Word2vec-based remotely supervised non-taxonomic relation extraction method and system

Abstract

The invention discloses a word2vec-based remotely supervised non-taxonomic relation extraction method and system, which can extract non-taxonomic relations in the field of vegetables. The method comprises the steps of crawling network vegetable field non-structured text data of a network encyclopedia and a large vegetable website to serve as corpora, and preprocessing the corpora in sequence to obtain primary training corpora; and training a word2vec model by utilizing the primary training corpora, and obtaining a spatial vector of each sentence by utilizing the word2vec model; aggregating the primary training corpora according to the types of the non-taxonomic relations, and for aggregated data of each relation, extracting a common sentence mode and an uncommon sentence mode; selecting two sentence spatial vectors accordant with two different modes as initial centers of a k-means clustering method, clustering all sentence spatial vectors, selecting a category accordant with the common sentence mode, and obtaining the training corpora with relatively high quality; and training a convolutional neural network model by the training corpora with relatively high quality, and through a fully connected softmax layer, extracting the non-taxonomic relations.

Description

technical field [0001] The invention relates to the field of weakly supervised classification, in particular to a word2vec-based far-supervised non-categorical relationship extraction method and system. Background technique [0002] At present, the research on ontology-like knowledge graphs in the agricultural field is still in its infancy, and there are relatively few relevant literature reports on non-categorical relationships (other relationships except the relationship between the classification relationship of the upper and lower relationships). Although there are literatures that also involve the learning of non-categorical relations in the fields of ancient agriculture and tea science, such as "Research on Semi-automatic Construction and Retrieval of Domain Ontology" by He Lin, "Research on Modeling of Ontology Learning for Vegetable Field" by Xu Jicheng etc., but they all use the most basic association rule method to discover the concept pairs with relationship. Not...

AI Technical Summary

Problems solved by technology

Moreover, the accuracy of the extracted non-categorical relationship is much lower than that of the general classification relationship.

Images