CIRS: Bursting Filter Bubbles by Counterfactual Interactive Recommender System

Abstract

While personalization increases the utility of recommender systems, it also brings the issue of filter bubbles. E.g., if the system keeps exposing and recommending the items that the user is interested in, it may also make the user feel bored and less satisfied. Existing work studies filter bubbles in static recommendation, where the effect of overexposure is hard to capture. In contrast, we believe it is more meaningful to study the issue in interactive recommendation and optimize long-term user satisfaction. Nevertheless, it is unrealistic to train the model online due to the high cost. As such, we have to leverage offline training data and disentangle the causal effect on user satisfaction.To achieve this goal, we propose a counterfactual interactive recommender system (CIRS) that augments offline reinforcement learning (offline RL) with causal inference. The basic idea is to first learn a causal user model on historical data to capture the overexposure effect of items on user satisfaction. It then uses the learned causal user model to help the planning of the RL policy. To conduct evaluation offline, we innovatively create an authentic RL environment (KuaiEnv) based on a real-world fully observed user rating dataset. The experiments show the effectiveness of CIRS in bursting filter bubbles and achieving long-term success in interactive recommendation. The implementation of CIRS is available via https://github.com/chongminggao/CIRS-codes.

A Comparative Analysis of Bias Amplification in Graph Neural Network Approaches for Recommender Systems

“…According to this regulation, AI systems should follow EU fundamental rights such as the right not to be discriminated against, respecting individuals' private life, and personal data protection [7]. Moreover, biased results in RSs can cause user dissatisfaction [38].…”

KuaiRand: An Unbiased Sequential Recommendation Dataset with Randomly Exposed Videos

“…The unbiased sequential data enables us to carry out the unbiased offline evaluation [20,29], thus facilitating the research of debiasing [3] in large-scale recommendation scenarios. Furthermore, with its distinctive features, KuaiRand can naturally facilitate a variety of other recommendation tasks, such as interactive recommendation [4,27], long sequential behavior modeling [17,18], and multi-task learning [2,28].…”

“…An interactive recommender system (IRS) is usually formulated as a decision-making process to pursue long-term success in recommendation. IRS built on reinforcement learning has shown superiority to the traditional supervised frameworks [4,26,27]. KuaiRand contains the sequential logs generated from different policies that can naturally support the research on IRSs with the help of OPE and user simulation techniques [5,10].…”

“…The biased data has catastrophic effects on researchers and practitioners of recommender systems. It leads to inconsistent online and offline results of the model [22] and gives rise to filter bubbles [4], which can incur heavy losses for the companies and hinder the development of recommender systems.…”

Alleviating Matthew Effect of Offline Reinforcement Learning in Interactive Recommendation

“…In recommendation, it means that the popular items or categories in previous data will get larger opportunities to be recommended later, whereas the unpopular ones get neglected. This is catastrophic since users desire diverse recommendations and the repetition of certain contents will incur the filter bubble issue, which in turn hurts users' satisfaction even though users favored them before [12,16,41,58]. We will show the Matthew effect in the existing offline RL-based recommender (Fig.…”