Authors:
Rafael Alonso
;
Philip Bramsen
and
Hua Li
Affiliation:
SET Corporation and a SAIC Company, United States
Keyword(s):
User Modeling, Machine Learning, Implicit Feedback, User.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Communication, Collaboration and Information Sharing
;
Intelligent Information Systems
;
Knowledge Management and Information Sharing
;
Knowledge-Based Systems
;
Social Networks and the Psychological Dimension
;
Symbolic Systems
Abstract:
We describe an innovative approach for incrementally learning user interests from multiple types of user behaviours or events. User interests are reflected in the concepts and their relations contained in these events. The concepts and relations form the structural elements of a user interest model. The relevance of each structural element is signified by a weight. Our user modeling algorithm builds dynamic user interest model with two concurrent processes. One process grows the model by intelligently incorporating concepts and relationships extracted from user events. Another process adapts the weights of these model elements by applying a novel combination of two mechanisms: reinforcement and forgetting, both important in modulating user interests. Our modeling algorithm supports incremental and real time modeling, and readily extends to new types of user events. One interesting application of user interest models is to identify a virtual interest group (VIG), which is an ordered
set of other system users who exhibit interests similar to those of the target user. As a result, we can evaluate our user modeling algorithm through a VIG identification task. In a formative NIST evaluation using intelligence analysts, we achieved 95% VIG identification precision and recall.
(More)