数字内容技术与服务研究中心知识库

This paper presents a method to calculate the semantic similarity with TongyiciCiLin and Word2vec. In the part of CiLin, the semantic similarity of words is calculated by using the distance of words as the main factor, the number of branches and the distance between branches as the fine-tuning parameters. In the part of Word2vec, this paper constructs a special Corpus based on movie review, and uses Word2vec model to calculate the semantic similarity of Chinese words. Then, the final semantic similarity is calculated by using the dynamic weighting strategy to fuse CiLin and Word2vec. The method makes full use of the semantic information of words in the knowledge base and Corpus. The experimental results show that the algorithm has better accuracy and more robust to domain sensitivity.