Keyword search technology over relational databases has become one of hot topics in the field of information retrieval. It can provide the users with little SQL knowledge a simple and friendly interface. But the algorithms of some existing keyword search systems are mainly based on database graph or schema graph. However, the efficiency of them which use database graph or schema graph separately is low. The accurate rate of results is also not high. This paper devises and implements a top-k keyword search system KWSDS (keyword search system based on database graph and schema graph). After users input the keywords, it can eliminate some dirty keywords through pre-processing. The method of combining database graph and schema graph together to solve the problem of keyword search is proposed for the first time. This paper also devises the search algorithms between same table and different tables, prove the correctness of the algorithms and analyze time complexity of them. A sorting method based on relevance is proposed at the same time. The algorithms of KWSDS system run in shorter time than the existing algorithms, the results are output by KWSDS system with high accuracy. The system has excellent query performance. Finally, KWSDS is verified by abundant experiments.