An Opinion Retrieval Method Based on Text Conceptualization

Deyuan LIU,Jingjing WEI,Yunbing WU,Xiangwen LIAO

DOI: https://doi.org/10.13451/j.cnki.shanxi.univ(nat.sci.).2018.02.006

2018-01-01

Abstract:Opinion retrieval is a hot topic in the research of Natural Language Processing.Most existing approaches for text opinion retrieval cannot extract knowledge and concept according to the context.This sparse feature space influences the performance of the opinion retrieval.We introduce a novel opinion retrieval method based on text conceptualization.In this method,we use knowledge graph to calculate the similarity between different texts in conceptual space to handle the sparse feature space.Firstly,we use the conceptual knowledge graph to conceptualize the queries and texts into the correct conceptual space.We calculate the similarity scores between queries and texts in the concept space simultaneously.Secondly,we capture the opinion scores of texts using the opinion mining methods based on opinion lexicon.Finally,the relevant opinion scores will be induced by the quadratic combination of relevant scores and opinion scores.The experiments on two datasets show that compared with benchmark methods,our approach improves the MAP scores by 4.6％ and 12.6％,respectively.