| Search Engine is a popular tool for information retrieval (IR) nowadays. However, the traditional search engine dose not do a job in identifying the individual's unique search goal, while it only returns the retrieval results associating with the query user provides. In order to overcome the problem, there have been many attempts to improve retrieval accuracy based on personalized information retrieval technology.Based on in-depth survey on the existing studies about personalized IR, this paper discusses separately on the short-term context and the long-term context personalized information. The thesis did research on how to use the context-based personalized information to improve the retrieval accuracy. Finally, a prototype system of client-side personalized search agent was built. The main work of the thesis includes:In terms of the short-term context personalized information, first we proposed a formal description of the personalized information retrieval. Second, based on the statistical language model and the implicit feedback technology, we proposed an approach to iteratively update the query model according to the short-term context such as query histories and clicked documents. This method was proved to be satisfied with the user ad hoc query intention. By the experiment, it found that the the mean average precision is increased by about 50%.In terms of the long-term context personalized information, first we discussed about long-term context, which is based on the research of the Peking University Network Center on the Skynet search engine. Second we improved the approach of building long-term user profile by interpolating the query history language models. Finally by the experiment, we found that this improved method has the good anti-noisy performance.Finally, a prototype system of personalized IR system was built, which used the context-sensitive personalized information. This system was built as a real experimental environment for the future research. |
PDF Full Text Request