A knowledge map construction method based on deep learning

Abstract

The invention relates to a method for constructing a knowledge map based on deep learning, which includes: given a target text sentence, using a bidirectional long-short-term memory cyclic neural network model and a conditional random field model to identify target entities in the target text sentence; using context-sensitive bidirectional long-short-term The memory recurrent neural network model and the forward neural network model extract the relationship between two target entities; use the vector space model to normalize the target entity, and map the normalized target entity to the concept; according to the target entity, the target entity The relationship and concept between construct knowledge graph. The present invention applies the deep learning technology to the construction of the knowledge graph graph, adopts the entity recognition model of the bidirectional cyclic neural network and the conditional random field to recognize the target entity in the target text sentence, and reduces the feature engineering in the process of entity recognition and relationship extraction , to reduce the burden and trouble caused by manual design and adjustment of features, and accurately mine the knowledge in the text.

Description

technical field [0001] The present invention relates to the technical field of natural language processing, in particular to a method for constructing a knowledge graph based on deep learning. Background technique [0002] As a huge, open, heterogeneous and dynamic information container, the Web generates and accommodates a huge amount of text, data, multimedia, temporary data and other types of information. Due to scattered resources and no unified management and structure, it is not easy to obtain relevant information, and the content that people are really interested in is often submerged in a lot of irrelevant information. Only by conducting in-depth semantic mining of network content and understanding users' interests from the semantic level can we provide people with high-quality Internet information. In this context, search engine companies such as Google, Baidu, and Sogou have built knowledge graphs based on this. Knowledge graphs aim to describe various entities o...

AI Technical Summary

Problems solved by technology

Feature-based models often need to design a large number of lexical, syntactic and semantic features, and then put them into classifiers such as support vector machines (SVM) for classification. The biggest problem with feature-based methods is that it takes a lot of time and time. Efforts to construct features

And another method based on kernel function, although there is no need to build a huge feature engineering, it is very difficult to design and choose a suitable kernel function

Images