Chinese HowNet-Based Multi-factor Word Similarity Algorithm Integrated of Result Modification

Abstract

In this paper, we firstly describe a novel approach to calculate the Chinese sememe similarity based on the HowNet hierarchical sememe tree. When we calculate the sememe similarity, we not only take Semantic Distance, Node Depth and Semantic Coincidence Degree into consideration, but also propose two impact factors named Node Environment Dense (NED) and Node Layer Ratio (NLR) to optimize the calculation process. Secondly, quite a few words described by identical concept definition in HowNet should have a certain discrimination according to human perception, so we propose a hybrid modification algorithm integrated of TongYiCi CiLin (hereinafter, CiLin) to deal with this case. Experiment results of the HowNet-based multi-factor similarity hybrid algorithm shows that this approach improves the similarity of independent sememe words and the words having identical concept descriptions in HowNet, while no large bias influence on the similarity of other words.