As the amount of information on the Web increases rapidly, search engines have already become the main tools for information retrieval. Meta search engines are proposed to increase the search coverage by combining several search engines. However, the problem of the information overload becomes more severe and most returned results are irrelevant to user' s interests. That affects the retrieval quality and increases the user cost. In order to improve the accuracy of returned results, Researches and experiments on personalized search for meta search engines are carried out in this paper.The state-of-the-art of meta search engines is firstly overviewed as well as that of personalized search. Then the analyses and comparisons of personalized search techniques that have been proposed for meta search engines are provided. Afterwards, based on user profiles, we propose several algorithms for personalized search. My main research work includes:(1) The method of mapping a user query to a set of categories of user interests is given. It deduces the categories that are likely to be of interest to the user on his/her query and profiles. These categories represent the user's search intention and provide the foundation for search engines scheduling and results merging.(2) A novel algorithm of database representation is proposed. It probes the usefulness of search engines by query-based classification sampling. And a personalized scheduling strategy incorporating user' s interests is designed to select most latently useful search engines for user query. The strategy promotes the effectiveness and efficiency of retrieval by computing the similarity between search engines and user query.(3) An algorithm of results merging based on user profiles is brought forward. It provides users with suitable results by removing the reduplicate results and estimating the similarity between results and user query.In the end, we design and implement a personalized meta search engine, PMSE. And on it a variety of experiments have been carried out. The analysises of experimental data indicate our methods are effective. |