An Agent-based System For Intelligent Information Filtering

With the fast development of Internet, the online inf6rmation is growing and changing rapidly. Nowadays people search information mainly by means of traditional search engines. However, there exist some challenges to the search engines for the form of online information isn't identical, for the knowledge hierarchy of users differs from each other, and for the queries vary with regard to each user. These search engines have the following distinct demerits. Firstly, as for the query composed of keywords, people are sure to get the same searching results if the inputting keywords are identical. Secondly, since the information on the Web is mutative, people have to inquire about the same topic in search engines frequently to keep pace with it. Thirdly, the coverage of any one engine is limited while its returning results are too many. Lastly, current search engines don't support the function of information sharing for the different users with similar interests. In this paper, we explore the notion of combining software agent and data mining technology to establish an agent-based system for intelligent information filtering. As for a single personal assistant agent (PAA), it can build user's interest profiles automatically by analyzing user's history visiting records and browser bookmark. In the meantime, PAA submits user's query to several search engines in parellel, and then collects the searching results. By merging the searching results from different search engines and compares them with user's interest profiles, web pages related to user's interests are recommended to users. Moreover, PAA tracks user's browsing behavior to obtain the relevancy information about the recommended pages, so we can revise user's interest profiles to keep up with user's interest changing and find out which search engine is more credible for one special topic that is in favor of the next searching for the same topic. In addition, PAA organizes user's favorite web sites, pages in a directory-tree bookmark, and with this method, the users can manage it more easily. This paper then expands the single agent-based system to multi-agent system in which we mainly provides the function of collaborative filtering. Users can log on the server by means of www while submitting their bookmark information. Thus, users can access to their information anywhere and search information manually or ask for recommendation. Furthermore, PAA can collaborate to filter information by agent communication language and communication protocol so that different users with similar interests can share information. In conclusion, the main contributions of this paper are as follows: (I) By adopting the agent-oriented technology, I construct the model of PAA composed of Information Filtering Model, Corrtmunication Model and Decision Model. (2) By analyzing user's history and browser bookmark, the user's interest profiles are automatically built that reflect user's mutative interests. (3) Design the merging algorithm that is used to compute the weight of each item and evaluate the credibility of each search engine with regard to some topics. (4) Propose the strategy of recommending relevant information by comparing the user's interest profiles and the merging results from several search engines. (5) Construct the collaborative filtering model for MAS, while proposing the static filtering strategy and dynamic filtering strategy.