Improved Semantic Similarity Computation in Question-Answering System

P. Jiang, H. Hu, F. Ren, and S. Kuroiwa (Japan)


Similarity, How-net, Sememe, Question-Answering System.


In this paper we propose an improved semantic similarity computation method for a Chinese question answering (QA) system [1]. The question answering system is capable to find out what answer are satisfy user’s requirements. Most successful QA systems aim at English. The conventional method used in Chinese QA systems for retrieving the related answers is Vector Space Model (VSM) in natural language processing (NLP). The VSM can only search by the keywords in the questions without the semantic relative information. The searching method in terns of semantic similarity considers not only the keywords but also the words related with keywords in semantics. We employ an improved method to use How net [2], a lexical knowledge base to calculate the word similarity and then the sentences similarity. The methods have been applied in our Japan Travel Question Answering System and the experiments are presented with satisfying results.

Important Links:

Go Back