Abstract: With the rapid development of Internet economy，POI（Point Of Interest）search becomes the core technology of spatial information services.The ultimate goal of POI search engines is undoubtedly to improve customer satisfaction.Through mining user access logs and establishing the feedback similarity model，the accuracy of search results is improved.The POI search engine is optimized.Through theoretical analysis，the method improves the accuracy of the search results without extra computation time.The method is applied to Tongmap（www.tongmap.cn） which is a map search engine developed by Geographic Information Center，Institute of Computing Technology，Chinese Academy of Sciences.Finally，simulation work shows that the method is effective in POI search engine optimization.