With the rapid growth of web services, how to efficiently locate the desired web services has become a very important problem. We observe that under different contexts the descriptions of services vary in their level of semantic ambiguity. So the traditional keyword search is insufficient in the very small text fragments in web services, and semantic of the web services is not exploited. We describe the algorithms based on the complex networks for web services. Firstly with the communities discovered from the complex networks, services sharing the similar context could be clustered into same groups. Then by targeting the users' queries to the relevant group, utilize similarity metric in this group to search services. The experiment shows the approach can improve the accuracy of service search.