Event relationship extraction method for massive data sets

Abstract

The invention discloses a mass data set event relationship extraction method, which comprises the steps of S1, establishing an association relationship and association strength among triples accordingto an association rule to form an undirected network; S2, connecting the preceding word vector, the posterior word vector and the entity type in the triad to serve as features of nodes in the undirected network; and S3, performing classification processing on each node in the undirected network, and extracting an entity relationship in the event. According to the method, the problems of extraction precision and speed of event relationships in a big data set or a mass data set are solved, and parallel operation can be carried out by using a convolution network, so that the problem of extraction speed is solved; and meanwhile, the convolutional network is used for extracting the features on the graph data structure, so that the defect of low precision caused by weak extracted features in the traditional method is overcome by utilizing the advantage of strong feature extraction of the convolutional network.

Description

technical field [0001] The invention relates to the field of event relationship extraction, in particular to a method for extracting event relationship from massive data sets. Background technique [0002] At present, in the development of the knowledge graph system, for the extraction of event relationships, the mainstream algorithms are based on remote supervision algorithms. This algorithm is more practical for small data sets. Once the number of entities in the data set reaches tens of millions, It faces shortcomings such as slow calculation, low accuracy of event relationship extraction, and the need for a large number of manually labeled training samples. It is difficult for the traditional remote supervision training algorithm to reach 60% on large and massive data sets, and the number of manually labeled samples required by the traditional method is basically 100,000 or even higher. Contents of the invention [0003] In order to solve the above problems, the prese...

AI Technical Summary

Problems solved by technology

[0002] At present, in the development of the knowledge graph system, for the extraction of event relationships, the mainstream algorithms are based on remote supervision algorithms. This algorithm is more practical for small data sets. Once the number of entities in the data set reaches tens of millions, Faced with the shortcomings of slow calculation, low accuracy of event relationship extraction, and the need for a large number of manually labeled training samples

It is difficult for traditional remote supervision training algorithms to reach 60% on large and massive data sets, and the number of manually labeled samples required by traditional methods is basically at the level of 100,000 or even higher

Images