A Reinforcement Profile Learning Agent for Documents Filtering

Y.M. Al Murtadha, N. bin Sulaiman, and Z. Muda (Malaysia)


Software Agent, Information Filtering, Profile Learning, Term Weighting


People spend an increasing amount of time using the web and the information sources. Some significant factors of that time are spent on navigation overhead like searching for relevant information through huge of irrelevant data retrieved. This paper presents RePLS, a Content-based reinforcement profile learning agent system that learns the user’s interests by analyzing the contents of the documents, build his profiles and block the irrelevant documents by filtering the incoming documents according to the learned user’s needs. The main idea is to select- when the documents is indexed, stemmed, represented and selected as relevant the best terms representing the profile which help to discriminate between profiles. The agent updates the profile with every selected document to meet the user interests. The learning mechanism used by the agents is relevance feedback and reinforcement learning. Agent Foundation Classes suite AFC is used to build the proposed agents under RETSINA architecture. RePLS efficiency is measured by using the linear utility T11SU described by TREC evaluation which shows a better documents categorization than previous profile learning methods, namely query zoning and incremental profile learning based on the reinforcement method. The used documents are TREC 2002 filtering track collections.

Important Links:

Go Back