How to obtain the high-level knowledge, such as semantic knowledge, to do deep processing of natural language is a focus in the field of natural language processing (NLP) these days. In this paper, we do research on the construction of semantic acknowledge base against the background of post-processing of speech recognition. Based on reviewing existing methods, authors present and realize a new domain-independent algorithm that applies to a single document and considers the impact of context. In the test, this algorithm shows well performance and the knowledge base can meet the need of project.