Intelligent Information Retrieval System Based On Multi-Agent

In recent years, as a branch of artificial intelligent (AI) the intelligent information retrieval has developed rapidly. With the development and the spread of Internet, more and more users use search engines to search for information. Although the development of the search engine has become mature, yet when using it, people find it more and more difficult to look for the information needed for their own. Faced on the current situation of network information service, people are searching for a kind of service model that the required information for users is recommended to them actively. That is the personalized active information service. The intelligent agent technology plays an important part in carrying on the personalized active information service.The paper has systematically studied the key technology dealing with the agent-based personalized active information service. Based on the system of Intelligent Information Retrieval Agent (IIR Agent) issued by Hsieh Chang Tu et al, we has made some improvements that three main functions of user interest modeling, information searching and information filtering are taken into separate agent. The system modified is called for Intelligent Information Retrieval based on Multi-Agent (IIR M-Agent). Agent community is discussed first. Agents of IIR M-Agent are also identified so that they can be designed and implemented separat(?)ly. How these agent collaborate to make some functions, such as intelligent search, auto-notification, navigation guide are illustrated and discussed.Three main parts in the system, which are user agent, information search agent and information filtering agent are particularly discussed. We should form and train the user agent through the expression of information requirement and information feedback by the user. Traditional user interest model adopts term frequency to express user interest on the binary group (term, weight). We have modified traditional method of term frequency. Taking frequency of words in a document into consideration, we also notice that one hand, different position of words indicates different importance of words; another hand, quoted or not in hyperlink can weigh importance of the document. Term frequency modified makes choice of interest terms more objective and exact and removes the words of high frequency but little meanings and relevance. We don't only improve traditional user interest model, but also propose a new one. Using term frequency modified to weigh user's interest, the system refers to the notion of term novelty. So user interest is expressed by the triple group (term, weight, novelty), which makes interest terms update more exact. Additionally, the user agent uses machine learning methods to study and update user interest.The information search agent through its inquiry agent connects Internet search engines, not only realizes meta-search, but also makes self-search on the web, when the recalls couldn't meet the needs of the user. The search algorithm starts its searching from the recalls of inquiry agent to reduce the range of searching and increase the speed of searching. Additionally, searching information on Internet, robot limits searching depth and also makes it dynamically adjusted. One hand, due to limited searching depth, search can't run into unlimited hyperlinks. Another, with dynamically adjusted searching depth, agent can stop itself on these links with irrelevant information, but continue to deeply search on those links with much relevant information. This makes searching inquality and efficiency high. According to the use ready-made information resources, information filtering agent analyses the user's favorite and adopts vector space method to carry out the personalized information filtering. Finally, the paper designed the entire structure of intelligent agent to ensure the close cooperation of the three models so that the information retrieval can be fully realized personalized, actively and intelligently.