Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Separating and Learning Latent Confounders to Enhancing User Preferences Modeling

Hangtong Xu, Yuanbo Xu, Yongjian Yang

TL;DR

This work tackles bias and confounding in recommender systems by treating the former recommender as a proxy for unmeasured confounders and proposing SLFR to disentangle confounders from true user preferences in a latent space. It introduces a two-stage variational approach where stage one learns user- and item-specific confounder representations via two VAEs with a modified KL objective to promote independence, and stage two estimates counterfactual true preferences by removing confounder effects in a Matrix Factorization scoring framework, optimizing the combined loss $\,\mathcal{L}_{SLFR} = \,\mathcal{L}_{normal} + \gamma \\mathcal{L}_{bias}$. The method is demonstrated on five real-world datasets, showing consistent improvements over strong baselines and compatibility with backbones like MF, LightGCN, DIN, and SASRec. By explicitly separating latent confounders from user preferences, SLFR provides a general debiasing strategy for more accurate true-preference modeling in recommender systems.

Abstract

Recommender models aim to capture user preferences from historical feedback and then predict user-specific feedback on candidate items. However, the presence of various unmeasured confounders causes deviations between the user preferences in the historical feedback and the true preferences, resulting in models not meeting their expected performance. Existing debias models either (1) specific to solving one particular bias or (2) directly obtain auxiliary information from user historical feedback, which cannot identify whether the learned preferences are true user preferences or mixed with unmeasured confounders. Moreover, we find that the former recommender system is not only a successor to unmeasured confounders but also acts as an unmeasured confounder affecting user preference modeling, which has always been neglected in previous studies. To this end, we incorporate the effect of the former recommender system and treat it as a proxy for all unmeasured confounders. We propose a novel framework, Separating and Learning Latent Confounders For Recommendation (SLFR), which obtains the representation of unmeasured confounders to identify the counterfactual feedback by disentangling user preferences and unmeasured confounders, then guides the target model to capture the true preferences of users. Extensive experiments in five real-world datasets validate the advantages of our method.

Separating and Learning Latent Confounders to Enhancing User Preferences Modeling

TL;DR

This work tackles bias and confounding in recommender systems by treating the former recommender as a proxy for unmeasured confounders and proposing SLFR to disentangle confounders from true user preferences in a latent space. It introduces a two-stage variational approach where stage one learns user- and item-specific confounder representations via two VAEs with a modified KL objective to promote independence, and stage two estimates counterfactual true preferences by removing confounder effects in a Matrix Factorization scoring framework, optimizing the combined loss . The method is demonstrated on five real-world datasets, showing consistent improvements over strong baselines and compatibility with backbones like MF, LightGCN, DIN, and SASRec. By explicitly separating latent confounders from user preferences, SLFR provides a general debiasing strategy for more accurate true-preference modeling in recommender systems.

Abstract

Recommender models aim to capture user preferences from historical feedback and then predict user-specific feedback on candidate items. However, the presence of various unmeasured confounders causes deviations between the user preferences in the historical feedback and the true preferences, resulting in models not meeting their expected performance. Existing debias models either (1) specific to solving one particular bias or (2) directly obtain auxiliary information from user historical feedback, which cannot identify whether the learned preferences are true user preferences or mixed with unmeasured confounders. Moreover, we find that the former recommender system is not only a successor to unmeasured confounders but also acts as an unmeasured confounder affecting user preference modeling, which has always been neglected in previous studies. To this end, we incorporate the effect of the former recommender system and treat it as a proxy for all unmeasured confounders. We propose a novel framework, Separating and Learning Latent Confounders For Recommendation (SLFR), which obtains the representation of unmeasured confounders to identify the counterfactual feedback by disentangling user preferences and unmeasured confounders, then guides the target model to capture the true preferences of users. Extensive experiments in five real-world datasets validate the advantages of our method.
Paper Structure (4 sections, 1 equation, 2 figures)

This paper contains 4 sections, 1 equation, 2 figures.

Table of Contents

  1. Introduction
  2. Problem Formulation
  3. Usual View of Recommendation Problem
  4. General Debiasing Recommendation Problem

Figures (2)

  • Figure 1: Causal Graph Views of Recommendation Problem. (a): General view without unmeasured confounders; (b): The view that takes into account unmeasured confounders and former recommender system; (c): The view uses the former recommender system as the proxy for confounders.
  • Figure 2: Analysis of the Reasoner dataset. After converting feedback to implicit feedback, we summarize as follows, left: false positive and false negative percent in Reasoner; right: percentage difference between positive and negative feedback for the Kuairec and Reasoner datasets.