A knowledge map generation method based on bidirectional LSTM deep neural network

Abstract

The invention discloses a method for generating a knowledge map based on a bidirectional LSTM deep neural network, which includes inputting a source entity and a target entity into a knowledge base to obtain an attribute matrix; inputting the attribute matrices corresponding to the two into a multilayer perceptron respectively , the features output by the multi-layer perceptron and the word vector of the corresponding entity form a joint expression of attributes; the joint expression of the two attributes is combined, and the combined feature matrix is ​​used as the input of the bidirectional LSTM deep neural network, and the output hidden layer vector ;Take the hidden layer vector as the input of the classifier, output the predicted probability distribution of the corresponding relationship in the relational dictionary, take the relationship with the highest predicted probability as the entity relationship between the source entity and the target entity, and complete the generation of the knowledge map. The invention weakens the upper and lower relationship between the source entity and the target entity, solves the expression bias of the candidate entity to a certain extent, and improves the accuracy of relationship prediction.

Description

technical field [0001] This application belongs to the technical field of knowledge graphs, and specifically relates to a method for generating knowledge graphs based on a bidirectional LSTM deep neural network. Background technique [0002] A knowledge graph is a collection of interrelated entities, which are essentially auxiliary or derived datasets: they are obtained by analyzing and filtering the original data. More specifically, the relationship between these pre-defined data points (entities) is also an important part of the knowledge graph. Therefore, in the knowledge graph, we can analyze not only entities, but also the relationship between entities. [0003] Existing representations of knowledge graphs basically rely on triples, namely {source entity, relation, target entity}. In this context, the construction of knowledge graphs is extremely dependent on the extraction of relationships between entities. Existing relationship extraction methods directly use the s...

AI Technical Summary

Problems solved by technology

[0007] However, the unidirectional LSTM network has the problem of the sequence of information input

In the task of triplet relationship extraction, although the source entity and the target entity do not have duality, the input order will have a certain impact on the expression of the LSTM network, that is, the expression bias between entities

Images