Recommender Systems: Value, Methods, Measurement [slides]
Dietmar Jannach, Univ. Klagenfurt, Austria
Recommender systems are one of the most visible and successful applications of AI/ Machine Learning today. They are not only helpful for end users to discover things they might be interested in, they can also lead to a substantial value for business, e.g., by stimulating more sales or by increasing customer engagement. In this talk, we will first discuss the various types of value recommenders can have for different stakeholders in practice. We will review case studies of recommender systems deployments and discuss how their effects can be measured. We will then move on to the academic perspective, where we describe common problem abstractions, the main technical approaches, as well as evaluation procedures. The talk ends with a discussion of open challenges in our field and an outlook on possible future directions.
FAT in Recommendation Systems [slides]
Denis Parra, Pontificia Univ. Católica de Chile
In recent years we have experienced an increasing deployment in our daily lives of applications based on Artificial Intelligence technology. Some of these applications are truly amazing, such as self driving cars, translation systems, and automatic detection of illnesses. However, we have also seen applications which are disrupting our daily lives with unintended consequences, such as recommender systems which polarize public opinion and face recognition applications which can hinder our privacy. As the researcher Michael Jordan stated: “Just as early buildings and bridges sometimes fell to the ground — in unforeseen ways and with tragic consequences — many of our early societal-scale inference-and-decision-making systems are already exposing serious conceptual flaws.” In this context, a serious and systematic study of FAT (Fairness, Accountability and Transparency) in AI is a critical. The popularity of the recently created FAT conference and several related workshops in academic conferences, especially those related to topics in artificial intelligence, show evidence of how researchers and developers, as well as the society at large, is becoming aware of the effects of AI technology. In terms of recommendation systems, algorithmic transparency has been a topic of study in for at least 10 years, but only recently this area has attracted strong attention in the community. In this tutorial, Prof. Parra will present a general overview of FAT in AI, to then focus on FAT for recommendation systems with a theoretical presentation as well as practical activities.
Recommender Systems Evaluation Beyond Accuracy [slides]
Pablo Castells, Univ. Autónoma de Madrid, Spain
Algorithmic development in recommender systems goes hand in hand with evaluation. For an initial period of time the recommendation problem was addressed as an issue of matching user preferences as accurately as possible. It was soon understood that while certainly a basic requirement, accuracy alone is not enough to procure useful recommendations. Novelty and diversity, for instance, have been identified as key properties to draw user value out of accurately predicted user interests. We will review notions, motivations, different angles and metrics developed in such broader perspectives, as well as some algorithmic implications. Time permitting, we will relate this outlook to recent advances in understanding and dealing with biases in recommender system evaluation and algorithms.
The past few years have seen the tremendous success of deep neural networks in a number of complex machine learning tasks such as computer vision, natural language processing and speech recognition. For these reasons, deep learning have been already well stablished in the area, while we now start to see deep neural networks deliver on their potential for improvement in recommendation systems technology. The aim of the tutorial is dual: 1) to briefly introduce deep learning techniques that have been and are used in recommender systems such as recurrent neural networks and convolutional networks 2) to present the current state-of-the-art collaborative filtering and content-based methods that use deep learning techniques to provide recommendations. The tutorial will focus on the deep learning methods that are mostly used in recommender systems, from recurrent networks to convolutional networks and transformer models. We will also touch upon the emerging topic of deep reinforcement learning for recommender systems. The tutorial does not require an in depth prior knowledge in deep learning since there will be an introduction to the relevant techniques, e.g., recurrent neural networks, convolutional networks, word2vec embeddings.
In today’s world, a recommendation system is an integral and expected component in any online service. This is especially true in a large scale entertainment streaming service such as Pandora. Users of such services have grown accustomed to a personalized experience not only fitting their expressed interest, but also their situations, through recommendations. In constructing these recommendation algorithms and systems, practitioners often need to consider a complex array of constraints such as different item formats, diverse preferences for degree of exploration, query originations (typed search in mobile, voice command), user tiers (free vs. premium) and a variety of clients. In addition, recommendations are expected to come with low latency and scale to millions of users. Our lives as practitioners are made simultaneously more interesting and more hectic by these constraints. We use four use case scenarios to illustrate the complexities faced by industrial scale recommendation systems, largely generalized from experience in building the largest streaming audio recommender systems at Pandora. Since 2005, Pandora’s music recommendation platform has powered the well known machine learning driven personalized radio product and premium on demand product, complete with support to a host of clients from web, mobile, and voice platforms for 65+ million monthly listeners in the U.S. Firstly, we share the content understanding aspect — genome tagging, keyword tagging, and combination of expert tagging and automatic tagging to form the most important knowledge building blocks of recommendations and explanations. Secondly, we share user modeling, from scalability issues from detailed user actions, to user feature extraction for ML systems. Thirdly, we discuss algorithms, ranging from core recommender model and their deployment, to ensemble re-ranking systems that could take shape of a contextual multi-armed bandit algorithm and possibly reinforcement learning. Finally, we discuss new challenges brought on by recently popular voice interaction platforms and our approaches. In this tutorial, we emphasize on the comprehensive view of a practical and scalable system, complete with offline and online evaluations. Let us learn, discuss and explore ideas together!
As we wrap up the Latin American School on Recommender Systems, we take a long-term look at recommender systems as tools for assisting and influencing human decision-making. Starting from the early origins of these system through today’s commercial applications, we look at how recommender systems design relates to human perception, cognition, and tasks. We then look at how recommender systems can be evaluated and tuned to achieve human-focused objectives, and explore open questions that can inform a future research agenda.