The current Natural Language Processing (NLP) means is chiefly grounded on the principle of language rules' description pattern, which lacks comprehensive semantic representation capability, and therefore it can not efficiently process the real texts in the layer of semantic meaning. This paper presents a semantic representation approach based on term connection samples. This method promotes the idea of describing fundamental language samples, and can analyze comprehensive semantic representations. So far it has been applied in project CAPC (Computer Aided Poetry Composing) funded by the Chinese Natural Science Foundation.