Collaborative Recommendations

The following sections are included:

• Preface
• Contents

This chapter provides a self-contained overview on the basics of collaborative filtering recommender systems. It covers two main classes of recommendation scenarios. In the classical matrix completion problem formulation, the task of an algorithm is to make longer-term relevance predictions given a user-item rating matrix. In session-based recommendation scenarios, the goal is to predict relevant items given a user’s observed short-term behavior. From an algorithmic perspective, the chapter particularly focuses on neighborhood-based methods, which were proposed in the early days of collaborative filtering and which are still relevant today. The chapter addresses the entire life-cycle of algorithm development and also discusses libraries, datasets, and implementation aspects. Furthermore, it covers evaluation issues and reflects on today’s research methodology in the field. Overall, the chapter shall serve as a starting point for readers, providing pointers to more detailed discussion of the various aspects regarding the design and evaluation of collaborative filtering recommender systems.

The following sections are included:

• Introduction
• Problem formulation
• SVD-based models
• Weighted low-rank approximation
• Advanced methods
• Practical aspects
• Conclusion
• References

After achieving great performance on complex domains, such as computer vision, natural language processing, speech recognition, reinforcement learning, etc., recommender systems research also turned towards deep learning recently. Although the recsys community was a late adopter and has been only focusing on deep learning since late 2015 / early 2016, it has already became one of the most promising and thus highly popular topic within the research community. Deep learning has the potential to revolutionize the field, just as it did with other domains. The aim of this chapter is to overview recent advances in the intersection of deep learning and recommender systems…

Hybrid collaborative recommender systems were developed to increase performance of single recommenders similarly to the ensemble machine learning methods. There are several types of hybridization that can be utilized in various scenarios. In this chapter, we summarize the overview of hybrid collaborative recommender systems. We focus on how they are beneficial with respect to the standard collaborative filtering techniques and how they can be utilized to solve the main issues of the collaborative filtering techniques. Examples of frameworks and libraries that implement existing recommendation approaches are complemented with the comprehensive list of datasets available (including state-of-the-art results reported on these datasets). Evaluation process of recommender systems is discussed followed by the examples of the evaluation frameworks. We summarize pros and cons of several hybrid approaches and conclude with suggestions for practical implementations.

With increasingly more complex information spaces on the Internet, recommender systems have emerged as critical tools to alleviate information overload and to assist decision making. Although such systems generate personalized recommendations based on users’ learned preferences, most recommenders do not consider the fact that a user’s decision or preferences may vary across contexts and situations. Context-aware recommender systems, on the other hand, not only adapt to user preferences, but also consider the contextual situations. This chapter provides a broad introduction to Context-Aware Recommender Systems (CARS). We especially focus on algorithmic approaches for integrating context into the recommendation framework, including approaches based on context selection and context-aware collaborative filtering. We also discuss evaluation strategies for CARS; available data sets and open-source libraries; and lessons learnt from the practical applications.

Group recommender systems have been developed to produce suggestions in contexts in which more than one person is involved in the recommendation process. As it happens with the systems that produce recommendation for single users, collaborative algorithms have been employed for most of the group-based solutions presented in the literature. Producing group recommendations, however, is not trivial and a system has to deal with additional aspects, such as the combination of the individual preferences into group preferences, or the prediction of the ratings straight for a group. In this chapter, we explore collaborative group recommender systems, analyzing the different families of approaches and architectural solutions available to build them, the algorithms that can be developed to produce the recommendations, the existing solutions in the literature, and the current open challenges in this area.

The process of collecting preferences from users is fundamental during the normal operational life of a recommender system. The preference elicitation strategy can affect both the “user utility” (how well the system can make good recommendations to the new user who is undergoing the elicitation process) and the “system utility” (how well the system can provide good recommendations to all users, given what it learns from the new users). Not only do recommender systems need to gather information from users; they also need this information to be reliable and noiseless, as inconsistencies in user preferences limit prediction accuracy…

A prerequisite for implementing collaborative filtering recommender systems is the availability of users’ preferences data. This data, typically in the form of ratings, is exploited to learn the tastes of the users and to serve them with personalized recommendations. However, there may be a lack of preference data, especially at the initial stage of the operations of a recommender system, i.e., in the Cold Start phase. In particular, when a new user has not yet rated any item, the system would be incapable of generating relevant recommendations for this user. Or, when a new item is added to the system catalogue and no user has rated it, the system cannot recommend this item to any user…

Recommender Systems have been a popular research topic within personalized systems and information retrieval since the mid nineties. Throughout this time, various models of recommendation have been developed, e.g., approaches using collaborative filtering for purposes such as retrieval of ranked lists of items for consumption, or for the rating prediction task (which was made very popular through the Netflix prize). Today, the use of recommender systems has spread to a very wide area of topics, including personalized healthcare, online news portals, food, social networks, exercise, jobs, investment, transportation, shopping, etc. Given the various situations recommendations can be applied to, it follows that evaluation of these systems needs to be tailored to the specific setting, domain, user-base, context, etc. This chapter aims to give an overview of some of the more commonly used evaluation methods and metrics used for various types of recommendation techniques. We also provide a summary of the available resources in this topic, in addition to some practical considerations, experimental results, and future directions about evaluation in recommendation.

Recommender systems research, traditionally focused on accuracy, is currently paying more and more attention to additional factors for evaluating the perceived quality and usefulness of recommendation lists. In this paper, we present a survey of the most important dimensions, other than accuracy, usually taken into account in the collaborative filtering literature. We survey beyond-accuracy objectives, i.e. novelty, diversity and serendipity, and the main techniques for increasing them. Moreover, we discuss possible undesired biases occurring in collaborative filtering algorithms, and how to effectively deal with them.

Recommender systems are ubiquitous; they are foundational to a wide variety of industries ranging from media companies such as Netflix to e-commerce companies such as Amazon. As recommender systems continue to permeate the marketplace, we observe two major shifts which must be addressed. First, the amount of data used to provide quality recommendations grows at an unprecedented rate. Secondly, modern computer architectures display great processing capabilities that significantly outpace memory speeds. These two trend shifts must be taken into account in order to design recommendation systems that can efficiently handle the amount of available data by distributing computations in order to take advantage of modern parallel architectures. In this chapter, we present ways to scale popular collaborative recommendation methods via parallel computing.

How robust is a recommender to the presence of corrupt data in the training dataset? To what extent will the recommendation algorithm be affected by such data? These are the basic questions of a analysis of recommender algorithms. They are particularly relevant in the context of maliciously injected corrupt data, where the data injection is designed to deliberately distort the recommendation output for particular sets of users and items. In the context of online recommenders, where the creation of fake identities may be relatively low-cost, such a robustness attack against a recommender system is of genuine concern. Robustness is also of some concern in the context of privacy preserving recommender algorithms in which user rating profiles are distorted in order to protect private data. In this chapter, we review research carried out since the early 2000s on robustness attacks and practical defences against them. We introduce a new library for robustness analysis incorporated into the Ranksys recommender systems framework, in which the main attacks proposed in the state-of-the-art are implemented and can be simply tested against a range of recommender algorithms.

Today, recommender systems are ubiquitous in our daily life and are becoming an indispensable decision supporting tool. Unfortunately, these systems are extremely greedy for (sensitive) personal data and they incur enormous economic and political incentives for malicious players to misbehave. An increasing number of privacy breaches and system abuses have been disclosed over the time. In this chapter, we aim at a comprehensive investigation into the privacy issues associated with recommender systems, with a focus on the subtle conflicting-yet-complementary relationships between privacy, robustness, and transparency. Taking several existing privacy-preserving solutions, including cryptographic and differential privacy based ones, we comparatively analyse their strengths and weaknesses. At the end, we outline some research directions to design pragmatic privacy-preserving recommender systems which also facilitate other important properties such as robustness and transparency.

With the ever-increasing number of TV shows and movies available to the users, facilitating content discovery is key toward creating a better user experience for an entertainment platform. Comcast is one of the largest cable providers in the United State with millions of customers. In this chapter, we provide an overview of the large scale recommendation system and its building blocks that power Comcast X1 entertainment platform. We discuss some of the challenges in building a TV and movie recommendation system and review a set of product features for content discovery on Comcast X1 platform. We briey describe the historical evolution of the recommendation system at Comcast and provide some insights and results on evaluating the recommendation system with offline and online metrics. We conclude this chapter with some recent areas of research and development.

Today’s online music services like Spotify provide their listeners with different types of music recommendations, e.g., in the form of weekly recommendations or personalized radio stations. Such recommendations are often based, at least in parts, on collaborative filtering techniques. In this chapter, we first review the different types of music recommendations that can be found in practice and discuss the specific challenges of the domain. Next, we discuss technical approaches for the problems of music discovery and next-track recommendation in more depth, with a specific focus on their practical application at Spotify. Finally, we further elaborate on open challenges in the field and revisit the specific problems of evaluating music recommendation systems in academic environments.

The increasingly fast development and expansion of recommender systems technology over the last two and a half decades, along with the exponential growth of online social networks in the last few years, has given place to the concurrence of the two areas in several directions. The present chapter focuses on a specific area within this confluence: the recommendation of people to connect with in social networks. We analyze the specifics of contact suggestion as a very particular recommendation task, where both the target users and the target items are people. We give an overview of the most relevant state of the art algorithms in this area, including methods that were originally developed with slightly different problems in mind. We present a global empirical comparison of the reviewed algorithms in order to get a perspective of their comparative performance. We conclude discussing future possible directions for research and development in this area.

There are several reasons why people are interested in getting job recommendations. They might be actively looking for a new job or want to stay informed about their own value in the job market, etc. Additionally, recruiters aim for the right people to see their job postings to find suitable candidates for open positions. Real world job recommender systems face several challenges, though. Job postings are only interesting as long as they are up-to-date and, thus, have to be recommended to the right users as soon as they are published. Furthermore, a job recommender system should aim to support the users in their professional growth and needs to find jobs that do not necessarily match the users’ current profile but have the potential to form the next step in the users’ career. In this chapter, we first introduce the professional social network XING and discuss the challenges it has to tackle when generating job recommendations. Thereafter, we present the recommender’s functionality and discuss how we evaluate new features or extensions. Finally, we conclude with an insight in the recommender system’s architecture that has to serve thousands of requests per minute.

Mendeley Suggest is a recommender system that helps researchers keep up to date with articles in their field and fill gaps in their knowledge. It leverages usage data from the Mendeley reference management system, in which millions of users have compiled personal libraries of relevant articles. In this chapter, we explain how recommendations are generated using user-based collaborative filtering. Much of the development of the live system has focused on keeping users engaged as they repeatedly interact with the system and adapting to their research interests as they add new articles to their libraries. The product has evolved over time through taking into account user feedback, experimenting with new approaches, and making data-driven decisions. As Mendeley Suggest continues to develop, we expect to incorporate additional domain-specific features to supplement collaborative filtering and provide even more useful recommendations.

Massive open online courses (MOOCs) have recently gained a huge users’ attention on the Web. They are considered as a highly promising form of teaching from leading universities such as Stanford and Berkeley. MoocRec.com is a web site that recommends courses to users so that, they can acquire those skills, that are expected from their ideal job posting. MoocRec’s recommendation engine is based on a Matrix Factorization (MF) model, which exploits information from external resources (i.e., job-skill, user-skill, course-skill, etc.), which are extractred from EdX, Coursera and Linkedin. Based on the aforementioned additional matrices we are able to predict course trends and to make rating predictions for users over courses. Based on these rating predictions, we find the similar neighbors of a target user and we provide to her top-N course recommendations.

The recommendation of food items is important for many reasons. Attaining cooking inspiration via digital sources is becoming evermore popular; as are systems, which recommend other types of food, such as meals in restaurants or products in supermarkets. Researchers have been studying these kinds of systems for many years, suggesting not only that can they be a means to help people find food they might want to eat, but also help them nourish themselves more healthily. This paper provides a summary of the state-of-the-art of so-called food recommender systems, highlighting both seminal and most recent approaches to the problem, as well as important specializations, such as food recommendation systems for groups of users or systems which promote healthy eating. We moreover discuss the diverse challenges involved in designing recsys for food, summarise the lessons learned from past research and outline what we believe to be important future directions and open questions for the field. In providing these contributions we hope to provide a useful resource for researchers and practitioners alike.

The following sections are included:

• Introduction
• Use Cases and General Approach
• Developed Algorithm: Learning to Rank
• System Architecture
• Addressed Challenges and Problems
• Experimental Evaluation
• Conclusions and Future Directions
• Acknowledgments
• References

The following section is included:

• Index