Text feature extraction method and knowledge graph construction method

Abstract

The invention discloses a text feature extraction method and a knowledge graph construction method. The text feature extraction method comprises the following steps: constructing negative samples on the basis of marked entities and relationships as positive samples: entity negative samples, relationship negative samples and anaphora disambiguation negative samples; performing mapping characterization on the positive sample and the negative sample to obtain a vector set formed by entity characterization, entity pair characterization and anaphora disambiguation entity pair characterization; analyzing the vector set to obtain entity features, anaphora disambiguation features and relation features; classifying the entity features, the anaphora disambiguation features and the relation features; setting loss function, and evaluating classification result. According to the method, the entity is taken as the center, anaphora disambiguation is assisted, the problem of entity loss caused by anaphora non-entity is solved, the accuracy of the whole model is improved, the relation characteristics are fused, and the accuracy of relation identification is further improved.

Description

Technical field [0001] The present invention relates to the field of natural language processing, and more particularly to a textual characteristic extraction method and a knowledge map construction method. Background technique [0002] Knowledge map is a semantic network that represents information and relationships in the form of a map data structure, thereby being used to further excavate the relationship between information and information hidden in the information. [0003] The graph in the knowledge map, consists of nodes and edges, where the nodes are used to represent concepts and entities, while indicating the relationship and attributes of things. How to extract the input or existing information (including entity extraction, relationship extraction, and attribute extraction), resulting in knowledge representation, is the basis and premise of the next construction of knowledge maps. [0004] In the prior art, there are two main methods, the first is two steps, which is t...

AI Technical Summary

Problems solved by technology

However, the existing one-step relationship extraction algorithm mainly designs the feature extraction model around the entity-relationship pair, and rarely focuses on the feature extraction design with the entity as the core, so the final accuracy of the relationship extraction is not high. , reducing the accuracy of subsequent relation classification tasks

[0007] In addition, the existing methods are difficult to deal with the reference problem of entities in the text, resulting in the loss of features

Images