Research On The Key Technology Of Semantic Search Based On Personalization

With the explosive growth of information capacity on Internet, information retrieval plays a more and more important role in the process of obtaining information resources. Internet information retrieval technology takes use of certain strategies and specific computer algorithms to collect information from the Internet, and return results to users after organization. Nowadays, search engines have become the main way for users to obtain information from Internet.However, today’s mainstream search engines like Google, Baidu, MSN, Bing, Yahoo, etc are based on keywords and word frequency statistics, which can not satisfy users in many aspects. First, too many useless search results. Second, due to the diversity of users, it is hard to meet each one’s special needs. At last, without considering the semantic correlation between the words, the search results are not effectively organized, wasting users’ time to browse and pick. Semantic search can solve these problems in some degree.Semantic search is a search method that seeks to improve search accuracy by understanding searchers’ intent and the contextual meaning of terms, to generate more relevant results. A series of researches have been made in recent years. These studies of semantic search have made their contribution, but some problems remains to be solved. First, implementing the method based on the classification of users’ queries requires harsh conditions, and the time consumption increased a lot. Secondly, they did not take the different significance of various elements into account. Finally, the organization structure of search results can be further improved.This paper presents a novel and practical approach for semantic search based on Multi-classification Semantic Analysis(MSA) and personalization through deep researches in frontier semantic retrieval techniques. The achievements of this research include the following aspects:In this approach, MSA method is presented to organize the documents, and with the help of Term Vector Database(TVDB) which constructed in MSA process, a series of personalization algorithms are proposed to "match" users’information. To evaluate our approach, several experiments are conducted with two different baselines. Results show that overall, our approach reaches a satisfactory performance with a relatively low time cost.