UBC Theses and Dissertations
Recommending user-generated item lists Liu, Yidan
Nowadays, more and more websites are providing users with the functionality to create item lists. For example, in Yelp, users can create restaurant lists he/she likes. In Goodreads, users can create booklists consisting of books they like. These user-generated item lists complement the main functionality of the corresponding application and intuitively become an alternative way for users to browse and discover interesting items. Existing recommender systems are not designed for recommending user-generated item lists directly. In this work, we study properties of these user-generated item lists and propose two different models for recommending user-generated lists. We first model the problem as a recommendation problem and propose a Bayesian ranking model, called Lire. The proposed model takes into consideration users' previous interactions with both item lists and with individual items. Furthermore, we propose in Lire a novel way of weighting items within item lists based on both positions of items, and personalized list consumption pattern. Through extensive experiments on a real item list dataset from Goodreads, we demonstrate the effectiveness of our proposed Lire model. We then summarize the lessons learned from the recommendation model and modelled the problem as an optimization problem. We show that the list recommendation optimization can be reduced to Maximum k-Coverage Problem, which is an NP-hard problem. We derive three different problem variants and propose algorithms for each of them. We then conduct experiments to compare their results. Lessons learned and possible application scenarios are also discussed in the hope of helping us and more people improve the work. Furthermore, we propose several potential directions for future work.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International