Query suggestion is an interactive approach for search engines to better understand user information need. A Term-Query bipartite graph was trained by extracting semantic relationships from snippet clicked by query. With the combination of Query-URL graph and Query-Flow graph, a heterogeneous Term-Query-URL information network was constructed. Random walk with restart (RWR) was performed on the information network for query suggestion. The relevance of long tail query suggestion was greatly improved by taking into account semantic information and log information. Term vector of query was constructed on the basis of probabilistic language model for query suggestion of new query. The experiment results have shown that our approach outperforms baseline methods by about 3% to 10%.