Improved search quality enhances users' satisfaction, which directly impacts sales growth of an E-Commerce (E-Com) platform. Traditional Learning to Rank (LTR) algorithms require relevance judgments on products. In E-Com, getting such judgments poses an immense challenge. In the literature, it is proposed to employ user feedback (such as clicks, add-to-basket (AtB) clicks and orders) to generate relevance judgments. It is done in two steps: first, query-product pair data are aggregated from the logs and then order rate etc are calculated for each pair in the logs. In this paper, we advocate counterfactual risk minimization (CRM) approach which circumvents the need of relevance judgements, data aggregation and is better suited for learning from logged data, i.e. contextual bandit feedback. Due to unavailability of public E-Com LTR dataset, we provide \textit{Mercateo dataset} from our platform. It contains more than 10 million AtB click logs and 1 million order logs from a catalogue of about 3.5 million products associated with 3060 queries. To the best of our knowledge, this is the first work which examines effectiveness of CRM approach in learning ranking model from real-world logged data. Our empirical evaluation shows that our CRM approach learns effectively from logged data and beats a strong baseline ranker ($\lambda$-MART) by a huge margin. Our method outperforms full-information loss (e.g. cross-entropy) on various deep neural network models. These findings demonstrate that by adopting CRM approach, E-Com platforms can get better product search quality compared to full-information approach. The code and dataset can be accessed at: https://github.com/ecom-research/CRM-LTR.
The limited availability of ground truth relevance labels has been a major impediment to the application of supervised methods to ad-hoc retrieval. As a result, unsupervised scoring methods, such as BM25, remain strong competitors to deep learning techniques which have brought on dramatic improvements in other domains, such as computer vision and natural language processing. Recent works have shown that it is possible to take advantage of the performance of these unsupervised methods to generate training data for learning-to-rank models. The key limitation to this line of work is the size of the training set required to surpass the performance of the original unsupervised method, which can be as large as $10^{13}$ training examples. Building on these insights, we propose two methods to reduce the amount of training data required. The first method takes inspiration from crowdsourcing, and leverages multiple unsupervised rankers to generate soft, or noise-aware, training labels. The second identifies harmful, or mislabeled, training examples and removes them from the training set. We show that our methods allow us to surpass the performance of the unsupervised baseline with far fewer training examples than previous works.
This tutorial covers and contrasts the two main methodologies in unbiased Learning to Rank (LTR): Counterfactual LTR and Online LTR. There has long been an interest in LTR from user interactions, however, this form of implicit feedback is very biased. In recent years, unbiased LTR methods have been introduced to remove the effect of different types of bias caused by user-behavior in search. For instance, a well addressed type of bias is position bias: the rank at which a document is displayed heavily affects the interactions it receives. Counterfactual LTR methods deal with such types of bias by learning from historical interactions while correcting for the effect of the explicitly modelled biases. Online LTR does not use an explicit user model, in contrast, it learns through an interactive process where randomized results are displayed to the user. Through randomization the effect of different types of bias can be removed from the learning process. Though both methodologies lead to unbiased LTR, their approaches differ considerably, furthermore, so do their theoretical guarantees, empirical results, effects on the user experience during learning, and applicability. Consequently, for practitioners the choice between the two is very substantial. By providing an overview of both approaches and contrasting them, we aim to provide an essential guide to unbiased LTR so as to aid in understanding and choosing between methodologies.
Learning to Rank (LTR) from user interactions is challenging as user feedback often contains high levels of bias and noise. At the moment, two methodologies for dealing with bias prevail in the field of LTR: counterfactual methods that learn from historical data and model user behavior to deal with biases; and online methods that perform interventions to deal with bias but use no explicit user models. For practitioners the decision between either methodology is very important because of its direct impact on end users. Nevertheless, there has never been a direct comparison between these two approaches to unbiased LTR. In this study we provide the first benchmarking of both counterfactual and online LTR methods under different experimental conditions. Our results show that the choice between the methodologies is consequential and depends on the presence of selection bias, and the degree of position bias and interaction noise. In settings with little bias or noise counterfactual methods can obtain the highest ranking performance; however, in other circumstances their optimization can be detrimental to the user experience. Conversely, online methods are very robust to bias and noise but require control over the displayed rankings. Our findings confirm and contradict existing expectations on the impact of model-based and intervention-based methods in LTR, and allow practitioners to make an informed decision between the two methodologies.
Search is a prominent channel for discovering products on an e-commerce platform. Ranking products retrieved from search becomes crucial to address customer's need and optimize for business metrics. While learning to Rank (LETOR) models have been extensively studied and have demonstrated efficacy in the context of web search; it is a relatively new research area to be explored in the e-commerce. In this paper, we present a framework for building LETOR model for an e-commerce platform. We analyze user queries and propose a mechanism to segment queries between broad and narrow based on user's intent. We discuss different types of features - query, product and query-product and discuss challenges in using them. We show that sparsity in product features can be tackled through a denoising auto-encoder while skip-gram based word embeddings help solve the query-product sparsity issues. We also present various target metrics that can be employed for evaluating search results and compare their robustness. Further, we build and compare performances of both pointwise and pairwise LETOR models on fashion category data set. We also build and compare distinct models for broad and narrow queries, analyze feature importance across these and show that these specialized models perform better than a combined model in the fashion world.
Computational models for blind image quality assessment (BIQA) are typically trained in well-controlled laboratory environments with limited generalizability to realistically distorted images. Similarly, BIQA models optimized for images captured in the wild cannot adequately handle synthetically distorted images. To face the cross-distortion-scenario challenge, we develop a BIQA model and an approach of training it on multiple IQA databases (of different distortion scenarios) simultaneously. A key step in our approach is to create and combine image pairs within individual databases as the training set, which effectively bypasses the issue of perceptual scale realignment. We compute a continuous quality annotation for each pair from the corresponding human opinions, indicating the probability of one image having better perceptual quality. We train a deep neural network for BIQA over the training set of massive image pairs by minimizing the fidelity loss. Experiments on six IQA databases demonstrate that the optimized model by the proposed training strategy is effective in blindly assessing image quality in the laboratory and wild, outperforming previous BIQA methods by a large margin.
We present a system, Spoke, for creating and searching internal knowledge base (KB) articles for organizations. Spoke is available as a SaaS (Software-as-a-Service) product deployed across hundreds of organizations with a diverse set of domains. Spoke continually improves search quality using conversational user feedback which allows it to provide better search experience than standard information retrieval systems without encoding any explicit domain knowledge. We achieve this by using a real-time online learning-to-rank (L2R) algorithm that automatically customizes relevance scoring for each organization deploying Spoke by using a query similarity kernel. The focus of this paper is on incorporating practical considerations into our relevance scoring function and algorithm that make Spoke easy to deploy and suitable for handling events that naturally happen over the life-cycle of any KB deployment. We show that Spoke outperforms competitive baselines by up to 41% in offline F1 comparisons.
This paper describes our system for The Microsoft AI Challenge India 2018: Ranking Passages for Web Question Answering. The system uses the biLSTM network with co-attention mechanism between query and passage representations. Additionally, we use self attention on embeddings to increase the lexical coverage by allowing the system to take union over different embeddings. We also incorporate hand-crafted features to improve the system performance. Our system achieved a Mean Reciprocal Rank (MRR) of 0.67 on eval-1 dataset.
Collaborative filtering is widely used in modern recommender systems. Recent research shows that variational autoencoders (VAEs) yield state-of-the-art performance by integrating flexible representations from deep neural networks into latent variable models, mitigating limitations of traditional linear factor models. VAEs are typically trained by maximizing the likelihood (MLE) of users interacting with ground-truth items. While simple and often effective, MLE-based training does not directly maximize the recommendation-quality metrics one typically cares about, such as top-N ranking. In this paper we investigate new methods for training collaborative filtering models based on actor-critic reinforcement learning, to directly optimize the non-differentiable quality metrics of interest. Specifically, we train a critic network to approximate ranking-based metrics, and then update the actor network (represented here by a VAE) to directly optimize against the learned metrics. In contrast to traditional learning-to-rank methods that require to re-run the optimization procedure for new lists, our critic-based method amortizes the scoring process with a neural network, and can directly provide the (approximate) ranking scores for new lists. Empirically, we show that the proposed methods outperform several state-of-the-art baselines, including recently-proposed deep learning approaches, on three large-scale real-world datasets. The code to reproduce the experimental results and figure plots is on Github: https://github.com/samlobel/RaCT_CF
Online Learning to Rank (OL2R) algorithms learn from implicit user feedback on the fly. The key of such algorithms is an unbiased estimation of gradients, which is often (trivially) achieved by uniformly sampling from the entire parameter space. This unfortunately introduces high-variance in gradient estimation, and leads to a worse regret of model estimation, especially when the dimension of parameter space is large. In this paper, we aim at reducing the variance of gradient estimation in OL2R algorithms. We project the selected updating direction into a space spanned by the feature vectors from examined documents under the current query (termed the "document space" for short), after interleaved test. Our key insight is that the result of interleaved test solely is governed by a user's relevance evaluation over the examined documents. Hence, the true gradient introduced by this test result should lie in the constructed document space, and components orthogonal to the document space in the proposed gradient can be safely removed for variance reduction. We prove that the projected gradient is an unbiased estimation of the true gradient, and show that this lower-variance gradient estimation results in significant regret reduction. Our proposed method is compatible with all existing OL2R algorithms which rank documents using a linear model. Extensive experimental comparisons with several state-of-the-art OL2R algorithms have confirmed the effectiveness of our proposed method in reducing the variance of gradient estimation and improving overall performance.
Recently, neural models for information retrieval are becoming increasingly popular. They provide effective approaches for product search due to their competitive advantages in semantic matching. However, it is challenging to use graph-based features, though proved very useful in IR literature, in these neural approaches. In this paper, we leverage the recent advances in graph embedding techniques to enable neural retrieval models to exploit graph-structured data for automatic feature extraction. The proposed approach can not only help to overcome the long-tail problem of click-through data, but also incorporate external heterogeneous information to improve search results. Extensive experiments on a real-world e-commerce dataset demonstrate significant improvement achieved by our proposed approach over multiple strong baselines both as an individual retrieval model and as a feature used in learning-to-rank frameworks.
According to common relevance-judgments regimes, such as TREC's, a document can be deemed relevant to a query even if it contains a very short passage of text with pertinent information. This fact has motivated work on passage-based document retrieval: document ranking methods that induce information from the document's passages. However, the main source of passage-based information utilized was passage-query similarities. We address the challenge of utilizing richer sources of passage-based information to improve document retrieval effectiveness. Specifically, we devise a suite of learning-to-rank-based document retrieval methods that utilize an effective ranking of passages produced in response to the query; the passage ranking is also induced using a learning-to-rank approach. Some of the methods quantify the ranking of the passages of a document. Others utilize the feature-based representation of passages used for learning a passage ranker. Empirical evaluation attests to the clear merits of our methods with respect to highly effective baselines. Our best performing method is based on learning a document ranking function using document-query features and passage-query features of the document's passage most highly ranked.
Researchers illustrate improvements in contextual encoding strategies via resultant performance on a battery of shared Natural Language Understanding (NLU) tasks. Many of these tasks are of a categorical prediction variety: given a conditioning context (e.g., an NLI premise), provide a label based on an associated prompt (e.g., an NLI hypothesis). The categorical nature of these tasks has led to common use of a cross entropy log-loss objective during training. We suggest this loss is intuitively wrong when applied to plausibility tasks, where the prompt by design is neither categorically entailed nor contradictory given the context. Log-loss naturally drives models to assign scores near 0.0 or 1.0, in contrast to our proposed use of a margin-based loss. Following a discussion of our intuition, we describe a confirmation study based on an extreme, synthetically curated task derived from MultiNLI. We find that a margin-based loss leads to a more plausible model of plausibility. Finally, we illustrate improvements on the Choice Of Plausible Alternative (COPA) task through this change in loss.
Uncoupled regression is the problem to learn a model from unlabeled data and the set of target values while the correspondence between them is unknown. Such a situation arises in predicting anonymized targets that involve sensitive information, e.g., one's annual income. Since existing methods for uncoupled regression often require strong assumptions on the true target function, and thus, their range of applications is limited, we introduce a novel framework that does not require such assumptions in this paper. Our key idea is to utilize pairwise comparison data, which consists of pairs of unlabeled data that we know which one has a larger target value. Such pairwise comparison data is easy to collect, as typically discussed in the learning-to-rank scenario, and does not break the anonymity of data. We propose two practical methods for uncoupled regression from pairwise comparison data and show that the learned regression model converges to the optimal model with the optimal parametric convergence rate when the target variable distributes uniformly. Moreover, we empirically show that for linear models the proposed methods are comparable to ordinary supervised regression with labeled data.
Non-stationarity appears in many online applications such as web search and advertising. In this paper, we study the online learning to rank problem in a non-stationary environment where user preferences change abruptly at an unknown moment in time. We consider the problem of identifying the K most attractive items and propose cascading non-stationary bandits, an online learning variant of the cascading model, where a user browses a ranked list from top to bottom and clicks on the first attractive item. We propose two algorithms for solving this non-stationary problem: CascadeDUCB and CascadeSWUCB. We analyze their performance and derive gap-dependent upper bounds on the n-step regret of these algorithms. We also establish a lower bound on the regret for cascading non-stationary bandits and show that both algorithms match the lower bound up to a logarithmic factor. Finally, we evaluate their performance on a real-world web search click dataset.
Accurate estimates of examination bias are crucial for unbiased learning-to-rank from implicit feedback in search engines and recommender systems, since they enable the use of Inverse Propensity Score (IPS) weighting techniques to address selection biases and missing data. Unfortunately, existing examination-bias estimators are limited to the Position-Based Model (PBM), where the examination bias may only depend on the rank of the document. To overcome this limitation, we propose a Contextual Position-Based Model (CPBM) where the examination bias may also depend on a context vector describing the query and the user. Furthermore, we propose an effective estimator for the CPBM based on intervention harvesting. A key feature of the estimator is that it does not require disruptive interventions but merely exploits natural variation resulting from the use of multiple historic ranking functions. Real-world experiments on the ArXiv search engine and semi-synthetic experiments on the Yahoo Learning-To-Rank dataset demonstrate the superior effectiveness and robustness of the new approach.
Learning to rank with biased click data is a well-known challenge. A variety of methods has been explored to debias click data for learning to rank such as click models, result interleaving and, more recently, the unbiased learning-to-rank framework based on inverse propensity weighting. Despite their differences, most existing studies separate the estimation of click bias (namely the \textit{propensity model}) from the learning of ranking algorithms. To estimate click propensities, they either conduct online result randomization, which can negatively affect the user experience, or offline parameter estimation, which has special requirements for click data and is optimized for objectives (e.g. click likelihood) that are not directly related to the ranking performance of the system. In this work, we address those problems by unifying the learning of propensity models and ranking models. We find that the problem of estimating a propensity model from click data is a dual problem of unbiased learning to rank. Based on this observation, we propose a Dual Learning Algorithm (DLA) that jointly learns an unbiased ranker and an \textit{unbiased propensity model}. DLA is an automatic unbiased learning-to-rank framework as it directly learns unbiased ranking models from biased click data without any preprocessing. It can adapt to the change of bias distributions and is applicable to online learning. Our empirical experiments with synthetic and real-world data show that the models trained with DLA significantly outperformed the unbiased learning-to-rank algorithms based on result randomization and the models trained with relevance signals extracted by click models.
LETOR is a package of benchmark data sets for research on LEarning TO Rank, which contains standard features, relevance judgments, data partitioning, evaluation tools, and several baselines. Version 1.0 was released in April 2007. Version 2.0 was released in Dec. 2007. Version 3.0 was released in Dec. 2008. This version, 4.0, was released in July 2009. Very different from previous versions (V3.0 is an update based on V2.0 and V2.0 is an update based on V1.0), LETOR4.0 is a totally new release. It uses the Gov2 web page collection (~25M pages) and two query sets from Million Query track of TREC 2007 and TREC 2008. We call the two query sets MQ2007 and MQ2008 for short. There are about 1700 queries in MQ2007 with labeled documents and about 800 queries in MQ2008 with labeled documents. If you have any questions or suggestions about the datasets, please kindly email us ([email omitted]). Our goal is to make the dataset reliable and useful for the community.
To enhance the performance of the recommender system, side information is extensively explored with various features (e.g., visual features and textual features). However, there are some demerits of side information: (1) the extra data is not always available in all recommendation tasks; (2) it is only for items, there is seldom high-level feature describing users. To address these gaps, we introduce the spectral features extracted from two hypergraph structures of the purchase records. Spectral features describe the \textit{similarity} of users/items in the graph space, which is critical for recommendation. We leverage spectral features to model the users' preference and items' properties by incorporating them into a Matrix Factorization (MF) model. In addition to modeling, we also use spectral features to optimize. Bayesian Personalized Ranking (BPR) is extensively leveraged to optimize models in implicit feedback data. However, in BPR, all missing values are regarded as negative samples equally while many of them are indeed unseen positive ones. We enrich the positive samples by calculating the similarity among users/items by the spectral features. The key ideas are: (1) similar users shall have similar preference on the same item; (2) a user shall have similar perception on similar items. Extensive experiments on two real-world datasets demonstrate the usefulness of the spectral features and the effectiveness of our spectrum-enhanced pairwise optimization. Our models outperform several state-of-the-art models significantly.
The Gradient Boosted Tree (GBT) algorithm is one of the most popular machine learning algorithms used in production, for tasks that include Click-Through Rate (CTR) prediction and learning-to-rank. To deal with the massive datasets available today, many distributed GBT methods have been proposed. However, they all assume a row-distributed dataset, addressing scalability only with respect to the number of data points and not the number of features, and increasing communication cost for high-dimensional data. In order to allow for scalability across both the data point and feature dimensions, and reduce communication cost, we propose block-distributed GBTs. We achieve communication efficiency by making full use of the data sparsity and adapting the Quickscorer algorithm to the block-distributed setting. We evaluate our approach using datasets with millions of features, and demonstrate that we are able to achieve multiple orders of magnitude reduction in communication cost for sparse data, with no loss in accuracy, while providing a more scalable design. As a result, we are able to reduce the training time for high-dimensional data, and allow more cost-effective scale-out without the need for expensive network communication.