Recommender systems help people make choices by suggesting items relevant to their needs and preferences. They trace their origins to the early 1990s. Bellcore hosted a Workshop on “High-Performance Information Filtering” in 1991, which led to a special issue of Communications of the ACM one year later. That special issue included a paper on a then-novel technique called collaborative filtering , which became a foundation of the field in the years that followed. As recommender systems enter their fourth decade, we look back at the field using a publication dataset of more than 2.5 million articles, including more than 50,000 core recommender systems articles. We use this dataset to explore the development of the field and its relationship to various adjacent research fields. We examine the research topics that have evolved within recommender systems over the years and the most influential research produced during this time. Our analysis reveals two largely separate communities of researchers, and we explore how these communities differ in terms of their output and impact. Finally, we discuss how we can identify recent emerging topics using measures of publication and citation momentum, and we present several emerging topics that appear destined to play a significant role in the immediate future of the field.

Preparing for the marathon involves many weeks of dedicated training. Achieving the right balance between building strength and endurance and the need for rest and recovery is a must, if a runner is to arrive at the start-line injury-free and ready to achieve their desired finish-time. However, because most recreational runners rely on generic training plans, they can struggle to find this balance, which can impact their motivation, health, and performance. In this paper, we describe a novel case-based reasoning approach to fine-tuning a runner’s training by recommending training adjustments based on the patterns of similar runners at corresponding points in their marathon training. The approach is designed to target training adjustments that are based on similar runners but with varying race goals, to allow runners to adjust their training for slower or faster finish-times, as their training progresses and motivations change. We evaluate the recommendations produced using a large-scale real-world dataset according to several factors: (i) the plausibility of the recommended training adjustment, (ii) the effectiveness of the adjustment when it comes to achieving a particular performance goal, and (iii) the safety of the adjustment in terms of the degree of risk that it will lead to an injury or otherwise disrupt training. Our findings suggest that plausible, effective, and safe recommendations can be generated for runners when evaluated against a range of race goals.

In this paper, we propose a novel communication-efficient algorithm for distributed matrix factorisation. Our goal is to find a good trade-off between the communication overhead and the overall model training time. In our setting, the training data is distributed across multiple servers that aim to learn a joint machine learning model. In contrast to standard distributed computation, due to privacy concerns, the participating servers are not allowed to share raw data, however, sharing of the non-personal model parameters is allowed. We investigate the draw-backs of traditional strongly coordinated distributed techniques and compare them to weakly coordinated gossip approaches. The advantage of strongly coordinated approaches is that the learning process closely mimics that of a centralised algorithm and hence this approach can keep the overall training time at a minimum. However, this is at the expense of a large communication footprint of the algorithm. On the other hand, the weakly coordinated gossip approach offers a communication efficient solution that can take a large amount of training time to reach a good accuracy. As a solution, we develop a hybrid approach combining the above two approaches. We apply the hybrid approach on a latent factor model solving a top-N recommendation problem and we show that the hybrid approach achieves good accuracy in relatively short training time with minimal communication overhead particularly on very sparse data.

In the area of natural language processing, various financial datasets have informed recent research and analysis including financial news, financial reports, social media, and audio data from earnings calls. We introduce a new, large-scale multi-modal, text-audio paired, earnings-call dataset named MAEC, based on S&P 1500 companies. We describe the main features of MAEC, how it was collected and assembled, paying particular attention to the text-audio alignment process used. We present the approach used in this work as providing a suitable framework for processing similar forms of data in the future. The resulting dataset is more than six times larger than those currently available to the research community and we discuss its potential in terms of current and future research challenges and opportunities. All resources of this work are available at https://github.com/Earnings-Call-Dataset/

Federated learning (FL) is quickly becoming the de facto standard for the distributed training of deep recommendation models, using on-device user data and reducing server costs. In a typical FL process, a central server tasks end-users to train a shared recommendation model using their local data. The local models are trained over several rounds on the users' devices and the server combines them into a global model, which is sent to the devices for the purpose of providing recommendations. Standard FL approaches use randomly selected users for training at each round, and simply average their local models to compute the global model. The resulting federated recommendation models require significant client effort to train and many communication rounds before they converge to a satisfactory accuracy. Users are left with poor quality recommendations until the late stages of training. We present a novel technique, FedFast, to accelerate distributed learning which achieves good accuracy for all users very early in the training process. We achieve this by sampling from a diverse set of participating clients in each training round and applying an active aggregation method that propagates the updated model to the other clients. Consequently, with FedFast the users benefit from far lower communication costs and more accurate models that can be consumed anytime during the training process even at the very early stages. We demonstrate the efficacy of our approach across a variety of benchmark datasets and in comparison to state-of-the-art recommendation techniques.

Training for the marathon is a complex problem. In order to run an optimal time, runners must find the right workload for their current abilities and identify the correct balance between the hard work and rest throughout their training programmes. We propose a recommender system that will help guide runners through the weeks leading up to the marathon. Using a large sample of marathon training data (8730 runners), we generate user profiles that capture both a runner’s current fitness and training levels, and leverage this information to generate tailored recommendations for future weeks of training. We investigate patterns of successful runners to determine how best to schedule recommendations and training to allow for improvement in fitness levels alongside adequate rest.

Millions of people participate in marathon events every year, typically devoting at least 12-16 weeks to building their endurance and fitness so that they can safely complete these gruelling 42.2km races. Most runners follow a training plan that is tailored to their expected finish-time (e.g. sub-4 hours or 4-5 hours), and these plans will prescribe a complex mixture of training sessions to help them achieve these times. However, such plans cannot adapt to the individual needs (fitness levels, changing goals, personal preferences) of runners, providing only broad training guidance rather than more personalised support. The development of wearable sensors and mobile fitness applications facilitates the collection of a large amount of training data from runners. In this paper, we propose a recommender system that utilizes such training data to deliver more personalised training advice to runners, using ideas from case-based reasoning to reuse and adapt the training habits of similar runners. Explainability plays a significant role in this type of system, and we also describe how the predictions and recommendation advice can be presented to runners. An initial off-line evaluation is presented based on a large-scale, real-world dataset.

Nowadays we commonly have multiple sources of data associated with items. Users may provide numerical ratings, or implicit interactions, but may also provide textual reviews. Although many algorithms have been proposed to jointly learn a model over both interactions and textual data, there is room to improve the many factorization models that are proven to work well on interactions data, but are not designed to exploit textual information. Our focus in this work is to propose a simple, yet easily applicable and effective, method to incorporate review data into such factorization models. In particular, we propose to build the user and item embeddings within the topic space of a topic model learned from the review data. This has several advantages: we observe that initializing the user and item embeddings in topic space leads to faster convergence of the factorization algorithm to a model that out-performs models initialized randomly, or with other state-of-the-art initialization strategies. Moreover, constraining user and item factors to topic space allows for the learning of an interpretable model that users can visualise.

The volatility forecasting task refers to predicting the amount of variability in the price of a financial asset over a certain period. It is an important mechanism for evaluating the risk associated with an asset and, as such, is of significant theoretical and practical importance in financial analysis. While classical approaches have framed this task as a time-series prediction one – using historical pricing as a guide to future risk forecasting – recent advances in natural language processing have seen researchers turn to complementary sources of data, such as analyst reports, social media, and even the audio data from earnings calls. This paper proposes a novel hierarchical, transformer, multi-task architecture designed to harness the text and audio data from quarterly earnings conference calls to predict future price volatility in the short and long term. This includes a comprehensive comparison to a variety of baselines, which demonstrates very significant improvements in prediction accuracy, in the range 17% - 49% compared to the current state-of-the-art. In addition, we describe the results of an ablation study to evaluate the relative contributions of each component of our approach and the relative contributions of text and audio data with respect to prediction accuracy.

We propose marathon running as a novel domain for recommender systems and machine learning. Using high-resolution marathon performance data from multiple marathon races (n = 7931), we build in-race recommendations for runners. We show that we can outperform the existing techniques which are currently employed for in-race finish-time prediction, and we demonstrate how such predictions may be used to make real time recommendations to runners. The recommendations are made at critical points in the race to provide personalised guidance so the runner can adjust their race strategy. Through the association of model features and the expert domain knowledge of marathon runners we generate explainable, adaptable pacing recommendations which can guide runners to their best possible finish time and help them avoid the potentially catastrophic effects of hitting the wall.

Conventional approaches to matrix factorisation (MF) typically rely on a centralised collection of user data for building a MF model. This approach introduces an increased risk when it comes to user privacy. In this short paper we propose an alternative, user-centric, privacy enhanced, decentralised approach to MF. Our method pushes the computation of the recommendation model to the user's device, and eliminates the need to exchange sensitive personal information; instead only the loss gradients of local (device-based) MF models need to be shared. Moreover, users can select the amount and type of information to be shared, for enhanced privacy. We demonstrate the effectiveness of this approach by considering different levels of user privacy in comparison with state-of-the-art alternatives.

Recommender systems (RS) share many features and objectives with reinforcement learning (RL) systems. The former aim to maximise user satisfaction by recommending the right items to the right users at the right time, the latter maximise future rewards by selecting state-changing actions in some environment. The concept of an RL gym has become increasingly important when it comes to supporting the development of RL models. A gym provides a simulation environment in which to test and develop RL agents, providing a state model, actions, rewards/penalties etc. In this paper we describe and demonstrate the PyRecGym gym, which is specifically designed for the needs of recommender systems research, by supporting standard test datasets (MovieLens, Yelp etc.), common input types (text, numeric etc.), and thereby offering researchers a reproducible research environment to accelerate experimentation and development of RL in RS.

Recommender systems are omni-present in our every day lives, guiding us through the vast amount of information available. However, in the academic world, personalised recommendations are less prominent, leaving students to navigate through the typically large space of available courses and modules manually. Since it is crucial for students to make informed choices about their learning pathways, we aim to improve the way students discover elective modules by developing a hybrid recommender system prototype that is specifically designed to help students find elective modules from a diverse set of subjects. We can improve the discoverability of long-tail options and help students broaden their horizons by combining notions of similarity and diversity.

We describe a novel, multi-task recommendation model, which jointly learns to perform rating prediction and recommendation explanation by combining matrix factorization, for rating prediction, and adversarial sequence to sequence learning for explanation generation. The result is evaluated using real-world datasets to demonstrate improved rating prediction performance, compared to state-of-the-art alternatives, while producing effective, personalized explanations.

Collaborative filtering (CF) is a common recommendation approach that relies on user-item ratings. However, the natural sparsity of user-item rating data can be problematic in many domains and settings, limiting the ability to generate accurate predictions and effective recommendations. Moreover, in some CF approaches latent features are often used to represent users and items, which can lead to a lack of recommendation transparency and explainability. User-generated, customer reviews are now commonplace on many websites, providing users with an opportunity to convey their experiences and opinions of products and services. As such, these reviews have the potential to serve as a useful source of recommendation data, through capturing valuable sentiment information about particular product features. In this paper, we present a novel deep learning recommendation model, which co-learns user and item information from ratings and customer reviews, by optimizing matrix factorization and an attention-based GRU network. Using real-world datasets we show a significant improvement in recommendation performance, compared to a variety of alternatives. Furthermore, the approach is useful when it comes to assigning intuitive meanings to latent features to improve the transparency and explainability of recommender systems.

We describe a novel application for recommender systems -- helping marathon runners to run a new personal-best race-time -- by predicting a challenging, but achievable target-time, and by recommending a tailored race-plan to achieve this time. A comprehensive evaluation of prediction accuracy and race-plan quality is provided using a large-scale dataset with almost 400,000 runners from the last 12 years of the Chicago marathon.

Recommender systems learn about our preferences to make targeted suggestions. In this paper we outline a novel game-with-a-purpose designed to infer preferences at scale as a side-effect of gameplay. We evaluate the utility of this data in a recommendation context as part of a small live-user trial.

This paper describes a novel approach to product recommendation that is based on opinionated product descriptions that are automatically mined from user-generated product reviews. We present a recommendation ranking strategy that combines similarity and sentiment to suggest products that are similar but superior to a query product according to the opinion of reviewers. We demonstrate the benefits of this approach across a variety of Amazon product domains.

In this paper we consider the application of content-based recommendation techniques to web curation services which allow users to curate and share topical collections of content (e.g. images, news, web pages etc.). Curation services like Pinterest are now a mainstay of the modern web and present a range of interesting recommendation challenges. In this paper we consider the task of recommending collections to users and evaluate a range of different content-based techniques across a variety of content signals. We present the results of a large-scale evaluation using data from the Scoop.it web page curation service

User generated reviews are now a familiar and valuable part of most e-commerce sites since high quality reviews are known to influence purchasing decisions. In this demonstration we describe work on the Reviewer's Assistant (RA), which is a recommendation system that is designed to help users to write better quality reviews. It does this by suggesting relevant topics that they may wish to discuss based on the product they are reviewing and the content of their review so far.

Our research involves developing technology and techniques that apply the vast sea of real-time web data to interesting problems and topics. In this demo, we will present the on- going development of a novel real-time search and discovery service named Yokie (http://yok.ie, early technology description originally published in [1]). Yokie uses the large volume of hyperlink-laden messages on social networks like Twitter as the basis of its content and ranking systems. Curated sets of users (or "Search Parties") form the basis of sourcing the content from the networks, and the metadata of the containing messages form the basis of ranking and contextual retrieval of the hyperlinks. Each hyperlink is in- dexed with a compound set of terms from multiple tweets (should the given hyperlink be shared more than once). This indexing step is a novel example of collaborative tagging of resources. The application is live with more than 100 users, who have performed approximately 1000 queries. We will demonstrate the main techniques and novel ranking and re- trieval techniques and user features.

The purpose of this paper is to provide a deployment update for the HeyStaks social search system which uses recommendation techniques to add collaboration to mainstream search engines such as Google, Bing, and Yahoo. We describe our the results of initial deployments, including an assessment of the quality of HeyStaks' recommendations, and highlight some lessons learned in the marketplace

The explosive growth of online social networks in recent times has presented a powerful source of information to be utilised in personalised recommendations. Unsurprisingly there has already been a large body of work completed in the recommender system field to incorporate this social information into the recommendation process. In this paper we examine the practice of leveraging a user's social graph in order to generate recommendations. Using various neighbourhood selection strategies, we examine the user satisfaction and the level of perceived trust in the recommendations received.

Although collaborative searching is not supported by mainstream search engines, recent research has highlighted the inherently collaborative nature of many Web search tasks. In this article, we describe HeyStaks, a collaborative Web search framework that is designed to complement mainstream search engines. At search time, HeyStaks learns from the search activities of other users and leverages this information to generate recommendations based on results that others have found relevant for similar searches. The key contribution of this article is to extend the HeyStaks social search model by considering the search expertise, or reputation, of HeyStaks users and using this information to enhance the result recommendation process. In particular, we propose a reputation model for HeyStaks users that utilise the implicit collaboration events that take place between users as recommendations are made and selected. We describe a live-user trial of HeyStaks that demonstrates the relevance of its core recommendations and the ability of the reputation model to further improve recommendation quality. Our findings indicate that incorporating reputation into the recommendation process further improves the relevance of HeyStaks recommendations by up to 40%.

No abstract available.

Poster presented at the 4th ACM Conference on Recommender Systems (RecSys 2010), Barcelona, Spain, September 26-30, 2010

Recently the world of the web has become more social and more real-time. Facebook and Twitter are perhaps the exemplars of a new generation of social, real-time web services and we believe these types of service provide a fertile ground for recommender systems research. In this paper we focus on one of the key features of the social web, namely the creation of relationships between users. Like recent research, we view this as an important recommendation problem -- for a given user, UT which other users might be recommended as followers/followees -- but unlike other researchers we attempt to harness the real-time web as the basis for profiling and recommendation. To this end we evaluate a range of different profiling and recommendation strategies, based on a large dataset of Twitter users and their tweets, to demonstrate the potential for effective and efficient followee recommendation.

User-generated reviews are a common and valuable source of product information, yet little attention has been paid as to how best to present them to end-users. In this paper, we describe a classification-based recommender system that is designed to recommend the most helpful reviews for a given product. We present a large-scale evaluation of our approach using TripAdvisor hotel reviews, and we show that our approach is capable of suggesting superior reviews compared to a number of alternative recommendation benchmarks.

Recommending news stories to users, based on their preferences, has long been a favourite domain for recommender systems research. In this paper, we describe a novel approach to news recommendation that harnesses real-time micro-blogging activity, from a service such as Twitter, as the basis for promoting news stories from a user's favourite RSS feeds. A preliminary evaluation is carried out on an implementation of this technique that shows promising results.

In this paper we report on our work to date concerning the development of a course recommender system for University College Dublin's on-line enrollment application. We outline the factors that influence student choices and propose solutions to address some of the key considerations that are identified. We empirically evaluate our approach using historical student enrolment data and show that promising performance is achieved with our initial design.

It is likely that mobile phones will soon come to rival more traditional devices as the primary platform for information access. Consequently, it is important to understand the emerging information access behavior of mobile Internet (MI) users especially in relation to their use of mobile handsets for information browsing and query-based search. In this article, we describe the results of a recent analysis of the MI habits of more than 600,000 European MI users, with a particular emphasis on the emerging interest in mobile search. We consider a range of factors including whether there are key differences between browsing and search behavior on the MI compared to the Web. We highlight how browsing continues to dominate mobile information access, but go on to show how search is becoming an increasingly popular information access alternative especially in relation to certain types of mobile handsets and information needs. Moreover, we show that sessions involving search tend to be longer and more data-rich than those that do not involve search. We also look at the type of queries used during mobile search and the way that these queries tend to be modified during the course of a mobile search session. Finally we examine the overlap among mobile search queries and the different topics mobile users are interested in.

We describe and evaluate an approach to capturing and re-using search expertise within a community of like minded searchers, such as the employees of a company or organisation. Within knowledge based industries, search expertise - the ability to quickly and accurately locate information according to a specific information need - is an important corporate asset and in our approach we attempt to capture this knowledge by mining the title and snippet texts of results that have been selected by community members in response to their queries. Our assumption is that the snippet text of a result must play a role in helping users to judge the initial relevance of that result and that the snippet terms of selected results must contain especially informative terms about the goals and preferences of the searchers. In other words, results are selected because the user recognises certain combinations of terms in their snippets which are related to their information needs. Our approach seeks to build a community-based snippet index that reflects the evolving interests of a group of searchers. This index is then used to re-rank the results returned by some underlying search engine by boosting the ranking of key results that have been frequently selected for similar queries by community members in the past.

We describe and evaluate an approach to personalizing Web search that involves post-processing the results returned by some underlying search engine so that they re .ect the interests of a community of like-minded searchers.To do this we leverage the search experiences of the community by mining the title and snippet texts of results that have been selected by community members in response to their queries. Our approach seeks to build a community-based snippet index that re .ects the evolving interests of a group of searchers. This index is then sed to re-rank the results returned by the underlying search engine by boosting the ranking of key results that have been freq ently selected for similar q eries by community members in the past.

Many systems use past behavior, preferences and environmental factors to attempt to predict user navigation on the Internet. However we believe that many of these models have shortcomings, in that they do not take into account that users may have many different sets of preferences. Here we investigate an environmental factor, namely time, in making predictions about user navigation. We present methods for creating temporal rules that describe user navigation patterns. We also show the benefit of using these rules to predict user navigation and also show the benefits of these models over traditional methods. An analysis is carried out on a sample of usage logs for Wireless Application Protocol (WAP) browsing, and the results of this analysis verify our hypothesis.

Search engine results are usually presented in some form of text summary (e.g., document title, some snippets of the page's content, a URL, etc). Based on the information contained within these summaries users make relevance judgments about what links best suit their information needs. Current research suggests that these relevance judgments are in the service of some search strategy. In this paper, we model two different search strategies (the comparison and threshold strategies) and determine how well they fit data gathered from an experiment on user search within a simulated Google environment.

Technological developments in content-based analysis of digital video information are undergoing much progress, with ideas for fully automatic systems now being proposed and demonstrated. Yet because we do not yet have robust operational video retrieval systems that can be deployed and used, the usual HCI practise of conducting a usage study and an informed iterative system design is thus not possible. Físchlár-News is one of the first automatic, content-based broadcast news analysis and archival systems that process broadcast news video so that users can search, browse, and play it in an easy-to-use manner with a conventional web browser. The system incorporates a number of state-of-the-art research components, some of which are not yet considered mature technology, yet it has been built to be robust enough to be deployed to users who are interested in access to daily news throughout a university campus. In this article we report and discuss a user-evaluation study conducted with 16 users, each of whom utilized the system freely for a one month period. Results from a detailed qualitative analysis are presented, looking at collected questionnaires, incident diaries, and interaction-log data. The findings suggest that our users employed the system in conjunction with their other news update methods, such as watching TV news at home and browsing online news websites at their workplace, their major concerns being up-to-dateness and coverage of the news content. They tried to accommodate the system to fit their established web browsing habits, and they found local news content and the ability to play self-contained news stories on their desktop as major values of the system. Our study also resulted in a detailed wishlist of new features which will help in the further development of both our and others' systems.

The I-SPY meta-search engine uses a technique called collaborative Web search to leverage the past search behaviour (queries and selections) of a community of users in order to promote search results that are relevant to the community. In this paper we describe recent studies to clarify the benefits of this approach in situations when the behaviour of users cannot be relied upon in terms of their ability to consistently select relevant results during search sessions.

There are many systems that attempt to predict user navigation on the Internet through the use of past behavior, preferences and environmental factors. We believe that many of these models have shortcomings, in that they do not take into account that users may have many different sets of preferences, specifically, we investigate time as an environmental factor in making predictions about user navigation. We present a method for segmenting log files in order to learn time dependent models to predict user navigation patterns and show the benefits of these models over traditional methods. An analysis is carried out on a sample of usage logs for wireless application protocol (WAP) browsing, and the results of this analysis verify our hypothesis.

We present an innovative approach to personalized Web search that exploits the search behaviour of a community of users to re-rank future result-lists according to the implied preferences of this group. Evaluation results demonstrate the precision and recall benefits of our collaborative search technique and we show how personalization can be achieved without the need for individual user profiling.

Recommender systems bring together ideas from information retrieval and filtering, user profiling, adaptive interfaces and machine learning in an attempt to offer users more personalized and responsive search systems. Conversational recommenders guide a user through a sequence of iterations, suggesting specific items, and using feedback from users to refine their suggestions in subsequent iterations. Different recommender systems look for different types of feedback from users. In this paper we examine the role of critiquing, a form of feedback in which the user indicates a preference over a particular feature of a recommended item. For example, when shopping for a PC a user might indicate that they like the current suggestion but they are looking for something "cheaper"; "cheaper" is a critique over the price feature of the PC case. Sometimes it is useful to critique multiple features simultaneously (compound critiques). In this paper we describe how a recommender can automatically discover useful compound critiques during the recommendation session and how these critiques can be used to improve recommendation efficiency.

Case based reasoning (CBR) is an artificial intelligence technique that emphasises the role of past experience during future problem solving. New problems are solved by retrieving and adapting the solutions to similar problems, solutions that have been stored and indexed for future reuse as cases in a case-base. The power of CBR is severely curtailed if problem solving is limited to the retrieval and adaptation of a single case, so most CBR systems dealing with complex problem solving tasks have to use multiple cases. The paper describes and evaluates the technique of hierarchical case based reasoning, which allows complex problems to be solved by reusing multiple cases at various levels of abstraction. The technique is described in the context of Deja Vu, a CBR system aimed at automating plant-control software design.