Personal adaptive web agent for information filtering

Thumbnail Image
Khan, Imran
Journal Title
Journal ISSN
Volume Title
This thesis presents one possible design solution for a Web agent which reduces information overload for Web users by autonomously retrieving documents that the user is interested in. The document provides a brief primer on software agents, presents a design solution for an unique model for a Personal Adaptive Web (PAW) Agent, and simulates the individual components of the model. The PAW agent is a personal assistant which learns different categories of Web documents that the user is interested in, then finds and suggests new similar documents to the user. Similar document vector representations and Inverse Document Frequency Weight (IDFW) techniques are employed as in other Information Retrieval Agents. However, this approach is otherwise quite new and produces an agent that is much more autonomous than the others. In fact, the only piece of information that must be supplied to the agent is the number of categories that the user would like to establish. The PAW agent performs seven subtasks to achieve its goal. It (i) monitors the user while she is browsing the Web, (ii) etermines the relevant documents that the user visits, (iii) textually analyses the relevant documents to obtain document vectors using a modified form of the IDFW technique, (iv) classifies the document vectors into categories using unsupervised competitive learning, (v) scans the Web for similar documents, (vi) classifies the new document vectors using the trained neural network, and (vii) decides whether the new documents should be referred to the user. (Abstract shortened by UMI.)