Towards Diversified Recommender Systems - RMGS

Description

Personalized recommender systems are playing an important role in modern society to reduce information/choice overload, satisfy personalized demands of users, and increase the revenue of service/product providers. Despite being helpful, recommender systems are criticized as a primary cause of information “filter bubble”, where users are trapped in an algorithm-created bubble without the opportunity to view diverse contents. This can lead to severely negative consequences like fake news proliferation and political polarization. Although recent studies have managed to improve the recommendation diversity and break filter bubbles, there still lack general solutions to address the accuracy-diversity dilemma.In this project, we will conduct systematic investigations on diversifying personalized recommendations and breaking filter bubbles in recommender systems. We plan to diversify recommended items against filter bubbles. Improving the recommendation diversity is a silver bullet, however, will induce the accuracy diversity dilemma. To solve this problem, we will build a model-agnostic multi-objective framework and treat diversity as one of the objectives. Moreover, we will also investigate the exploitation-exploration trade-off of contextual multi-armed bandit to solve the accuracy-diversity dilemma.