Reinforcement Learning Agent Applied Research In Personalized Information

Since 1990s', Internet has been developed rapidly. The World Wide Web(WWW) provides users with a massive valuable information source, and gradually becomes a huge distributed information space. It becomes a problem needed to solve urgently for users to find information rapidly and accurately form millions of web pages. Traditional searching engines don't satisfy users' need of high quality information, though they solve some problems of information orientation. Therefore, many researchers are trying to apply the theory of Artificial Intelligence to information searching to improve recall and precision of information searching. In this paper, we elaborate some domains an information agent involves, introduce reinforce learning into dynamic scheduling of searching engines, and realize intelligent scheduling of searching engines. We have developed an intelligent, personal information agent on the basis of studying and making use of the technology of home and oversea web searching.In this paper, our work mainly focuses on the following aspects:1. Study comparatively wholly the domain knowledge of information agent involves, and introduce in detail information filtering, interest learning, and some technologies of meta-searching engine.2. Study comparatively profoundly the theory of reinforce leaning, its main algorithms, and methods of improve its velocity, make use of modeling method of reinforce learning to build a scheduling model of searching engines, and gets better results.3. Make use of the technology of information filtering based on TFIDF to filter the results returned by the searching engine.4. Adopt the method of hidden interest learning based on VSM, and it can learn user's interests without users' direct guidance.Our information agent consists of four main function modules: user interface agent, searching agent, interest-learning agent, and result-disposing agent. Four modules are independent, and collaborate with each other. They are integrated as a whole, and realize better the intelligence and self- adaptability of the system.