Online controlled experiments are a crucial tool to allow for confident decision-making in technology companies. A North Star metric is defined (such as long-term revenue or user retention), and system variants that statistically significantly improve on this metric in an A/B-test can be considered superior. North Star metrics are typically delayed and insensitive. As a result, the cost of experimentation is high: experiments need to run for a long time, and even then, type-II errors (i.e. false negatives) are prevalent. We propose to tackle this by learning metrics from short-term signals that directly maximise the statistical power they harness with respect to the North Star. We show that existing approaches are prone to overfitting, in that higher average metric sensitivity does not imply improved type-II errors, and propose to instead minimise the $p$-values a metric would have produced on a log of past experiments. We collect such datasets from two social media applications with over 160 million Monthly Active Users each, totalling over 153 A/B-pairs. Empirical results show that we are able to increase statistical power by up to 78% when using our learnt metrics stand-alone, and by up to 210% when used in tandem with the North Star. Alternatively, we can obtain constant statistical power at a sample size that is down to 12% of what the North Star requires, significantly reducing the cost of experimentation.
Online controlled experiments, such as A/B-tests, are commonly used by modern tech companies to enable continuous system improvements. Despite their paramount importance, A/B-tests are expensive: by their very definition, a percentage of traffic is assigned an inferior system variant. To ensure statistical significance on top-level metrics, online experiments typically run for several weeks. Even then, a considerable amount of experiments will lead to inconclusive results (i.e. false negatives, or type-II error). The main culprit for this inefficiency is the variance of the online metrics. Variance reduction techniques have been proposed in the literature, but their direct applicability to commonly used ratio metrics (e.g. click-through rate or user retention) is limited. In this work, we successfully apply variance reduction techniques to ratio metrics on a large-scale short-video platform: ShareChat. Our empirical results show that we can either improve A/B-test confidence in 77% of cases, or can retain the same level of confidence with 30% fewer data points. Importantly, we show that the common approach of including as many covariates as possible in regression is counter-productive, highlighting that control variates based on Gradient-Boosted Decision Tree predictors are most effective. We discuss the practicalities of implementing these methods at scale and showcase the cost reduction they beget.
Many platforms on the web present ranked lists of content to users, typically optimized for engagement-, satisfaction- or retention- driven metrics. Advances in the Learning-to-Rank (LTR) research literature have enabled rapid growth in this application area. Several popular interfaces now include nested lists, where users can enter a 2nd-level feed via any given 1st-level item. Naturally, this has implications for evaluation metrics, objective functions, and the ranking policies we wish to learn. We propose a theoretically grounded method to incorporate 2nd-level feedback into any 1st-level ranking model. Online experiments on a large-scale recommendation system confirm our theoretical findings.
Practitioners who wish to build real-world applications that rely on ranking models, need to decide which modelling paradigm to follow. This is not an easy choice to make, as the research literature on this topic has been shifting in recent years. In particular, whilst Gradient Boosted Decision Trees (GBDTs) have reigned supreme for more than a decade, the flexibility of neural networks has allowed them to catch up, and recent works report accuracy metrics that are on par. Nevertheless, practical systems require considerations beyond mere accuracy metrics to decide on a modelling approach. This work describes our experiences in balancing some of the trade-offs that arise, presenting a case study on a short-video recommendation application. We highlight (1) neural networks' ability to handle large training data size, user- and item-embeddings allows for more accurate models than GBDTs in this setting, and (2) because GBDTs are less reliant on specialised hardware, they can provide an equally accurate model at a lower cost. We believe these findings are of relevance to researchers in both academia and industry, and hope they can inspire practitioners who need to make similar modelling choices in the future.
Ad-load balancing is a critical challenge in online advertising systems, particularly in the context of social media platforms, where the goal is to maximize user engagement and revenue while maintaining a satisfactory user experience. This requires the optimization of conflicting objectives, such as user satisfaction and ads revenue. Traditional approaches to ad-load balancing rely on static allocation policies, which fail to adapt to changing user preferences and contextual factors. In this paper, we present an approach that leverages off-policy learning and evaluation from logged bandit feedback. We start by presenting a motivating analysis of the ad-load balancing problem, highlighting the conflicting objectives between user satisfaction and ads revenue. We emphasize the nuances that arise due to user heterogeneity and the dependence on the user's position within a session. Based on this analysis, we define the problem as determining the optimal ad-load for a particular feed fetch. To tackle this problem, we propose an off-policy learning framework that leverages unbiased estimators such as Inverse Propensity Scoring (IPS) and Doubly Robust (DR) to learn and estimate the policy values using offline collected stochastic data. We present insights from online A/B experiments deployed at scale across over 80 million users generating over 200 million sessions, where we find statistically significant improvements in both user satisfaction metrics and ads revenue for the platform.
Off-Policy Estimation (OPE) methods allow us to learn and evaluate decision-making policies from logged data. This makes them an attractive choice for the offline evaluation of recommender systems, and several recent works have reported successful adoption of OPE methods to this end. An important assumption that makes this work is the absence of unobserved confounders: random variables that influence both actions and rewards at data collection time. Because the data collection policy is typically under the practitioner's control, the unconfoundedness assumption is often left implicit, and its violations are rarely dealt with in the existing literature. This work aims to highlight the problems that arise when performing off-policy estimation in the presence of unobserved confounders, specifically focusing on a recommendation use-case. We focus on policy-based estimators, where the logging propensities are learned from logged data. We characterise the statistical bias that arises due to confounding, and show how existing diagnostics are unable to uncover such cases. Because the bias depends directly on the true and unobserved logging propensities, it is non-identifiable. As the unconfoundedness assumption is famously untestable, this becomes especially problematic. This paper emphasises this common, yet often overlooked issue. Through synthetic data, we empirically show how na\"ive propensity estimation under confounding can lead to severely biased metric estimates that are allowed to fly under the radar. We aim to cultivate an awareness among researchers and practitioners of this important problem, and touch upon potential research directions towards mitigating its effects.
Approaches to recommendation are typically evaluated in one of two ways: (1) via a (simulated) online experiment, often seen as the gold standard, or (2) via some offline evaluation procedure, where the goal is to approximate the outcome of an online experiment. Several offline evaluation metrics have been adopted in the literature, inspired by ranking metrics prevalent in the field of Information Retrieval. (Normalised) Discounted Cumulative Gain (nDCG) is one such metric that has seen widespread adoption in empirical studies, and higher (n)DCG values have been used to present new methods as the state-of-the-art in top-$n$ recommendation for many years. Our work takes a critical look at this approach, and investigates when we can expect such metrics to approximate the gold standard outcome of an online experiment. We formally present the assumptions that are necessary to consider DCG an unbiased estimator of online reward and provide a derivation for this metric from first principles, highlighting where we deviate from its traditional uses in IR. Importantly, we show that normalising the metric renders it inconsistent, in that even when DCG is unbiased, ranking competing methods by their normalised DCG can invert their relative order. Through a correlation analysis between off- and on-line experiments conducted on a large-scale recommendation platform, we show that our unbiased DCG estimates strongly correlate with online reward, even when some of the metric's inherent assumptions are violated. This statement no longer holds for its normalised variant, suggesting that nDCG's practical utility may be limited.
Modern web-based platforms show ranked lists of recommendations to users, attempting to maximise user satisfaction or business metrics. Typically, the goal of such systems boils down to maximising the exposure probability for items that are deemed "reward-maximising" according to a metric of interest. This general framing comprises streaming applications, as well as e-commerce or job recommendations, and even web search. Position bias or user models can be used to estimate exposure probabilities for each use-case, specifically tailored to how users interact with the presented rankings. A unifying factor in these diverse problem settings is that typically only one or several items will be engaged with (clicked, streamed,...) before a user leaves the ranked list. Short-video feeds on social media platforms diverge from this general framing in several ways, most notably that users do not tend to leave the feed after e.g. liking a post. Indeed, seemingly infinite feeds invite users to scroll further down the ranked list. For this reason, existing position bias or user models tend to fall short in such settings, as they do not accurately capture users' interaction modalities. In this work, we propose a novel and probabilistically sound personalised position bias model for feed recommendations. We focus on a 1st-level feed in a hierarchical structure, where users may enter a 2nd-level feed via any given 1st-level item. We posit that users come to the platform with a scrolling budget drawn according to some distribution, and show how the survival function of said distribution can be used to obtain closed-form estimates for personalised exposure probabilities. Empirical insights from a large-scale social media platform show how our probabilistic position bias model more accurately captures empirical exposure than existing models, and paves the way for unbiased evaluation and learning-to-rank.
In this paper we propose RecFusion, which comprise a set of diffusion models for recommendation. Unlike image data which contain spatial correlations, a user-item interaction matrix, commonly utilized in recommendation, lacks spatial relationships between users and items. We formulate diffusion on a 1D vector and propose binomial diffusion, which explicitly models binary user-item interactions with a Bernoulli process. We show that RecFusion approaches the performance of complex VAE baselines on the core recommendation setting (top-n recommendation for binary non-sequential feedback) and the most common datasets (MovieLens and Netflix). Our proposed diffusion models that are specialized for 1D and/or binary setups have implications beyond recommendation systems, such as in the medical domain with MRI and CT scans.