Odysseus

CLARITY is carrying out research in conjunction with an Irish software company on a project involving search and recommendation techniques. The project is based around a platform which connects consumers to suppliers of services using a wide range of social network and web 2.0 technologies.

As part of this service supplier platform, various different types of recommendations need to be made to the consumer. Since this platform uses hand-built profiles, social networks and other web 2.0 technologies, the range of features that can be used to influence recommendations are many. Also, different features need to be used in different strengths in different recommendation scenarios. Deciding on which features and the effect these features have on overall recommendations is a large research challenge.

In this project, the student will develop and test automatic feature-weight learning techniques which will be incorporated into the recommendation algorithms. The student will apply machine learning techniques to learn feature selection policies and feature-weight settings that deliver optimum performance in a variety of recommendation scenarios.

The student will benefit from access to a number of CLARITY researchers who are working on related issues. CLARITY will also provide access to real-world data for testing and development. This project allows the student to work on research solutions which will be incorporated into an industry platform, giving the student a valuable and unique insight into aspects of both research and industry.

Workplan

Week 1: Background Research. The student will look into related research in the areas of recommendation. The student will also get up-to-speed with other developments in this research-industry partnership.

Weeks 2-3: Automatic feature selection. The student will develop techniques to automatically learn which features are most suitable to drive specific recommendations.

Weeks 4-5: Automatic feature-weighting. Once the selection of features is complete the student will concentrate on automatically learning how much influence the selected features should have on the recommendation results.

Weeks 6-8: Recommender Viewer/Manager Interface. The student will develop a tool for testing, managing and viewing results of their automatic feature weight learning strategies.

Weeks 9-10: Testing and fine tuning. The student will test their strategies across a wide range of data and fine tune the parameters of their generation techniques.

Supporting Material:

B. SMYTH. "Case-based recommendation" In The Adaptive Web (2007), P. Brusilovsky, A. Kobsa, and W. Nejdl, Eds., Springer, pp. 342–376.

M. SALAMO, L. MCGINTY, B. SMYTH. "Knowledge discovery from user preferences in conversational recommendation" In Proceedings of the 9th European Conference on Principles and Practice of Knowledge Discovery in Databases (PKDD 2005) (2005), A. Jorge, L. Torgo, P. Brazdil, R. Camacho, and J. Gama, Eds., vol. 3721 of Lecture Notes in Computer Science, Springer, pp. 228–239. Porto, Portugal.