Stochastic learning to rank (LTR) is a recent branch in the LTR field that concerns the optimization of probabilistic ranking models. Their probabilistic behavior enables certain ranking qualities that are impossible with deterministic models. For example, they can increase the diversity of displayed documents, increase fairness of exposure over documents, and better balance exploitation and exploration through randomization. A core difficulty in LTR is gradient estimation, for this reason, existing stochastic LTR methods have been limited to differentiable ranking models (e.g., neural networks). This is in stark contrast with the general field of LTR where Gradient Boosted Decision Trees (GBDTs) have long been considered the state-of-the-art. In this work, we address this gap by introducing the first stochastic LTR method for GBDTs. Our main contribution is a novel estimator for the second-order derivatives, i.e., the Hessian matrix, which is a requirement for effective GBDTs. To efficiently compute both the first and second-order derivatives simultaneously, we incorporate our estimator into the existing PL-Rank framework, which was originally designed for first-order derivatives only. Our experimental results indicate that stochastic LTR without the Hessian has extremely poor performance, whilst the performance is competitive with the current state-of-the-art with our estimated Hessian. Thus, through the contribution of our novel Hessian estimation method, we have successfully introduced GBDTs to stochastic LTR.
The powerful generative abilities of large language models (LLMs) show potential in generating relevance labels for search applications. Previous work has found that directly asking about relevancy, such as ``How relevant is document A to query Q?", results in sub-optimal ranking. Instead, the pairwise ranking prompting (PRP) approach produces promising ranking performance through asking about pairwise comparisons, e.g., ``Is document A more relevant than document B to query Q?". Thus, while LLMs are effective at their ranking ability, this is not reflected in their relevance label generation. In this work, we propose a post-processing method to consolidate the relevance labels generated by an LLM with its powerful ranking abilities. Our method takes both LLM generated relevance labels and pairwise preferences. The labels are then altered to satisfy the pairwise preferences of the LLM, while staying as close to the original values as possible. Our experimental results indicate that our approach effectively balances label accuracy and ranking performance. Thereby, our work shows it is possible to combine both the ranking and labeling abilities of LLMs through post-processing.
Most Recommender Systems (RecSys) do not provide an indication of confidence in their decisions. Therefore, they do not distinguish between recommendations of which they are certain, and those where they are not. Existing confidence methods for RecSys are either inaccurate heuristics, conceptually complex or computationally very expensive. Consequently, real-world RecSys applications rarely adopt these methods, and thus, provide no confidence insights in their behavior. In this work, we propose learned beta distributions (LBD) as a simple and practical recommendation method with an explicit measure of confidence. Our main insight is that beta distributions predict user preferences as probability distributions that naturally model confidence on a closed interval, yet can be implemented with the minimal model-complexity. Our results show that LBD maintains competitive accuracy to existing methods while also having a significantly stronger correlation between its accuracy and confidence. Furthermore, LBD has higher performance when applied to a high-precision targeted recommendation task. Our work thus shows that confidence in RecSys is possible without sacrificing simplicity or accuracy, and without introducing heavy computational complexity. Thereby, we hope it enables better insight into real-world RecSys and opens the door for novel future applications.
Since its inception, the field of unbiased learning to rank (ULTR) has remained very active and has seen several impactful advancements in recent years. This tutorial provides both an introduction to the core concepts of the field and an overview of recent advancements in its foundations along with several applications of its methods. The tutorial is divided into four parts: Firstly, we give an overview of the different forms of bias that can be addressed with ULTR methods. Secondly, we present a comprehensive discussion of the latest estimation techniques in the ULTR field. Thirdly, we survey published results of ULTR in real-world applications. Fourthly, we discuss the connection between ULTR and fairness in ranking. We end by briefly reflecting on the future of ULTR research and its applications. This tutorial is intended to benefit both researchers and industry practitioners who are interested in developing new ULTR solutions or utilizing them in real-world applications.
Counterfactual learning to rank (CLTR) relies on exposure-based inverse propensity scoring (IPS), a LTR-specific adaptation of IPS to correct for position bias. While IPS can provide unbiased and consistent estimates, it often suffers from high variance. Especially when little click data is available, this variance can cause CLTR to learn sub-optimal ranking behavior. Consequently, existing CLTR methods bring significant risks with them, as naively deploying their models can result in very negative user experiences. We introduce a novel risk-aware CLTR method with theoretical guarantees for safe deployment. We apply a novel exposure-based concept of risk regularization to IPS estimation for LTR. Our risk regularization penalizes the mismatch between the ranking behavior of a learned model and a given safe model. Thereby, it ensures that learned ranking models stay close to a trusted model, when there is high uncertainty in IPS estimation, which greatly reduces the risks during deployment. Our experimental results demonstrate the efficacy of our proposed method, which is effective at avoiding initial periods of bad performance when little data is available, while also maintaining high performance at convergence. For the CLTR field, our novel exposure-based risk minimization method enables practitioners to adopt CLTR methods in a safer manner that mitigates many of the risks attached to previous methods.
The goal of this work is to help mitigate the already existing gender wage gap by supplying unbiased job recommendations based on resumes from job seekers. We employ a generative adversarial network to remove gender bias from word2vec representations of 12M job vacancy texts and 900k resumes. Our results show that representations created from recruitment texts contain algorithmic bias and that this bias results in real-world consequences for recommendation systems. Without controlling for bias, women are recommended jobs with significantly lower salary in our data. With adversarially fair representations, this wage gap disappears, meaning that our debiased job recommendations reduce wage discrimination. We conclude that adversarial debiasing of word representations can increase real-world fairness of systems and thus may be part of the solution for creating fairness-aware recommendation systems.
Optimizing recommender systems based on user interaction data is mainly seen as a problem of dealing with selection bias, where most existing work assumes that interactions from different users are independent. However, it has been shown that in reality user feedback is often influenced by earlier interactions of other users, e.g. via average ratings, number of views or sales per item, etc. This phenomenon is known as the bandwagon effect. In contrast with previous literature, we argue that the bandwagon effect should not be seen as a problem of statistical bias. In fact, we prove that this effect leaves both individual interactions and their sample mean unbiased. Nevertheless, we show that it can make estimators inconsistent, introducing a distinct set of problems for convergence in relevance estimation. Our theoretical analysis investigates the conditions under which the bandwagon effect poses a consistency problem and explores several approaches for mitigating these issues. This work aims to show that the bandwagon effect poses an underinvestigated open problem that is fundamentally distinct from the well-studied selection bias in recommendation.
Click-based learning to rank (LTR) tackles the mismatch between click frequencies on items and their actual relevance. The approach of previous work has been to assume a model of click behavior and to subsequently introduce a method for unbiasedly estimating preferences under that assumed model. The success of this approach is evident in that unbiased methods have been found for an increasing number of behavior models and types of bias. This work aims to uncover the implicit limitations of the high-level prevalent approach in the counterfactual LTR field. Thus, in contrast with limitations that follow from explicit assumptions, our aim is to recognize limitations that the field is currently unaware of. We do this by inverting the existing approach: we start by capturing existing methods in generic terms, and subsequently, from these generic descriptions we derive the click behavior for which these methods can be unbiased. Our inverted approach reveals that there are indeed implicit limitations to the counterfactual LTR approach: we find counterfactual estimation can only produce unbiased methods for click behavior based on affine transformations. In addition, we also recognize previously undiscussed limitations of click-modelling and pairwise approaches to click-based LTR. Our findings reveal that it is impossible for existing approaches to provide unbiasedness guarantees for all plausible click behavior models.
Methods for reinforcement learning for recommendation (RL4Rec) are increasingly receiving attention as they can quickly adapt to user feedback. A typical RL4Rec framework consists of (1) a state encoder to encode the state that stores the users' historical interactions, and (2) an RL method to take actions and observe rewards. Prior work compared four state encoders in an environment where user feedback is simulated based on real-world logged user data. An attention-based state encoder was found to be the optimal choice as it reached the highest performance. However, this finding is limited to the actor-critic method, four state encoders, and evaluation-simulators that do not debias logged user data. In response to these shortcomings, we reproduce and expand on the existing comparison of attention-based state encoders (1) in the publicly available debiased RL4Rec SOFA simulator with (2) a different RL method, (3) more state encoders, and (4) a different dataset. Importantly, our experimental results indicate that existing findings do not generalize to the debiased SOFA simulator generated from a different dataset and a Deep Q-Network (DQN)-based method when compared with more state encoders.