A clickstream-based collaborative filtering personalization model: towards a better performance

  • Authors:

  • Dong-Ho Kim;Vijayalakshmi Atluri;Michael Bieber;Nabil Adam;Yelena Yesha

  • Affiliations:

  • Rutgers University;Rutgers University;New Jersey Institute of Technology;Rutgers University;University of Maryland at Baltimore County

  • Venue:
  • Proceedings of the 6th annual ACM international workshop on Web information and data management
  • Year:
  • 2004

Quantified Score

Hi-index 0.00

Visualization

Abstract

In recent years, clickstream-based Web personalization models for collaborative filtering recommendation have received much attention mainly due to their scalability [10,16,19]. The common personalization models are the Markov model, (sequential) association rule, and clustering. These models have shown strengths and weaknesses in their performance: for instance, the Markov model has higher precision and lower recall than (sequential) association rule and clustering, and vice versa [22]. In order to address the trade-off relationship of precision and recall, some study has combined two or more different models [22] or applied multi-order models [24,27]. The performance increases by these models, however, are at best marginal and still there is room for improving the performance because of their first order (one model type) application in making recommendation. We propose a new hybrid model for improving the performance, especially recall. The proposed hybrid model applies four prediction models - the Markov model, sequential association rule, association rule, and a default model [1,17] - in tandem in their precision order. We evaluated our model with Web usage data, and the result is promising.