Syntactic and Semantic Features Based Relation Extraction in Agriculture Domain

Abstract

Relation extraction plays an important role in many natural language processing tasks, such as knowledge graph and question answering system. Up to the present, many of the former relation methods work directly on the raw word sequence, so it is often subject to a major limitation: the lack of semantic information, which leads to the problem of the wrong category. This paper presents a novel method to extract relation from Chinese agriculture text by incorporating syntactic parsing feature and word embedding feature. This paper uses word embedding to capture the semantic information of the word. On the basis of the traditional method, this paper integrates the dependency parsing, the core predicate and word embedding features, using the naive Bayes model, support vector machine (SVM) and decision tree to build experiment. We use the websites knowledge base to construct a dataset and evaluate our approach. Experimental results show that our proposed method achieves good performance on the agriculture dataset. The dataset and the word vectors trained by Word2Vec are available at Github (https://github.com/A-MuMu/agriculture.git).