Relation extraction system integrated with dynamic word vector

Abstract

The invention provides an entity relationship extraction method and system integrated with a dynamic word vector technology. According to the system, an existing knowledge base corresponds to rich unstructured data through a remote supervision method, so that a large amount of training data can be generated, the problem that manual annotation corpora are insufficient can be solved, dependence on annotation data can be reduced through the system, and therefore the labor cost can be effectively reduced. In order to obtain feature information between entities as much as possible, the infrastructure of a model adopts a piecewise convolutional neural network; and a dynamic word vector technology is fused to further extract semantic information of the example sentences.

Description

technical field [0001] The invention relates to the field of information extraction, in particular, to mining the semantic relationship between entities. Background technique [0002] Information extraction aims to extract structured information from large-scale unstructured or semi-structured natural language texts. The main tasks are entity extraction, relation extraction, and event extraction. Among them, the main content of relation extraction (Relation Extraction, RE) research is to excavate the semantic relationship between entities from the text content, and use the relation extraction technology to excavate the deep relationship structure between entities, which has profound theoretical significance and Huge research value, it is also the basic work for optimizing search engines, establishing knowledge graphs, and developing intelligent question answering systems. [0003] Practice has proved that supervised learning methods can extract more effective features, and ...

AI Technical Summary

Problems solved by technology

[0004] The use of neural network models mainly faces two problems: (1) There is not enough labeled data, the training data set has a low coverage of entities and entity relationships, and it does not perform well in terms of versatility

Moreover, it takes a lot of time and energy to manually label training data; (2) Since word usage is complex and changeable in semantics and grammar, existing models use pre-trained word vectors that are "static" and cannot follow the language changing environment

As a result, its ability to represent is limited

Images