Investigating and Evaluating Exploratory Recommender Systems

From PAWS Lab
Jump to: navigation, search

Overview

This project is focused on a popular category of interactive Recommender Systems (RecSys) known as user-controlled RecSys. The idea of user-controlled RecSys is to engage the user in controlling some aspects of the recommendation process such as choosing peers for collaborative filtering, fusing several recommender algorithms, or deducing user current preferences. While the traditional data-driven approaches estimate these parameters from data trying to optimize the results across many users and contexts, in many cases user data is not sufficient to make a reliable “guess”. As demonstrated by many existing examples, leaving control over some aspects of the recommendation process to the user avoids guessing and leads to a better user experience in each specific case. A good example of simple user control is the carousel-based interface in some industrial recommenders. This approach leaves it to the user to choose which category of items they are currently interested in instead of the traditional attempt to “guess” it with too little information and offer users a single ranked list.

The target of this proposal is algorithms for user-controlled recommendation. While recent work on user-controlled RecSys demonstrated the value of this approach, this stream of research predominantly used simple recommer algorithms. In other words, the focus was on the user at the expense of the AI. In this project, we want to restore the balance by developing a better generation of recommender algorithms that can benefit from user control.

Fundning

The project is supported by Amazon Research Award to Peter Brusilovsky

Publications

  • Rahdari, B., Brusilovsky, P., and Sabet, A. J. (2021) Controlling Personalized Recommendations in Two Dimensions with a Carousel-Based Interface. In: Proceedings of Joint Workshop on Interfaces and Human Decision Making for Recommender Systems (IntRS’21) at 2021 ACM Conference on Recommender Systems (RecSys’21), September 25, 2021, CEUR, pp. 112-122.
  • Rahdari, B., Brusilovsky, P., and Sabet, A. J. (2021) Connecting Students with Research Advisors Through User-Controlled Recommendation. In: Proceedings of Fifteenth ACM Conference on Recommender Systems, Amsterdam, Netherlands, 27 September 2021 - 1 October 2021, ACM, pp. 745-748.
  • Rahdari, B. and Brusilovsky, P. (2022) Simulation-Based Evaluation of Interactive Recommender Systems. In: Proceedings of 9th Joint Workshop on Interfaces and Human Decision Making for Recommender Systems co-located with the 16th ACM Conference on Recommender Systems (RecSys 2022), Seattle, WA, September 22, 2022, CEUR, pp. 122-136.
  • Rahdari, B., Brusilovsky, P., He, D., Thaker, K., Luo, Z., and Lee, Y. J. (2022) Helper: an interactive recommender system for ovarian cancer patients and caregivers. In: Proceedings of 16th ACM Conference on Recommender Systems, Seattle, WA, ACM, pp. 644-647.
  • Rahdari, B., Brusilovsky, P., and Kveton, B. (2022) Towards Increasing the Coverage of Interactive Recommendations. In: Proceedings of The Thirty-Fifth International Florida Artificial Intelligence Research Society Conference (FLAIRS-35), Jensen Beach, FL, May 15-18, 2022, Association for the Advancement of Artificial Intelligence.
  • Rahdari, B., Kveton, B., and Brusilovsky, P. (2022) The Magic of Carousels: Single vs. Multi-List Recommender Systems. In: Proceedings of Proceedings of the 33rd ACM Conference on Hypertext and Social Media, Barcelona, Spain, June 28-July 1, 2022, ACM, pp. 166-174.
Retrieved from "http://adapt2.sis.pitt.edu/w/index.php?title=Investigating_and_Evaluating_Exploratory_Recommender_Systems&oldid=4599"