Enhanced Doubly Robust Learning for Debiasing Post-Click Conversion Rate Estimation

Abstract

Post-click conversion, as a strong signal indicating the user preference, is salutary for building recommender systems. However, accurately estimating the post-click conversion rate (CVR) is challenging due to the selection bias, i.e., the observed clicked events usually happen on users' preferred items. Currently, most existing methods utilize counterfactual learning to debias recommender systems. Among them, the doubly robust (DR) estimator has achieved competitive performance by combining the error imputation based (EIB) estimator and the inverse propensity score (IPS) estimator in a doubly robust way. However, inaccurate error imputation may result in its higher variance than the IPS estimator. Worse still, existing methods typically use simple model-agnostic methods to estimate the imputation error, which are not sufficient to approximate the dynamically changing model-correlated target (i.e., the gradient direction of the prediction model). To solve these problems, we first derive the bias and variance of the DR estimator. Based on it, a more robust doubly robust (MRDR) estimator has been proposed to further reduce its variance while retaining its double robustness. Moreover, we propose a novel double learning approach for the MRDR estimator, which can convert the error imputation into the general CVR estimation. Besides, we empirically verify that the proposed learning scheme can further eliminate the high variance problem of the imputation learning. To evaluate its effectiveness, extensive experiments are conducted on a semi-synthetic dataset and two real-world datasets. The results demonstrate the superiority of the proposed approach over the state-of-the-art methods. The code is available at https://github.com/guosyjlu/MRDR-DL. CCS CONCEPTS• Information systems → Recommender systems.

Doubly Robust Collaborative Targeted Learning for Debiased Recommendations

“…MovieLens 100K 5 (ML-100K) is a dataset of 100,000 MNAR ratings from 943 users and 1,682 movies collected from movie recommendation ratings. Following the standard procedure of previous studies [27,36,22,7], we performed the following preprocessing steps to carry out the semi-synthetic experiments.…”

“…[36] proposes a doubly robust joint learning approach that improves the IPS method. A series of variants of DR methods are developed, such as more robust doubly robust (MRDR) method [7] and multi-task learning [43]. In addition, [2,13,5,37] design new debiasing algorithms via using a small uniform dataset.…”

“…Throughout, we adopt the RCT-free technique to estimate the propensity, which differs from the existing studies (e.g., Naive Bayes). In this section, following the previous studies [27,36,22,7], we aim to answer the following research question (RQ) on the semi-synthetic dataset: RQ1. Do the proposed TMLE estimators in estimating the ideal loss have both the statistical properties of relatively lower bias and variance in the presence of selection bias?…”

“…Causal inference methods are increasingly being employed in recommender system (RS) [41], such as post-view click-through rate prediction [7], post-click conversion rate prediction [7,43], and uplift modeling [23,25,26]. Recommendation based on causality has shown its great potential in both numeric experiments and theoretical analyses in various literature [4,36,41].…”

“…In RS, data missing not at random (MNAR) is a common problem, which can be interpreted from a causal perspective as "what would the feedback be, if recommending an item to a user", requiring to answer the counterfactual problem. To address this question, many methods have been proposed, such as inverse propensity score (IPS) [24,27], self-normalized inverse propensity score (SNIPS) [27,30], error imputation based (EIB) [8,28] learning, doubly robust (DR) [36,7,5,37] learning. Among them, the DR method and its variants show superior performance.…”