Some existing estimators and regularizers attempt to achieve unbiased estimation to improve the predictive performance.

In recommender systems, the collected data used for training is always subject to selection bias, which poses a great challenge for unbiased learning.

Neural Information Processing Systems http://nips.cc/

In recommender systems, the collected data used for training is always subject to selection bias, which poses a great challenge for unbiased learning. Previous studies proposed various debiasing methods based on observed user and item features, but ignored the effect of hidden confounding. To address this problem, recent works suggest the use of sensitivity analysis for worst-case control of the unknown true propensity, but only valid when the true propensity is near to the nominal propensity within a finite bound. In this paper, we first perform theoretical analysis to reveal the possible failure of previous approaches, including propensity-based, multi-task learning, and bi-level optimization methods, in achieving unbiased learning when hidden confounding is present. Then, we propose a unified multi-task learning approach to remove hidden confounding, which uses a few unbiased ratings to calibrate the learned nominal propensities and nominal error imputations from biased data. We conduct extensive experiments on three publicly available benchmark datasets containing a fully exposed large-scale industrial dataset, validating the effectiveness of the proposed methods in removing hidden confounding.

Causality-based recommendation systems focus on the causal effects of user-item interactions resulting from item exposure (i.e., which items are recommended or exposed to the user), as opposed to conventional correlation-based recommendation. They are gaining popularity due to their multi-sided benefits to users, sellers and platforms alike. However, existing causality-based recommendation methods require additional input in the form of exposure data and/or propensity scores (i.e., the probability of exposure) for training. Such data, crucial for modeling causality in recommendation, are often not available in real-world situations due to technical or privacy constraints. In this paper, we bridge the gap by proposing a new framework, called Propensity Estimation for Causality-based Recommendation (PropCare). It can estimate the propensity and exposure from a more practical setup, where only interaction data are available any ground truth on exposure or propensity in training and inference. We demonstrate that, by relating the pairwise characteristics between propensity and item popularity, PropCare enables competitive causality-based recommendation given only the conventional interaction data. We further present a theoretical analysis on the bias of the causal effect under our model estimation. Finally, we empirically evaluate PropCare through both quantitative and qualitative experiments.

Neural Information Processing Systems http://nips.cc/

Unlike previous approaches that assume unconfoundedness, i.e., no unobserved confounders affected both treatment assignment and outcomes, we calibrate policy learning for realistic violations of this unverifiable assumption with uncertainty sets motivated by sensitivity analysis in causal inference. Our framework for confounding-robust policy improvement optimizes the minimax regret of a candidate policy against a baseline or reference "status quo" policy, over an uncertainty set around nominal propensity weights.

As AI systems become more capable, deceptive behaviors can undermine evaluation and mislead users at deployment. Recent work has shown that lie detectors can accurately classify deceptive behavior, but they are not typically used in the training pipeline due to concerns around contamination and objective hacking. We examine these concerns by incorporating a lie detector into the labelling step of LLM post-training and evaluating whether the learned policy is genuinely more honest, or instead learns to fool the lie detector while remaining deceptive. Using DolusChat, a novel 65k-example dataset with paired truthful/deceptive responses, we identify three key factors that determine the honesty of learned policies: amount of exploration during preference learning, lie detector accuracy, and KL regularization strength. We find that preference learning with lie detectors and GRPO can lead to policies which evade lie detectors, with deception rates of over 85\%. However, if the lie detector true positive rate (TPR) or KL regularization is sufficiently high, GRPO learns honest policies. In contrast, off-policy algorithms (DPO) consistently lead to deception rates under 25\% for realistic TPRs. Our results illustrate a more complex picture than previously assumed: depending on the context, lie-detector-enhanced training can be a powerful tool for scalable oversight, or a counterproductive method encouraging undetectable misalignment.