Contributions

The work presented in this thesis uses techniques from the field of Interface Agents to solve the problem of information filtering. Interface Agents are computer programs that learn the tastes and preferences of users and automate repetitive and predictable computer related tasks for them [28]. An information filtering agent learns the information related preferences of the user and automates some of the filtering tasks for the user. The contributions made by this thesis are as follows:

• This thesis refines the statement of the personalized information filtering problem. Personalized filtering systems must be specialized to user interests, adaptive to preference changes and explore newer information domains. While the first two criteria have been addressed earlier [39][7], exploration has not been sufficiently emphasized before.

• The vector space model for document representation has been generalized for use with documents containing more than just text. The distance metric for computation of document score and the effect of relevance feedback has also been generalized.

• Programming by demonstration [8] is proposed as an additional method of providing feedback to the filtering system. The user is no longer constrained to providing feedback only to the documents retrieved by the profiles, but can also pro-actively train the system using documents she found.

• This thesis validates the use of Genetic Algorithms [18] for modeling adaptive and exploratory behavior in Filtering systems.

• Experimental results show that using only relevance feedback is sufficient for specializing to user interests, but not satisfactory for modeling adaptive behavior.