Online learning to rank (OL2R) has attracted great research interests in recent years, thanks to its advantages in avoiding expensive relevance labeling as required in offline supervised ranking model learning. Such a solution explores the unknowns (e.g., intentionally present selected results on top positions) to improve its relevance estimation. This however triggers concerns on its ranking fairness: different groups of items might receive differential treatments during the course of OL2R. But existing fair ranking solutions usually require the knowledge of result relevance or a performing ranker beforehand, which contradicts with the setting of OL2R and thus cannot be directly applied to guarantee fairness. In this work, we propose a general framework to achieve fairness defined by group exposure in OL2R. The key idea is to calibrate exploration and exploitation for fairness control, relevance learning and online ranking quality. In particular, when the model is exploring a set of results for relevance feedback, we confine the exploration within a subset of random permutations, where fairness across groups is maintained while the feedback is still unbiased. Theoretically we prove such a strategy introduces minimum distortion in OL2R's regret to obtain fairness. Extensive empirical analysis is performed on two public learning to rank benchmark datasets to demonstrate the effectiveness of the proposed solution compared to existing fair OL2R solutions.
Nowadays, Online Travel Agencies provide the main service for booking holidays, business trips, accommodations, etc. As in many e-commerce services where users, items, and preferences are involved, the use of a Recommender System facilitates the navigation of the marketplaces. One of the main challenges when productizing machine learning models (and in this case, Learning-to-Rank models) is the need of, not only consistent pre-processing transformations, but also input features maintaining a similar scale both at training and prediction time. However, the features' scale does not necessarily stay the same in the real-world production environment, which could lead to unexpected ranking order. Normalization techniques such as feature standardization, batch normalization and layer normalization are commonly used to tackle the scaling issue. However, these techniques. To address this issue, in this paper we propose a novel scale-invariant ranking function (dubbed as SIR) which is accomplished by combining a deep and a wide neural network. We incorporate SIR with five state-of-the-art Learning-to-Rank models and compare the performance of the combined models with the classic algorithms on a large data set containing 56 million booked searches from the Hotels.com website. Besides, we simulate four real-world scenarios where the features' scale at the test set is inconsistent with that at the training set. The results reveal that when the features' scale is inconsistent at prediction time, Learning-To-Rank methods incorporating SIR outperform their original counterpart in all scenarios (with performance difference up to 14.7%), while when the features' scale at the training and test set are consistent our proposal achieves comparable accuracy to the classic algorithms.
Finding relevant scientific articles is crucial for advancing knowledge. Recommendation systems are helpful for such purpose, although they have only been applied to science recently. This article describes EILEEN (Exploratory Innovator of LitEraturE Networks), a recommendation system for scientific publications and grants with open source code and datasets. We describe EILEEN's architecture for ingesting and processing documents and modeling the recommendation system and keyphrase estimator. Using a unique dataset of log-in user behavior, we validate our recommendation system against Latent Semantic Analysis (LSA) and the standard ranking from Elasticsearch (Lucene scoring). We find that a learning-to-rank with Random Forest achieves an AUC of 0.9, significantly outperforming both baselines. Our results suggest that we can substantially improve science recommendations and learn about scientists' behavior through their search behavior. We make our system available through eileen.io
In multiple-choice exams, students select one answer from among typically four choices and can explain why they made that particular choice. Students are good at understanding natural language questions and based on their domain knowledge can easily infer the question's answer by 'connecting the dots' across various pertinent facts. Considering automated reasoning for elementary science question answering, we address the novel task of generating explanations for answers from human-authored facts. For this, we examine the practically scalable framework of feature-rich support vector machines leveraging domain-targeted, hand-crafted features. Explanations are created from a human-annotated set of nearly 5,000 candidate facts in the WorldTree corpus. Our aim is to obtain better matches for valid facts of an explanation for the correct answer of a question over the available fact candidates. To this end, our features offer a comprehensive linguistic and semantic unification paradigm. The machine learning problem is the preference ordering of facts, for which we test pointwise regression versus pairwise learning-to-rank. Our contributions are: (1) a case study in which two preference ordering approaches are systematically compared; (2) it is a practically competent approach that can outperform some variants of BERT-based reranking models; and (3) the human-engineered features make it an interpretable machine learning model for the task.
We consider language modelling (LM) as a multi-label structured prediction task by re-framing training from solely predicting a single ground-truth word to ranking a set of words which could continue a given context. To avoid annotating top-$k$ ranks, we generate them using pre-trained LMs: GPT-2, BERT, and Born-Again models. This leads to a rank-based form of knowledge distillation (KD). We also develop a method using $N$-grams to create a non-probabilistic teacher which generates the ranks without the need of a pre-trained LM. We confirm the hypotheses that we can treat LMing as a ranking task and that we can do so without the use of a pre-trained LM. We show that rank-based KD generally improves perplexity (PPL), often with statistical significance, when compared to Kullback-Leibler-based KD. Surprisingly, given the simplicity of the method, $N$-grams act as competitive teachers and achieve similar performance as using either BERT or a Born-Again model teachers. GPT-2 always acts as the best teacher, though, and using it and a Transformer-XL student on Wiki-02, rank-based KD reduces a cross-entropy baseline from 65.27 to 55.94 and against a KL-based KD of 56.70.
Ranking system is the core part of modern retrieval and recommender systems, where the goal is to rank candidate items given user contexts. Optimizing ranking systems online means that the deployed system can serve user requests, e.g., queries in the web search, and optimize the ranking policy by learning from user interactions, e.g., clicks. Bandit is a general online learning framework and can be used in our optimization task. However, due to the unique features of ranking, there are several challenges in designing bandit algorithms for ranking system optimization. In this dissertation, we study and propose solutions for four challenges in optimizing ranking systems online: effectiveness, safety, nonstationarity, and diversification. First, the effectiveness is related to how fast the algorithm learns from interactions. We study the effective online ranker evaluation task and propose the MergeDTS algorithm to solve the problem effectively. Second, the deployed algorithm should be safe, which means the algorithm only displays reasonable content to user requests. To solve the safe online learning to rank problem, we propose the BubbleRank algorithm. Third, as users change their preferences constantly, the algorithm should handle the nonstationarity. We formulate this nonstationary online learning to rank problem as cascade non-stationary bandits and propose CascadeDUCB and CascadeSWUCB algorithms to solve the problem. Finally, the contents in ranked lists should be diverse. We consider the results diversification task and propose the CascadeHybird algorithm that considers both the item relevance and results diversification when learning from user interactions.
In this paper, we present a new approach to estimate the layout of a room from its single image. While recent approaches for this task use robust features learnt from data, they resort to optimization for detecting the final layout. In addition to using learnt robust features, our approach learns an additional ranking function to estimate the final layout instead of using optimization. To learn this ranking function, we propose a framework to train a CNN using max-margin structure cost. Also, while most approaches aim at detecting cuboidal layouts, our approach detects non-cuboidal layouts for which we explicitly estimates layout complexity parameters. We use these parameters to propose layout candidates in a novel way. Our approach shows state-of-the-art results on standard datasets with mostly cuboidal layouts and also performs well on a dataset containing rooms with non-cuboidal layouts.
We explore a novel perspective of knowledge distillation (KD) for learning to rank (LTR), and introduce Self-Distilled neural Rankers (SDR), where student rankers are parameterized identically to their teachers. Unlike the existing ranking distillation work which pursues a good trade-off between performance and efficiency, SDR is able to significantly improve ranking performance of students over the teacher rankers without increasing model capacity. The key success factors of SDR, which differs from common distillation techniques for classification are: (1) an appropriate teacher score transformation function, and (2) a novel listwise distillation framework. Both techniques are specifically designed for ranking problems and are rarely studied in the existing knowledge distillation literature. Building upon the state-of-the-art neural ranking structure, SDR is able to push the limits of neural ranking performance above a recent rigorous benchmark study and significantly outperforms traditionally strong gradient boosted decision tree based models on 7 out of 9 key metrics, the first time in the literature. In addition to the strong empirical results, we give theoretical explanations on why listwise distillation is effective for neural rankers, and provide ablation studies to verify the necessity of the key factors in the SDR framework.
We present an overview of the second edition of the CheckThat! Lab at CLEF 2019. The lab featured two tasks in two different languages: English and Arabic. Task 1 (English) challenged the participating systems to predict which claims in a political debate or speech should be prioritized for fact-checking. Task 2 (Arabic) asked to (A) rank a given set of Web pages with respect to a check-worthy claim based on their usefulness for fact-checking that claim, (B) classify these same Web pages according to their degree of usefulness for fact-checking the target claim, (C) identify useful passages from these pages, and (D) use the useful pages to predict the claim's factuality. CheckThat! provided a full evaluation framework, consisting of data in English (derived from fact-checking sources) and Arabic (gathered and annotated from scratch) and evaluation based on mean average precision (MAP) and normalized discounted cumulative gain (nDCG) for ranking, and F1 for classification. A total of 47 teams registered to participate in this lab, and fourteen of them actually submitted runs (compared to nine last year). The evaluation results show that the most successful approaches to Task 1 used various neural networks and logistic regression. As for Task 2, learning-to-rank was used by the highest scoring runs for subtask A, while different classifiers were used in the other subtasks. We release to the research community all datasets from the lab as well as the evaluation scripts, which should enable further research in the important tasks of check-worthiness estimation and automatic claim verification.
The ability to collect and store ever more massive databases has been accompanied by the need to process them efficiently. In many cases, most observations have the same behavior, while a probable small proportion of these observations are abnormal. Detecting the latter, defined as outliers, is one of the major challenges for machine learning applications (e.g. in fraud detection or in predictive maintenance). In this paper, we propose a methodology addressing the problem of outlier detection, by learning a data-driven scoring function defined on the feature space which reflects the degree of abnormality of the observations. This scoring function is learnt through a well-designed binary classification problem whose empirical criterion takes the form of a two-sample linear rank statistics on which theoretical results are available. We illustrate our methodology with preliminary encouraging numerical experiments.
Given the recent proliferation of false claims online, there has been a lot of manual fact-checking effort. As this is very time-consuming, human fact-checkers can benefit from tools that can support them and make them more efficient. Here, we focus on building a system that could provide such support. Given an input document, it aims to detect all sentences that contain a claim that can be verified by some previously fact-checked claims (from a given database). The output is a re-ranked list of the document sentences, so that those that can be verified are ranked as high as possible, together with corresponding evidence. Unlike previous work, which has looked into claim retrieval, here we take a document-level perspective. We create a new manually annotated dataset for this task, and we propose suitable evaluation measures. We further experiment with a learning-to-rank approach, achieving sizable performance gains over several strong baselines. Our analysis demonstrates the importance of modeling text similarity and stance, while also taking into account the veracity of the retrieved previously fact-checked claims. We believe that this research would be of interest to fact-checkers, journalists, media, and regulatory authorities.
Search engines answer users' queries by listing relevant items (e.g. documents, songs, products, web pages, ...). These engines rely on algorithms that learn to rank items so as to present an ordered list maximizing the probability that it contains relevant item. The main challenge in the design of learning-to-rank algorithms stems from the fact that queries often have different meanings for different users. In absence of any contextual information about the query, one often has to adhere to the {\it diversity} principle, i.e., to return a list covering the various possible topics or meanings of the query. To formalize this learning-to-rank problem, we propose a natural model where (i) items are categorized into topics, (ii) users find items relevant only if they match the topic of their query, and (iii) the engine is not aware of the topic of an arriving query, nor of the frequency at which queries related to various topics arrive, nor of the topic-dependent click-through-rates of the items. For this problem, we devise LDR (Learning Diverse Rankings), an algorithm that efficiently learns the optimal list based on users' feedback only. We show that after $T$ queries, the regret of LDR scales as $O((N-L)\log(T))$ where $N$ is the number of all items. We further establish that this scaling cannot be improved, i.e., LDR is order optimal. Finally, using numerical experiments on both artificial and real-world data, we illustrate the superiority of LDR compared to existing learning-to-rank algorithms.
In counterfactual learning to rank (CLTR) user interactions are used as a source of supervision. Since user interactions come with bias, an important focus of research in this field lies in developing methods to correct for the bias of interactions. Inverse propensity scoring (IPS) is a popular method suitable for correcting position bias. Affine correction (AC) is a generalization of IPS that corrects for position bias and trust bias. IPS and AC provably remove bias, conditioned on an accurate estimation of the bias parameters. Estimating the bias parameters, in turn, requires an accurate estimation of the relevance probabilities. This cyclic dependency introduces practical limitations in terms of sensitivity, convergence and efficiency. We propose a new correction method for position and trust bias in CLTR in which, unlike the existing methods, the correction does not rely on relevance estimation. Our proposed method, mixture-based correction (MBC), is based on the assumption that the distribution of the CTRs over the items being ranked is a mixture of two distributions: the distribution of CTRs for relevant items and the distribution of CTRs for non-relevant items. We prove that our method is unbiased. The validity of our proof is not conditioned on accurate bias parameter estimation. Our experiments show that MBC, when used in different bias settings and accompanied by different LTR algorithms, outperforms AC, the state-of-the-art method for correcting position and trust bias, in some settings, while performing on par in other settings. Furthermore, MBC is orders of magnitude more efficient than AC in terms of the training time.
Predictor-based algorithms have achieved remarkable performance in the Neural Architecture Search (NAS) tasks. However, these methods suffer from high computation costs, as training the performance predictor usually requires training and evaluating hundreds of architectures from scratch. Previous works along this line mainly focus on reducing the number of architectures required to fit the predictor. In this work, we tackle this challenge from a different perspective - improve search efficiency by cutting down the computation budget of architecture training. We propose NOn-uniform Successive Halving (NOSH), a hierarchical scheduling algorithm that terminates the training of underperforming architectures early to avoid wasting budget. To effectively leverage the non-uniform supervision signals produced by NOSH, we formulate predictor-based architecture search as learning to rank with pairwise comparisons. The resulting method - RANK-NOSH, reduces the search budget by ~5x while achieving competitive or even better performance than previous state-of-the-art predictor-based methods on various spaces and datasets.
One of the biggest challenges that complicates applied supervised machine learning is the need for huge amounts of labeled data. Active Learning (AL) is a well-known standard method for efficiently obtaining labeled data by first labeling the samples that contain the most information based on a query strategy. Although many methods for query strategies have been proposed in the past, no clear superior method that works well in general for all domains has been found yet. Additionally, many strategies are computationally expensive which further hinders the widespread use of AL for large-scale annotation projects. We, therefore, propose ImitAL, a novel query strategy, which encodes AL as a learning-to-rank problem. For training the underlying neural network we chose Imitation Learning. The required demonstrative expert experience for training is generated from purely synthetic data. To show the general and superior applicability of \ImitAL{}, we perform an extensive evaluation comparing our strategy on 15 different datasets, from a wide range of domains, with 10 different state-of-the-art query strategies. We also show that our approach is more runtime performant than most other strategies, especially on very large datasets.
Recently, pre-trained language models such as BERT have been applied to document ranking for information retrieval, which first pre-train a general language model on an unlabeled large corpus and then conduct ranking-specific fine-tuning on expert-labeled relevance datasets. Ideally, an IR system would model relevance from a user-system dualism: the user's view and the system's view. User's view judges the relevance based on the activities of "real users" while the system's view focuses on the relevance signals from the system side, e.g., from the experts or algorithms, etc. Inspired by the user-system relevance views and the success of pre-trained language models, in this paper we propose a novel ranking framework called Pre-Rank that takes both user's view and system's view into consideration, under the pre-training and fine-tuning paradigm. Specifically, to model the user's view of relevance, Pre-Rank pre-trains the initial query-document representations based on large-scale user activities data such as the click log. To model the system's view of relevance, Pre-Rank further fine-tunes the model on expert-labeled relevance data. More importantly, the pre-trained representations, are fine-tuned together with handcrafted learning-to-rank features under a wide and deep network architecture. In this way, Pre-Rank can model the relevance by incorporating the relevant knowledge and signals from both real search users and the IR experts. To verify the effectiveness of Pre-Rank, we showed two implementations by using BERT and SetRank as the underlying ranking model, respectively. Experimental results base on three publicly available benchmarks showed that in both of the implementations, Pre-Rank can respectively outperform the underlying ranking models and achieved state-of-the-art performances.
Learning to rank systems has become an important aspect of our daily life. However, the implicit user feedback that is used to train many learning to rank models is usually noisy and suffered from user bias (i.e., position bias). Thus, obtaining an unbiased model using biased feedback has become an important research field for IR. Existing studies on unbiased learning to rank (ULTR) can be generalized into two families-algorithms that attain unbiasedness with logged data, offline learning, and algorithms that achieve unbiasedness by estimating unbiased parameters with real-time user interactions, namely online learning. While there exist many algorithms from both families, there lacks a unified way to compare and benchmark them. As a result, it can be challenging for researchers to choose the right technique for their problems or for people who are new to the field to learn and understand existing algorithms. To solve this problem, we introduced ULTRA, which is a flexible, extensible, and easily configure ULTR toolbox. Its key features include support for multiple ULTR algorithms with configurable hyperparameters, a variety of built-in click models that can be used separately to simulate clicks, different ranking model architecture and evaluation metrics, and simple learning to rank pipeline creation. In this paper, we discuss the general framework of ULTR, briefly describe the algorithms in ULTRA, detailed the structure, and pipeline of the toolbox. We experimented on all the algorithms supported by ultra and showed that the toolbox performance is reasonable. Our toolbox is an important resource for researchers to conduct experiments on ULTR algorithms with different configurations as well as testing their own algorithms with the supported features.
Recently, we have witnessed the bloom of neural ranking models in the information retrieval (IR) field. So far, much effort has been devoted to developing effective neural ranking models that can generalize well on new data. There has been less attention paid to the robustness perspective. Unlike the effectiveness which is about the average performance of a system under normal purpose, robustness cares more about the system performance in the worst case or under malicious operations instead. When a new technique enters into the real-world application, it is critical to know not only how it works in average, but also how would it behave in abnormal situations. So we raise the question in this work: Are neural ranking models robust? To answer this question, firstly, we need to clarify what we refer to when we talk about the robustness of ranking models in IR. We show that robustness is actually a multi-dimensional concept and there are three ways to define it in IR: 1) The performance variance under the independent and identically distributed (I.I.D.) setting; 2) The out-of-distribution (OOD) generalizability; and 3) The defensive ability against adversarial operations. The latter two definitions can be further specified into two different perspectives respectively, leading to 5 robustness tasks in total. Based on this taxonomy, we build corresponding benchmark datasets, design empirical experiments, and systematically analyze the robustness of several representative neural ranking models against traditional probabilistic ranking models and learning-to-rank (LTR) models. The empirical results show that there is no simple answer to our question. While neural ranking models are less robust against other IR models in most cases, some of them can still win 1 out of 5 tasks. This is the first comprehensive study on the robustness of neural ranking models.
Autocomplete (a.k.a "Query Auto-Completion", "AC") suggests full queries based on a prefix typed by customer. Autocomplete has been a core feature of commercial search engine. In this paper, we propose a novel context-aware neural network based pairwise ranker (DeepPLTR) to improve AC ranking, DeepPLTR leverages contextual and behavioral features to rank queries by minimizing a pairwise loss, based on a fully-connected neural network structure. Compared to LambdaMART ranker, DeepPLTR shows +3.90% MeanReciprocalRank (MRR) lift in offline evaluation, and yielded +0.06% (p < 0.1) Gross Merchandise Value (GMV) lift in an Amazon's online A/B experiment.
Exposure bias is a well-known issue in recommender systems where items and suppliers are not equally represented in the recommendation results. This is especially problematic when bias is amplified over time as a few popular items are repeatedly over-represented in recommendation lists. This phenomenon can be viewed as a recommendation feedback loop: the system repeatedly recommends certain items at different time points and interactions of users with those items will amplify bias towards those items over time. This issue has been extensively studied in the literature on model-based or neighborhood-based recommendation algorithms, but less work has been done on online recommendation models such as those based on multi-armed Bandit algorithms. In this paper, we study exposure bias in a class of well-known bandit algorithms known as Linear Cascade Bandits. We analyze these algorithms on their ability to handle exposure bias and provide a fair representation for items and suppliers in the recommendation results. Our analysis reveals that these algorithms fail to treat items and suppliers fairly and do not sufficiently explore the item space for each user. To mitigate this bias, we propose a discounting factor and incorporate it into these algorithms that controls the exposure of items at each time step. To show the effectiveness of the proposed discounting factor on mitigating exposure bias, we perform experiments on two datasets using three cascading bandit algorithms and our experimental results show that the proposed method improves the exposure fairness for items and suppliers.