Abstract: The current knowledge retrieval is retrieval speed is slow, the problem of low efficiency. Put forward a way of knowledge retrieval based on natural language processing, the use of a semi-supervised algorithms for natural language description, to construct knowledge retrieve objects for all undirected graph, in the undirected graph, transmission method to design a label, there will be passed to the rest of the data sample no marking data samples, complete all the data in the undirected graph marking process. Using TextRank way, in view of the data to select keywords after all the tags, according to the keywords in retrieve objects to realize the process of knowledge retrieval. The experimental results show that the algorithm presented in this paper a knowledge retrieval, can effectively improve the speed of knowledge retrieval, in a short period of time to complete the process of knowledge retrieval, improve user satisfaction.