A Potential Semantic Analysis Model that Incorporates Chinese Dependent Syntactic Analysis

Sentence similarity computation is one of the important foundations of Natural Language Processing(NLP). Latent Semantic Analysis(LSA) is a corpus-based method that evaluates the similarity of text by mining the semantic relationships between words and sentences through Singular Value Decomposition(SVD), but LSA ignores syntactic construction and cannot deal with the problem of contextually similar and semantic opposites, nor can it distinguish whether the sentence structure is complete, i.e., there is a syntactic blindness problem. Moreover, Chinese is very different from English in structure as well as expression, so the traditional LSA has more limitations in calculating Chinese sentence similarity. To solve this problem, this paper proposes a new latent semantic analysis algorithm as eLSA, which uses Dependency Parsing(DP) to extract the core words of Chinese sentences, i.e., subject, predicate and object, and combines with latent semantic analysis algorithm to analyze the semantics of the sentences. By testing on contextually similar sentences with different semantics and comparing with the traditional latent semantic analysis algorithm, the experimental results show that the algorithm alleviates the problem of syntactic blindness to a certain extent, and the sentence-to-sentence similarity scores are lower than those provided by the latent semantic analysis algorithm, providing a more realistic semantic similarity score.