Title: Dynamically constructing user profiles with similarity-based online incremental clustering

Authors: Roman Y. Shtykh, Qun Jin

Addresses: Networked Information Systems Laboratory, Faculty of Human Sciences, Waseda University, Japan. ' Networked Information Systems Laboratory, Faculty of Human Sciences, Waseda University, Japan

Abstract: User profiling is a widely used technique to analyse and store user interests and preferences to apply this knowledge to improve user experiences with information systems. In this research paper, we present an approach for dynamically constructing user profiles, particularly from uniform relevance feedback in information-seeking activities. We propose an inference method for user interests, which we call High-Similarity Sequence Data-Driven (H2S2D) clustering and discuss its peculiarities and show its superiority for the creation of high-quality concepts, which are the elementary constituents of user profiles. To reflect the volatility of user interests and emphasise the steadiness of persistent preferences, we adopt recency, frequency and persistency as the three main criteria for multi-layered dynamic profile construction and update.

Keywords: user models; multi-layered user profiles; interest dynamics; relevance feedback; high similarity sequence data-driven clustering; incremental clustering; user profiling; user preferences.

DOI: 10.1504/IJAIP.2009.026760

International Journal of Advanced Intelligence Paradigms, 2009 Vol.1 No.4, pp.377 - 397

Published online: 25 Jun 2009 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article