Text language association extraction method and system based on recurrent neural network

Abstract

The invention discloses a text language association extraction method and system based on a recurrent neural network. According to the method, complex context features are automatically extracted based on a recurrent neural network (bidirectional long-short-term memory network), and semantic information of contexts is encoded; finding a definition mode in the document through a rule-based entity expression extractor, identifying definitions related to non-standard expressions in the document, and extracting defined standard expressions and non-standard expressions belonging to the same entityconcept; encoding the extracted features of the entity expression pair, and embedding information about entity normalization into a low-dimensional entity expression vector; the entity expression vector and the context feature coding vector are connected and subjected to dimension conversion to obtain a final code; and decoding a globally optimal state sequence as a final output sequence based ona decoder of the conditional random field in combination with the transition probability between the features learned by the encoder and the state. According to the invention, the entity identification performance can be effectively improved.

Description

technical field [0001] The invention belongs to the field of artificial intelligence, and relates to extracting information from massive unstructured data using natural language processing technology, specifically refers to identifying entities and extracting entity associations from text, which is a key technology for information extraction. Background technique [0002] Text entity extraction is to identify meaningful entities from text, such as person names, place names, organization names, etc. It is a key technology for extracting information from massive unstructured data, and is the cornerstone of many complex natural language processing applications, such as intelligent question answering, knowledge graph, automatic summarization, machine translation, etc. [0003] Due to the rich expression forms of natural language, the same entity may have many different expressions, such as the full name, abbreviation and alias of the entity. The phenomenon of "multiple words wi...

AI Technical Summary

Problems solved by technology

The cost of these feature engineering is high, and it is difficult to expand on large-scale data sets. It does not play the role of massive data well, and it is not data-driven.

Moreover, many hidden high-order interactions of contextual features cannot be covered by artificially constructed features.

Moreover, the entity normalization modules in these methods all rely on existing dictionaries, and there is an unreasonable assumption that "standard entity representations already exist in dictionaries".

However, the coverage of existing dictionaries is limited, and many corpora lack dictionaries in corresponding fields.

Especially in today's highly developed information technology, some new entities often appear in the text of the news media, such as reports on newly established institutions, newly issued bonds, new events, etc. These new entities do not exist in the existing In some dictionaries and knowledge bases, dictionary-dependent methods cannot normalize the names of such new entities

Images