User Modelling on Semi-Structured Documents

Download

Postscript Download (497Kb)

Description/Abstract

We present a new approach that makes use of the embedded structural information of the documents that a user frequently refers to for deriving a personalized concept hierarchy and for identifying user preferences concerning document searching and browsing. In contrast with conventional methods that ignore the distribution of structural elements, our approach accepts semantic clues defined in the structural tags, so that it takes full advantage of the textual and structural information. Formal concept analysis theory is applied to define semantic relationships among the concepts, and reinforcement learning is employed to unobtrusively adapt to individualized information needs. Given a user’s query, the personal ontology and user model maximize their knowledge bases to present the relevant documents in a prioritized order, one which the user prefers to browse. We demonstrate the practicability of our approach through two experiments. The former showed significantly improved performance results compared to that of a flat vector model. The second experiment compared the performance accuracy of the structure elements exploited by four users and thereby demonstrated the diversity of user preferences.