally obtains a complete triple;
• determining the relation between the sentences of the text;
• determining, for a sentence of the legal text that has not been split into short sentences, the relation between the short sentences;
• and
• combining triple and visualizing graph;
• obtaining a combined triple corresponding to the legal text according to the triple extracted by the model and the relation between sentences of the text;
• knowledge graph generation and storage;
• generating the legal knowledge graph based on the combined triple, and storing the legal knowledge graph in a memory for subsequent retrieval and invocation;
• application-oriented invocation and visualization:
• executing computer-implemented legal applications based on the stored legal knowledge graph, comprising responding to question-and-answer reasoning requests and/or outputting relevant recommendation results, and visually displaying the legal knowledge graph.