Ranking Model In Search Engines Based On Different Objects And Performance Evaluation

Nowadays, the size of Internet is growing rapidly. Internet is one of the most important methods for user to get information, but the problem of information overload has come, which means users are difficult to gather relative information due to too much irrelative information on the Internet. Therefore, search engines are created. In the development of search engine, there exists two challenging research scope now. First, traditional search engines are based on Web pages, but beyond Web pages, now there exists other forms to represent information on the Internet. Therefore, we need new technique of search engine. Second, there are many search engines, and their quality varies considerably. Users want to know which search engine can be trusted in which scope, so we need a technique for performance evaluation of search engines.The main contribution of this paper can be summarized as follows. First, this paper presents a new ranking model for object-level search. This model is a generic model because it is based on the linkage of objects, and can be used on varies types of objects. Second, this paper presents the concept of UIG, which is more accurate than traditional DCG in measuring information in result list. Finally, this paper presents an ideal distribution family of search engine's clicks, and this distribution can be used in search engine performance evaluation. The parameters in the distribution family can be fit based on search engine's click data. Moreover, the difference between actual click distribution and ideal distribution reflexes the defect and points out possible improvement in search engine's ranking algorithm.The object-level ranking model represented is based on the link graph of objects, calculates the object relevance to single term, and merges those relevance in case of multiple terms. The model is verified in ACM Portal data set and gets satisfying results. The click distribution model represented is based on Weibull distribution, and modified due to some features of search engine. The model is verified on AOL, Sogou and Microsoft search engines, and gets satisfying results.