This paper considers the problem of document ranking in information retrieval systems by Learning to Rank. We propose ConvRankNet combining a Siamese Convolutional Neural Network encoder and the RankNet ranking model which could be trained in an end-to-end fashion. We prove a general result justifying the linear test-time complexity of pairwise Learning to Rank approach. Experiments on the OHSUMED dataset show that ConvRankNet outperforms systematically existing feature-based models.
In Online Learning to Rank (OLTR) the aim is to find an optimal ranking model by interacting with users. When learning from user behavior, systems must interact with users while simultaneously learning from those interactions. Unlike other Learning to Rank (LTR) settings, existing research in this field has been limited to linear models. This is due to the speed-quality tradeoff that arises when selecting models: complex models are more expressive and can find the best rankings but need more user interactions to do so, a requirement that risks frustrating users during training. Conversely, simpler models can be optimized on fewer interactions and thus provide a better user experience, but they will converge towards suboptimal rankings. This tradeoff creates a deadlock, since novel models will not be able to improve either the user experience or the final convergence point, without sacrificing the other. Our contribution is twofold. First, we introduce a fast OLTR model called Sim-MGD that addresses the speed aspect of the speed-quality tradeoff. Sim-MGD ranks documents based on similarities with reference documents. It converges rapidly and, hence, gives a better user experience but it does not converge towards the optimal rankings. Second, we contribute Cascading Multileave Gradient Descent (C-MGD) for OLTR that directly addresses the speed-quality tradeoff by using a cascade that enables combinations of the best of two worlds: fast learning and high quality final convergence. C-MGD can provide the better user experience of Sim-MGD while maintaining the same convergence as the state-of-the-art MGD model. This opens the door for future work to design new models for OLTR without having to deal with the speed-quality tradeoff.
Deep neural networks have recently shown promise in the ad-hoc retrieval task. However, such models have often been based on one field of the document, for example considering document title only or document body only. Since in practice documents typically have multiple fields, and given that non-neural ranking models such as BM25F have been developed to take advantage of document structure, this paper investigates how neural models can deal with multiple document fields. We introduce a model that can consume short text fields such as document title and long text fields such as document body. It can also handle multi-instance fields with variable number of instances, for example where each document has zero or more instances of incoming anchor text. Since fields vary in coverage and quality, we introduce a masking method to handle missing field instances, as well as a field-level dropout method to avoid relying too much on any one field. As in the studies of non-neural field weighting, we find it is better for the ranker to score the whole document jointly, rather than generate a per-field score and aggregate. We find that different document fields may match different aspects of the query and therefore benefit from comparing with separate representations of the query text. The combination of techniques introduced here leads to a neural ranker that can take advantage of full document structure, including multiple instance and missing instance data, of variable length. The techniques significantly enhance the performance of the ranker, and also outperform a learning to rank baseline with hand-crafted features.
When applying learning to rank algorithms to Web search, a large number of features are usually designed to capture the relevance signals. Most of these features are computed based on the extracted textual elements, link analysis, and user logs. However, Web pages are not solely linked texts, but have structured layout organizing a large variety of elements in different styles. Such layout itself can convey useful visual information, indicating the relevance of a Web page. For example, the query-independent layout (i.e., raw page layout) can help identify the page quality, while the query-dependent layout (i.e., page rendered with matched query words) can further tell rich structural information (e.g., size, position and proximity) of the matching signals. However, such visual information of layout has been seldom utilized in Web search in the past. In this work, we propose to learn rich visual features automatically from the layout of Web pages (i.e., Web page snapshots) for relevance ranking. Both query-independent and query-dependent snapshots are considered as the new inputs. We then propose a novel visual perception model inspired by human's visual search behaviors on page viewing to extract the visual features. This model can be learned end-to-end together with traditional human-crafted features. We also show that such visual features can be efficiently acquired in the online setting with an extended inverted indexing scheme. Experiments on benchmark collections demonstrate that learning visual features from Web page snapshots can significantly improve the performance of relevance ranking in ad-hoc Web retrieval tasks.
This paper evaluates existing and newly proposed answer selection methods based on pre-trained word embeddings. Word embeddings are highly effective in various natural language processing tasks and their integration into traditional information retrieval (IR) systems allows for the capture of semantic relatedness between questions and answers. Empirical results on three publicly available data sets show significant gains over traditional term frequency based approaches in both supervised and unsupervised settings. We show that combining these word embedding features with traditional learning-to-rank techniques can achieve similar performance to state-of-the-art neural networks trained for the answer selection task.
We describe a new deep learning architecture for learning to rank question answer pairs. Our approach extends the long short-term memory (LSTM) network with holographic composition to model the relationship between question and answer representations. As opposed to the neural tensor layer that has been adopted recently, the holographic composition provides the benefits of scalable and rich representational learning approach without incurring huge parameter costs. Overall, we present Holographic Dual LSTM (HD-LSTM), a unified architecture for both deep sentence modeling and semantic matching. Essentially, our model is trained end-to-end whereby the parameters of the LSTM are optimized in a way that best explains the correlation between question and answer representations. In addition, our proposed deep learning architecture requires no extensive feature engineering. Via extensive experiments, we show that HD-LSTM outperforms many other neural architectures on two popular benchmark QA datasets. Empirical studies confirm the effectiveness of holographic composition over the neural tensor layer.
With the popularity of Linked Open Data (LOD) and the associated rise in freely accessible knowledge that can be accessed via LOD, exploiting LOD for recommender systems has been widely studied based on various approaches such as graph-based or using different machine learning models with LOD-enabled features. Many of the previous approaches require construction of an additional graph to run graph-based algorithms or to extract path-based features by combining user- item interactions (e.g., likes, dislikes) and background knowledge from LOD. In this paper, we investigate Factorization Machines (FMs) based on particularly lightweight LOD-enabled features which can be directly obtained via a public SPARQL Endpoint without any additional effort to construct a graph. Firstly, we aim to study whether using FM with these lightweight LOD-enabled features can provide competitive performance compared to a learning-to-rank approach leveraging LOD as well as other well-established approaches such as kNN-item and BPRMF. Secondly, we are interested in finding out to what extent each set of LOD-enabled features contributes to the recommendation performance. Experimental evaluation on a standard dataset shows that our proposed approach using FM with lightweight LOD-enabled features provides the best performance compared to other approaches in terms of five evaluation metrics. In addition, the study of the recommendation performance based on different sets of LOD-enabled features indicate that property-object lists and PageRank scores of items are useful for improving the performance, and can provide the best performance through using them together for FM. We observe that subject-property lists of items does not contribute to the recommendation performance but rather decreases the performance.
Recommendation systems are being explored by Cable TV operators to improve user satisfaction with services, such as Live TV and Video on Demand (VOD) services. More recently, Catch-up TV has been introduced, allowing users to watch recent broadcast content whenever they want to. These services give users a large set of options from which they can choose from, creating an information overflow problem. Thus, recommendation systems arise as essential tools to solve this problem by helping users in their selection, which increases not only user satisfaction but also user engagement and content consumption. In this paper we present a learning to rank approach that uses contextual information and implicit feedback to improve recommendation systems for a Cable TV operator that provides Live and Catch-up TV services. We compare our approach with existing state-of-the-art algorithms and show that our approach is superior in accuracy, while maintaining high scores of diversity and serendipity.
Mobile search has recently been shown to be the major contributor to the growing search market. The key difference between mobile search and desktop search is that information presentation is limited to the screen space of the mobile device. Thus, major search engines have adopted a new type of search result presentation, known as \textit{information cards}, in which each card presents summarized results from one domain/vertical, for a given query, to augment the standard blue-links search results. While it has been widely acknowledged that information cards are particularly suited to mobile user experience, it is also challenging to optimize such result sets. Typically, user engagement metrics like query reformulation are based on whole ranked list of cards for each query and most traditional learning to rank algorithms require per-item relevance labels. In this paper, we investigate the possibility of interpreting query reformulation into effective relevance labels for query-card pairs. We inherit the concept of conventional learning-to-rank, and propose pointwise, pairwise and listwise interpretations for query reformulation. In addition, we propose a learning-to-label strategy that learns the contribution of each card, with respect to a query, where such contributions can be used as labels for training card ranking models. We utilize a state-of-the-art ranking model and demonstrate the effectiveness of proposed mechanisms on a large-scale mobile data from a major search engine, showing that models trained from labels derived from user engagement can significantly outperform ones trained from human judgment labels.
Deciding which vocabulary terms to use when modeling data as Linked Open Data (LOD) is far from trivial. Choosing too general vocabulary terms, or terms from vocabularies that are not used by other LOD datasets, is likely to lead to a data representation, which will be harder to understand by humans and to be consumed by Linked data applications. In this technical report, we propose TermPicker: a novel approach for vocabulary reuse by recommending RDF types and properties based on exploiting the information on how other data providers on the LOD cloud use RDF types and properties to describe their data. To this end, we introduce the notion of so-called schema-level patterns (SLPs). They capture how sets of RDF types are connected via sets of properties within some data collection, e.g., within a dataset on the LOD cloud. TermPicker uses such SLPs and generates a ranked list of vocabulary terms for reuse. The lists of recommended terms are ordered by a ranking model which is computed using the machine learning approach Learning To Rank (L2R). TermPicker is evaluated based on the recommendation quality that is measured using the Mean Average Precision (MAP) and the Mean Reciprocal Rank at the first five positions (MRR@5). Our results illustrate an improvement of the recommendation quality by 29% - 36% when using SLPs compared to the beforehand investigated baselines of recommending solely popular vocabulary terms or terms from the same vocabulary. The overall best results are achieved using SLPs in conjunction with the Learning To Rank algorithm Random Forests.
In the medical domain, a Systematic Literature Review (SLR) attempts to collect all empirical evidence, that fit pre-specified eligibility criteria, in order to answer a specific research question. The process of preparing an SLR consists of multiple tasks that are labor-intensive and time-consuming, involving large monetary costs. Technology-assisted review (TAR) methods automate the different processes of creating an SLR and they are particularly focused on reducing the burden of screening for reviewers. We present a novel method for TAR that implements a full pipeline from the research protocol to the screening of the relevant papers. Our pipeline overcomes the need of a Boolean query constructed by specialists and consists of three different components: the primary retrieval engine, the inter-review ranker and the intra-review ranker, combining learning-to-rank techniques with a relevance feedback method. In addition, we contribute an updated version of the Task 2 of the CLEF 2019 eHealth Lab dataset, which we make publicly available. Empirical results on this dataset show that our approach can achieve state-of-the-art results.
Recommender systems is one of the most successful AI technologies applied in the internet cooperations. Popular internet products such as TikTok, Amazon, and YouTube have all integrated recommender systems as their core product feature. Although recommender systems have received great success, it is well known for highly skewed datasets, engineers and researchers need to adjust their methods to tackle the specific problem to yield good results. Inability to deal with highly skewed dataset usually generates hard computational problems for big data clusters and unsatisfactory results for customers. In this paper, we propose a new algorithm solving the problem in the framework of matrix factorization. We model the data skewness factors in the theoretic modeling of the approach with easy to interpret and easy to implement formulas. We prove in experiments our method generates comparably favorite results with popular recommender system algorithms such as Learning to Rank , Alternating Least Squares and Deep Matrix Factorization.
We propose Conditional Imputation GAN, an extended missing data imputation method based on Generative Adversarial Networks (GANs). The motivating use case is learning-to-rank, the cornerstone of modern search, recommendation system, and information retrieval applications. Empirical ranking datasets do not always follow standard Gaussian distributions or Missing Completely At Random (MCAR) mechanism, which are standard assumptions of classic missing data imputation methods. Our methodology provides a simple solution that offers compatible imputation guarantees while relaxing assumptions for missing mechanisms and sidesteps approximating intractable distributions to improve imputation quality. We prove that the optimal GAN imputation is achieved for Extended Missing At Random (EMAR) and Extended Always Missing At Random (EAMAR) mechanisms, beyond the naive MCAR. Our method demonstrates the highest imputation quality on the open-source Microsoft Research Ranking (MSR) Dataset and a synthetic ranking dataset compared to state-of-the-art benchmarks and across various feature distributions. Using a proprietary Amazon Search ranking dataset, we also demonstrate comparable ranking quality metrics for ranking models trained on GAN-imputed data compared to ground-truth data.
Learning to rank with implicit feedback is one of the most important tasks in many real-world information systems where the objective is some specific utility, e.g., clicks and revenue. However, we point out that existing methods based on probabilistic ranking principle do not necessarily achieve the highest utility. To this end, we propose a novel ranking framework called U-rank that directly optimizes the expected utility of the ranking list. With a position-aware deep click-through rate prediction model, we address the attention bias considering both query-level and item-level features. Due to the item-specific attention bias modeling, the optimization for expected utility corresponds to a maximum weight matching on the item-position bipartite graph. We base the optimization of this objective in an efficient Lambdaloss framework, which is supported by both theoretical and empirical analysis. We conduct extensive experiments for both web search and recommender systems over three benchmark datasets and two proprietary datasets, where the performance gain of U-rank over state-of-the-arts is demonstrated. Moreover, our proposed U-rank has been deployed on a large-scale commercial recommender and a large improvement over the production baseline has been observed in an online A/B testing.
Neural approaches to learning term embeddings have led to improved computation of similarity and ranking in information retrieval (IR). So far neural representation learning has not been extended to meta-textual information that is readily available for many IR tasks, for example, patent classes in prior-art retrieval, topical information in Wikipedia articles, or product categories in e-commerce data. We present a framework that learns embeddings for meta-textual categories, and optimizes a pairwise ranking objective for improved matching based on combined embeddings of textual and meta-textual information. We show considerable gains in an experimental evaluation on cross-lingual retrieval in the Wikipedia domain for three language pairs, and in the Patent domain for one language pair. Our results emphasize that the mode of combining different types of information is crucial for model improvement.
Large scale eCommerce platforms such as eBay carry a wide variety of inventory and provide several buying choices to online shoppers. It is critical for eCommerce search engines to showcase in the top results the variety and selection of inventory available, specifically in the context of the various buying intents that may be associated with a search query. Search rankers are most commonly powered by learning-to-rank models which learn the preference between items during training. However, they score items independent of other items at runtime. Although the items placed at top of the results by such scoring functions may be independently optimal, they can be sub-optimal as a set. This may lead to a mismatch between the ideal distribution of items in the top results vs what is actually impressed. In this paper, we present methods to address the purchase-impression gap observed in top search results on eCommerce sites. We establish the ideal distribution of items based on historic shopping patterns. We then present a sequential reranker that methodically reranks top search results produced by a conventional pointwise scoring ranker. The reranker produces a reordered list by sequentially selecting candidates trading off between their independent relevance and potential to address the purchase-impression gap by utilizing specially constructed features that capture impression distribution of items already added to a reranked list. The sequential reranker enables addressing purchase impression gap with respect to multiple item aspects. Early version of the reranker showed promising lifts in conversion and engagement metrics at eBay. Based on experiments on randomly sampled validation datasets, we observe that the reranking methodology presented produces around 10% reduction in purchase-impression gap at an average for the top 20 results, while making improvements to conversion metrics.
Expert finding is an information retrieval task that is concerned with the search for the most knowledgeable people with respect to a specific topic, and the search is based on documents that describe people's activities. The task involves taking a user query as input and returning a list of people who are sorted by their level of expertise with respect to the user query. Despite recent interest in the area, the current state-of-the-art techniques lack in principled approaches for optimally combining different sources of evidence. This article proposes two frameworks for combining multiple estimators of expertise. These estimators are derived from textual contents, from graph-structure of the citation patterns for the community of experts, and from profile information about the experts. More specifically, this article explores the use of supervised learning to rank methods, as well as rank aggregation approaches, for combing all of the estimators of expertise. Several supervised learning algorithms, which are representative of the pointwise, pairwise and listwise approaches, were tested, and various state-of-the-art data fusion techniques were also explored for the rank aggregation framework. Experiments that were performed on a dataset of academic publications from the Computer Science domain attest the adequacy of the proposed approaches.
The task of expert finding has been getting increasing attention in information retrieval literature. However, the current state-of-the-art is still lacking in principled approaches for combining different sources of evidence in an optimal way. This paper explores the usage of learning to rank methods as a principled approach for combining multiple estimators of expertise, derived from the textual contents, from the graph-structure with the citation patterns for the community of experts, and from profile information about the experts. Experiments made over a dataset of academic publications, for the area of Computer Science, attest for the adequacy of the proposed approaches.
This paper studies a new cognitively motivated semantic typing task, multi-axis event process typing, that, given an event process, attempts to infer free-form type labels describing (i) the type of action made by the process and (ii) the type of object the process seeks to affect. This task is inspired by computational and cognitive studies of event understanding, which suggest that understanding processes of events is often directed by recognizing the goals, plans or intentions of the protagonist(s). We develop a large dataset containing over 60k event processes, featuring ultra fine-grained typing on both the action and object type axes with very large ($10^3\sim 10^4$) label vocabularies. We then propose a hybrid learning framework, P2GT, which addresses the challenging typing problem with indirect supervision from glosses1and a joint learning-to-rank framework. As our experiments indicate, P2GT supports identifying the intent of processes, as well as the fine semantic type of the affected object. It also demonstrates the capability of handling few-shot cases, and strong generalizability on out-of-domain event processes.
We present a new framework for self-supervised representation learning by formulating it as a ranking problem in an image retrieval context on a large number of random views (augmentations) obtained from images. Our work is based on two intuitions: first, a good representation of images must yield a high-quality image ranking in a retrieval task; second, we would expect random views of an image to be ranked closer to a reference view of that image than random views of other images. Hence, we model representation learning as a learning to rank problem for image retrieval. We train a representation encoder by maximizing average precision (AP) for ranking, where random views of an image are considered positively related, and that of the other images considered negatives. The new framework, dubbed S2R2, enables computing a global objective on multiple views, compared to the local objective in the popular contrastive learning framework, which is calculated on pairs of views. In principle, by using a ranking criterion, we eliminate reliance on object-centric curated datasets. When trained on STL10 and MS-COCO, S2R2 outperforms SimCLR and the clustering-based contrastive learning model, SwAV, while being much simpler both conceptually and at implementation. On MS-COCO, S2R2 outperforms both SwAV and SimCLR with a larger margin than on STl10. This indicates that S2R2 is more effective on diverse scenes and could eliminate the need for an object-centric large training dataset for self-supervised representation learning.