The Research Of Intelligent Database Selection Based On User Model

Search engine face the problem that recall and precision are decline ,the emergence of vertical search service improve the precision which user search the special topic . But as the number and diversity of vertical search engine and universal search engine based on the special topic on the Internet exponentially increase. User can not decide directly which can provide better service. Therefore , it is an efficient way to improve recall and precision via constructing meta search engine that integrate vertical search service and universal search service based on the special topic. But, constructing meta search engine will face many challenges.database selection is one of the most important problem. Database selection is aim at selecting a few search engines that contain as many as relvent documents so as to response user query. But it's difficulty is that databases are heterogeneous and uncooperated. Heterogeneity appears in different forms: query languages,communication protocols,document models ; uncooperation appears in that database is not provided statistical metadate information. For the reason we know little about the heterogeneous and uncooperated database , we can't estimate the relevent document distribution exactly.This dissertation makes a deep research into the intelligent database selection based on user model. Firstly, to construct user model we use statistical language modeling via collecting user implicit feedback information , and then using language model and topic hierarchy tree to construct database model. Secondly, to match topic database adaptively via user model updateing actively on the time axis. On the basis of the two points above, this dissertation propose two-stage database selection based user model. Firstly,integrating user preference into the first-stage,for improving topic hierarchy database selection precision. For improving the estimation that revelent document distribution ,this dissertation modify the UUM (unified utility maximization) framework,And then,in the second stage,Using modified UUM framework to improve the relevant document distribution estimation . select the database that contain as many as relevent documents.This dissertation compare with the traditional database selection by experimentation. It is conclused that improving topic hierarchy database selection precision by integrating user model,at the same time,improving the revelent document distribution estimation between similar topic database. Effective database selection can improve the distributed information retrieval performance, expecially can improve the recall and precision.