Broadway: A Case-Based Browsing Advisor for the Web

Abstract

The World Wide Web (WWW) is an hypermedia of heterogeneous and dynamic documents, frequently referred to be the world wide digital library. This virtual space is growing more and more every day, offering to the user a huge amount of data. Two kinds of tasks can be handled to locate a relevant document through this space: querying and browsing. Querying is appropriate when the user has a clear goal which should usually be expressed through a list of keywords. Different servers on the WWW (such as Yahoo, Lycos, Altavista) can be then used to retrieve matching documents based on their indexing capability. Browsing is well suited when the user cannot express his goal explicitly or when query formulation by keywords is not adequate. Then, the user must navigate through this space, moving from one node to another, looking for a relevant document. These two tasks can be mixed so that querying gives a list of reasonable starting points for browsing.

However, the huge size and the structure of this space make difficult the indexing of the documents required by querying access methods and could disorient the user during a browsing session. This poster focuses on the assistance given to a group of users during their browsing session, and more precisely on the design of a browsing advisor or recommendation system. A browsing advisor is able to follow the user during a browsing session to infer his goal, and then must advise him of potentially relevant documents to visit next. Three main approaches for recommendation computation can be distinguished: content-based recommendation, profile-similarity based recommendation and behaviour-similarity based recommendation.