Research On Chinese Hyponymy Relation Automatic Extraction

There are a variety of relations between words, such as Hyponymy, synonymy, antonymy, whole-Part relation etc. As one of the most important part, Hyponymy describes the base classification methods between objects. In the field of NLP(Natural Language Processing), the hypernymy is a kind of subordinate relation in semantic. For instance, those words which satisfy the conditions of Hyponymy can be described as "Word B is a kind of word A". We can say A is the hypernym of B, or B is the hyponym of A. Besides, A is the class of B, and B is one of instances of A.One of basic tasks of semantic extraction on words is how to extract the relation of hypernymy correctly and efficiently. For supporting the advanced knowledge extraction, this task try to convert those Non-formatting information into a hierarchy. In the fields of Machine Translation, textual Entailment and Information retrieval, Hyponymy plays a important part in supporting the task like ontology knowledge database extension, correct detection and improvement.This paper attempts to combine a series of methods to solve the problem of hyponymy acquisition and validation.For the acquisition task, by combined with the algorithm of pattern self-extension and Chinese word definition of Wikipedia, we propose a hyponymy extraction method based on Latent Dirichlet Allocation (LDA) modal.For the validation task, by calculating the Contextual Feature Similarity (SimCF) and Brown Clustering Similarity (SimBrown), we present a novel approach of hyponymy validation based on combination of Contextual Feature and Brown Clustering. Evaluation on CCF NLP&CC2012 Word Semantic Relation corpus shows that the approach achieved a good result.