Personal digital traces are constantly produced by connected devices, internet services and interactions. These digital traces are typically small, heterogeneous and stored in various locations in the cloud or on local devices, making it a challenge for users to interact with and search their own data. By adopting a multidimensional data model based on the six natural questions -- what, when, where, who, why and how -- to represent and unify heterogeneous personal digital traces, we can propose a learning-to-rank approach using the state of the art LambdaMART algorithm and frequency-based features that leverage the correlation between content (what), users (who), time (when), location (where) and data source (how) to improve the accuracy of search results. Due to the lack of publicly available personal training data, a combination of known-item query generation techniques and an unsupervised ranking model (field-based BM25) is used to build our own training sets. Experiments performed over a publicly available email collection and a personal digital data trace collection from a real user show that the frequency-based learning approach improves search accuracy when compared with traditional search tools.
This paper proposes an iterative inference algorithm for multi-hop explanation regeneration, that retrieves relevant factual evidence in the form of text snippets, given a natural language question and its answer. Combining multiple sources of evidence or facts for multi-hop reasoning becomes increasingly hard when the number of sources needed to make an inference grows. Our algorithm copes with this by decomposing the selection of facts from a corpus autoregressively, conditioning the next iteration on previously selected facts. This allows us to use a pairwise learning-to-rank loss. We validate our method on datasets of the TextGraphs 2019 and 2020 Shared Tasks for explanation regeneration. Existing work on this task either evaluates facts in isolation or artificially limits the possible chains of facts, thus limiting multi-hop inference. We demonstrate that our algorithm, when used with a pre-trained transformer model, outperforms the previous state-of-the-art in terms of precision, training time and inference efficiency.
We study Label Smoothing (LS), a widely used regularization technique, in the context of neural learning to rank (L2R) models. LS combines the ground-truth labels with a uniform distribution, encouraging the model to be less confident in its predictions. We analyze the relationship between the non-relevant documents-specifically how they are sampled-and the effectiveness of LS, discussing how LS can be capturing "hidden similarity knowledge" between the relevantand non-relevant document classes. We further analyze LS by testing if a curriculum-learning approach, i.e., starting with LS and after anumber of iterations using only ground-truth labels, is beneficial. Inspired by our investigation of LS in the context of neural L2R models, we propose a novel technique called Weakly Supervised Label Smoothing (WSLS) that takes advantage of the retrieval scores of the negative sampled documents as a weak supervision signal in the process of modifying the ground-truth labels. WSLS is simple to implement, requiring no modification to the neural ranker architecture. Our experiments across three retrieval tasks-passage retrieval, similar question retrieval and conversation response ranking-show that WSLS for pointwise BERT-based rankers leads to consistent effectiveness gains. The source code is available at https://anonymous.4open.science/r/dac85d48-6f71-4261-a7d8-040da6021c52/.
The success of a cross-sectional systematic strategy depends critically on accurately ranking assets prior to portfolio construction. Contemporary techniques perform this ranking step either with simple heuristics or by sorting outputs from standard regression or classification models, which have been demonstrated to be sub-optimal for ranking in other domains (e.g. information retrieval). To address this deficiency, we propose a framework to enhance cross-sectional portfolios by incorporating learning-to-rank algorithms, which lead to improvements of ranking accuracy by learning pairwise and listwise structures across instruments. Using cross-sectional momentum as a demonstrative case study, we show that the use of modern machine learning ranking algorithms can substantially improve the trading performance of cross-sectional strategies -- providing approximately threefold boosting of Sharpe Ratios compared to traditional approaches.
A key challenge with machine learning approaches for ranking is the gap between the performance metrics of interest and the surrogate loss functions that can be optimized with gradient-based methods. This gap arises because ranking metrics typically involve a sorting operation which is not differentiable w.r.t. the model parameters. Prior works have proposed surrogates that are loosely related to ranking metrics or simple smoothed versions thereof, and often fail to scale to real-world applications. We propose PiRank, a new class of differentiable surrogates for ranking, which employ a continuous, temperature-controlled relaxation to the sorting operator based on NeuralSort [1]. We show that PiRank exactly recovers the desired metrics in the limit of zero temperature and further propose a divide and-conquer extension that scales favorably to large list sizes, both in theory and practice. Empirically, we demonstrate the role of larger list sizes during training and show that PiRank significantly improves over comparable approaches on publicly available internet-scale learning-to-rank benchmarks.
Ranking systems form the basis for online search engines and recommendation services. They process large collections of items, for instance web pages or e-commerce products, and present the user with a small ordered selection. The goal of a ranking system is to help a user find the items they are looking for with the least amount of effort. Thus the rankings they produce should place the most relevant or preferred items at the top of the ranking. Learning to rank is a field within machine learning that covers methods which optimize ranking systems w.r.t. this goal. Traditional supervised learning to rank methods utilize expert-judgements to evaluate and learn, however, in many situations such judgements are impossible or infeasible to obtain. As a solution, methods have been introduced that perform learning to rank based on user clicks instead. The difficulty with clicks is that they are not only affected by user preferences, but also by what rankings were displayed. Therefore, these methods have to prevent being biased by other factors than user preference. This thesis concerns learning to rank methods based on user clicks and specifically aims to unify the different families of these methods. As a whole, the second part of this thesis proposes a framework that bridges many gaps between areas of online, counterfactual, and supervised learning to rank. It has taken approaches, previously considered independent, and unified them into a single methodology for widely applicable and effective learning to rank from user clicks.
Optimizing ranking systems based on user interactions is a well-studied problem. State-of-the-art methods for optimizing ranking systems based on user interactions are divided into online approaches - that learn by directly interacting with users - and counterfactual approaches - that learn from historical interactions. Existing online methods are hindered without online interventions and thus should not be applied counterfactually. Conversely, counterfactual methods cannot directly benefit from online interventions. We propose a novel intervention-aware estimator for both counterfactual and online Learning to Rank (LTR). With the introduction of the intervention-aware estimator, we aim to bridge the online/counterfactual LTR division as it is shown to be highly effective in both online and counterfactual scenarios. The estimator corrects for the effect of position bias, trust bias, and item-selection bias by using corrections based on the behavior of the logging policy and on online interventions: changes to the logging policy made during the gathering of click data. Our experimental results, conducted in a semi-synthetic experimental setup, show that, unlike existing counterfactual LTR methods, the intervention-aware estimator can greatly benefit from online interventions.
Search and recommendation systems, such as search engines, recruiting tools, online marketplaces, news, and social media, output ranked lists of content, products, and sometimes, people. Credit ratings, standardized tests, risk assessments output only a score, but are also used implicitly for ranking. Bias in such ranking systems, especially among the top ranks, can worsen social and economic inequalities, polarize opinions, and reinforce stereotypes. On the other hand, a bias correction for minority groups can cause more harm if perceived as favoring group-fair outcomes over meritocracy. In this paper, we formulate the problem of underranking in group-fair rankings, which was not addressed in previous work. Most group-fair ranking algorithms post-process a given ranking and output a group-fair ranking. We define underranking based on how close the group-fair rank of each item is to its original rank, and prove a lower bound on the trade-off achievable for simultaneous underranking and group fairness in ranking. We give a fair ranking algorithm that takes any given ranking and outputs another ranking with simultaneous underranking and group fairness guarantees comparable to the lower bound we prove. Our algorithm works with group fairness constraints for any number of groups. Our experimental results confirm the theoretical trade-off between underranking and group fairness, and also show that our algorithm achieves the best of both when compared to the state-of-the-art baselines.
Online Learning to Rank (OLTR) methods optimize ranking models by directly interacting with users, which allows them to be very efficient and responsive. All OLTR methods introduced during the past decade have extended on the original OLTR method: Dueling Bandit Gradient Descent (DBGD). Recently, a fundamentally different approach was introduced with the Pairwise Differentiable Gradient Descent (PDGD) algorithm. To date the only comparisons of the two approaches are limited to simulations with cascading click models and low levels of noise. The main outcome so far is that PDGD converges at higher levels of performance and learns considerably faster than DBGD-based methods. However, the PDGD algorithm assumes cascading user behavior, potentially giving it an unfair advantage. Furthermore, the robustness of both methods to high levels of noise has not been investigated. Therefore, it is unclear whether the reported advantages of PDGD over DBGD generalize to different experimental conditions. In this paper, we investigate whether the previous conclusions about the PDGD and DBGD comparison generalize from ideal to worst-case circumstances. We do so in two ways. First, we compare the theoretical properties of PDGD and DBGD, by taking a critical look at previously proven properties in the context of ranking. Second, we estimate an upper and lower bound on the performance of methods by simulating both ideal user behavior and extremely difficult behavior, i.e., almost-random non-cascading user models. Our findings show that the theoretical bounds of DBGD do not apply to any common ranking model and, furthermore, that the performance of DBGD is substantially worse than PDGD in both ideal and worst-case circumstances. These results reproduce previously published findings about the relative performance of PDGD vs. DBGD and generalize them to extremely noisy and non-cascading circumstances.
In this work, we propose FM-Pair, an adaptation of Factorization Machines with a pairwise loss function, making them effective for datasets with implicit feedback. The optimization model in FM-Pair is based on the BPR (Bayesian Personalized Ranking) criterion, which is a well-established pairwise optimization model. FM-Pair retains the advantages of FMs on generality, expressiveness and performance and yet it can be used for datasets with implicit feedback. We also propose how to apply FM-Pair effectively on two collaborative filtering problems, namely, context-aware recommendation and cross-domain collaborative filtering. By performing experiments on different datasets with explicit or implicit feedback we empirically show that in most of the tested datasets, FM-Pair beats state-of-the-art learning-to-rank methods such as BPR-MF (BPR with Matrix Factorization model). We also show that FM-Pair is significantly more effective for ranking, compared to the standard FMs model. Moreover, we show that FM-Pair can utilize context or cross-domain information effectively as the accuracy of recommendations would always improve with the right auxiliary features. Finally we show that FM-Pair has a linear time complexity and scales linearly by exploiting additional features.
Presentation bias is one of the key challenges when learning from implicit feedback in search engines, as it confounds the relevance signal. While it was recently shown how counterfactual learning-to-rank (LTR) approaches \cite{Joachims/etal/17a} can provably overcome presentation bias when observation propensities are known, it remains to show how to effectively estimate these propensities. In this paper, we propose the first method for producing consistent propensity estimates without manual relevance judgments, disruptive interventions, or restrictive relevance modeling assumptions. First, we show how to harvest a specific type of intervention data from historic feedback logs of multiple different ranking functions, and show that this data is sufficient for consistent propensity estimation in the position-based model. Second, we propose a new extremum estimator that makes effective use of this data. In an empirical evaluation, we find that the new estimator provides superior propensity estimates in two real-world systems -- Arxiv Full-text Search and Google Drive Search. Beyond these two points, we find that the method is robust to a wide range of settings in simulation studies.
Nowadays, where multimedia data is continuously generated, stored, and distributed, multimedia indexing, with its purpose of group- ing similar data, becomes more important than ever. Understanding the gist (=message) of multimedia instances is framed in related work as a ranking of concepts from a knowledge base, i.e., Wikipedia. We cast the task of multimedia indexing as a gist understanding problem. Our pipeline benefits from external knowledge and two subsequent learning- to-rank (l2r) settings. The first l2r produces a ranking of concepts rep- resenting the respective multimedia instance. The second l2r produces a mapping between the concept representation of an instance and the targeted class topic(s) for the multimedia indexing task. The evaluation on an established big size corpus (MIRFlickr25k, with 25,000 images), shows that multimedia indexing benefits from understanding the gist. Finally, with a MAP of 61.42, it can be shown that the multimedia in- dexing task benefits from understanding the gist. Thus, the presented end-to-end setting outperforms DBM and competes with Hashing-based methods.
Understanding searchers' queries is an essential component of semantic search systems. In many cases, search queries involve specific attributes of an entity in a knowledge base (KB), which can be further used to find query answers. In this study, we aim to move forward the understanding of queries by identifying their related entity attributes from a knowledge base. To this end, we introduce the task of entity attribute identification and propose two methods to address it: (i) a model based on Markov Random Field, and (ii) a learning to rank model. We develop a human annotated test collection and show that our proposed methods can bring significant improvements over the baseline methods.
The popular approaches to recommendation and ad-hoc retrieval tasks are largely distinct in the literature. In this work, we argue that many recommendation problems can also be cast as ad-hoc retrieval tasks. To demonstrate this, we build a solution for the RecSys 2018 Spotify challenge by combining standard ad-hoc retrieval models and using popular retrieval tools sets. We draw a parallel between the playlist continuation task and the task of finding good expansion terms for queries in ad-hoc retrieval, and show that standard pseudo-relevance feedback can be effective as a collaborative filtering approach. We also use ad-hoc retrieval for content-based recommendation by treating the input playlist title as a query and associating all candidate tracks with meta-descriptions extracted from the background data. The recommendations from these two approaches are further supplemented by a nearest neighbor search based on track embeddings learned by a popular neural model. Our final ranked list of recommendations is produced by a learning to rank model. Our proposed solution using ad-hoc retrieval models achieved a competitive performance on the music recommendation task at RecSys 2018 challenge---finishing at rank 7 out of 112 participating teams and at rank 5 out of 31 teams for the main and the creative tracks, respectively.
Recommending venues plays a critical rule in satisfying users' needs on location-based social networks. Recent studies have explored the idea of adopting collaborative ranking (CR) for recommendation, combining the idea of learning to rank and collaborative filtering. However, CR suffers from the sparsity problem, mainly because it associates similar users based on exact matching of the venues in their check-in history. Even though research in collaborative filtering has shown that considering auxiliary information such as geographical influence, helps the model to alleviate the sparsity problem, the same direction still needs to be explored in CR. In this work, we present a CR framework that focuses on the top of the ranked list while integrating an arbitrary number of similarity functions between venues as it learns the model's parameters. We further introduce three example similarity measures based on venues' contents and locations. Incorporating cross-venue similarity measures into the model enhances the latent associations between users as similar venues are also taken into account while associating users with each other. Our experiments on the TREC Contextual Suggestion dataset show that our proposed CR model beats other state-of-the-art venue suggestion methods.
Predictive models are omnipresent in automated and assisted decision making scenarios. But for the most part they are used as black boxes which output a prediction without understanding partially or even completely how different features influence the model prediction avoiding algorithmic transparency. Rankings are ordering over items encoding implicit comparisons typically learned using a family of features using learning-to-rank models. In this paper we focus on how best we can understand the decisions made by a ranker in a post-hoc model agnostic manner. We operate on the notion of interpretability based on explainability of rankings over an interpretable feature space. Furthermore we train a tree based model (inherently interpretable) using labels from the ranker, called secondary training data to provide explanations. Consequently, we attempt to study how well does a subset of features, potentially interpretable, explain the full model under different training sizes and algorithms. We do experiments on the learning to rank datasets with 30k queries and report results that serve show in certain settings we can learn a faithful interpretable ranker.
Personalized recommendation of Points of Interest (POIs) plays a key role in satisfying users on Location-Based Social Networks (LBSNs). In this paper, we propose a probabilistic model to find the mapping between user-annotated tags and locations' taste keywords. Furthermore, we introduce a dataset on locations' contextual appropriateness and demonstrate its usefulness in predicting the contextual relevance of locations. We investigate four approaches to use our proposed mapping for addressing the data sparsity problem: one model to reduce the dimensionality of location taste keywords and three models to predict user tags for a new location. Moreover, we present different scores calculated from multiple LBSNs and show how we incorporate new information from the mapping into a POI recommendation approach. Then, the computed scores are integrated using learning to rank techniques. The experiments on two TREC datasets show the effectiveness of our approach, beating state-of-the-art methods.
Neural network approaches have recently shown to be effective in several information retrieval (IR) tasks. However, neural approaches often require large volumes of training data to perform effectively, which is not always available. To mitigate the shortage of labeled data, training neural IR models with weak supervision has been recently proposed and received considerable attention in the literature. In weak supervision, an existing model automatically generates labels for a large set of unlabeled data, and a machine learning model is further trained on the generated "weak" data. Surprisingly, it has been shown in prior art that the trained neural model can outperform the weak labeler by a significant margin. Although these obtained improvements have been intuitively justified in previous work, the literature still lacks theoretical justification for the observed empirical findings. In this position paper, we propose to theoretically study weak supervision, in particular for IR tasks, e.g., learning to rank. We briefly review a set of our recent theoretical findings that shed light on learning from weakly supervised data, and provide guidelines on how train learning to rank models with weak supervision.
For many queries in the Web retrieval setting there is an on-going ranking competition: authors manipulate their documents so as to promote them in rankings. Such competitions can have unwarranted effects not only in terms of retrieval effectiveness, but also in terms of ranking robustness. A case in point, rankings can (rapidly) change due to small indiscernible perturbations of documents. While there has been a recent growing interest in analyzing the robustness of classifiers to adversarial manipulations, there has not yet been a study of the robustness of relevance-ranking functions. We address this challenge by formally analyzing different definitions and aspects of the robustness of learning-to-rank-based ranking functions. For example, we formally show that increased regularization of linear ranking functions increases ranking robustness. This finding leads us to conjecture that decreased variance of any ranking function results in increased robustness. We propose several measures for quantifying ranking robustness and use them to analyze ranking competitions between documents' authors. The empirical findings support our formal analysis and conjecture for both RankSVM and LambdaMART.
With the rapid advance of the Internet, search engines (e.g., Google, Bing, Yahoo!) are used by billions of users for each day. The main function of a search engine is to locate the most relevant webpages corresponding to what the user requests. This report focuses on the core problem of information retrieval: how to learn the relevance between a document (very often webpage) and a query given by user. Our analysis consists of two parts: 1) we use standard statistical methods to select important features among 137 candidates given by information retrieval researchers from Microsoft. We find that not all the features are useful, and give interpretations on the top-selected features; 2) we give baselines on prediction over the real-world dataset MSLR-WEB by using various learning algorithms. We find that models of boosting trees, random forest in general achieve the best performance of prediction. This agrees with the mainstream opinion in information retrieval community that tree-based algorithms outperform the other candidates for this problem.