The present natural language processing (NLP) methods mainly employ bottom-up language analysis ways, which have only limited semantic analysis capabilities. Therefore it is impossible to carry out semantic processing for large amounts of real texts. This paper presents a semantic analysis approach and corresponding algorithms based on term connections. The approach has rooted in semantics based on term connections and taken a major sentence cut-in, bottom-up semantic analyzing process. The main algorithm is to find the optimal sentence tree based on the semantic meaning conformity of term connections. So far, this approach has been applied in project CAPC (Computer Aided Poetry Composing) funded by the Chinese Natural Science Foundation.