In sequential recommendation scenarios, user intent is a key driver of consumption behavior. However, consumption intents are usually latent and hence, difficult to leverage for recommender systems. Additionally, intents can be of repeated nature (e.g. yearly shopping for christmas gifts or buying a new phone), which has not been exploited by previous approaches. To navigate these impediments we propose the HyperHawkes model which models user sessions via hypergraphs and extracts user intents via contrastive clustering. We use Hawkes Processes to model the temporal dynamics of intents, namely repeated consumption patterns and long-term interests of users. For short-term interest adaption, which is more fine-grained than intent-level modeling, we use a multi-level attention mixture network and fuse long-term and short-term signals. We use the generalized expectation-maximization (EM) framework for training the model by alternating between intent representation learning and optimizing parameters of the long- and short-term modules. Extensive experiments on four real-world datasets from different domains show that HyperHawkes significantly outperforms existing state-of-the-art methods.

Session-based recommendation aims to predict the next item based on a set of anonymous sessions. Capturing user intent from a short interaction sequence imposes a variety of challenges since no user profiles are available and interaction data is naturally sparse. Recent approaches relying on graph neural networks (GNNs) for session-based recommendation use global item relations to explore collaborative information from different sessions. These methods capture the topological structure of the graph and rely on multi-hop information aggregation in GNNs to exchange information along edges. Consequently, graph-based models suffer from noisy item relations in the training data and introduce high complexity for large item catalogs. We propose to explicitly model the multi-hop information aggregation mechanism over multiple layers via shortest-path edges based on knowledge from the sequential recommendation domain. Our approach does not require multiple layers to exchange information and ignores unreliable item-item relations. Furthermore, to address inherent data sparsity, we are the first to apply supervised contrastive learning by mining data-driven positive and hard negative item samples from the training data. Extensive experiments on four different datasets show that the proposed approach outperforms almost all of the state-of-the-art methods.

Explainability in recommender systems is both crucial and challenging. Among the state-of-the-art explanation strategies, counterfactual explanation provides intuitive and easily understandable insights into model predictions by illustrating how a small change in the input can lead to a different outcome. Recently, this approach has garnered significant attention, with various studies employing different metrics to evaluate the performance of these explanation methods. In this paper, we investigate the metrics used for evaluating counterfactual explainers for recommender systems. Through extensive experiments, we demonstrate that the performance of recommenders has a direct effect on counterfactual explainers and ignoring it results in inconsistencies in the evaluation results of explainer methods. Our findings highlight an additional challenge in evaluating counterfactual explainer methods and underscore the need to report the recommender performance or consider it in evaluation metrics.

With the RecSys conference now turning 18 years old, the recommender systems (RS) discipline ventures into adulthood. This workshop serves as a platform for introspection, examining the evolution of RS from its origins in CHI to its current state heavily influenced by and focusing on machine learning. The INTROSPECTIVES workshop aims to foster discussions on the past, present, and future of the RS discipline, inviting the community to reflect on key questions such as the maturation of RS, shifts in research focus, and the impact and success of RS in practice. Topics include the changing landscape of RS problems, the evolving role of RS in addressing choice overload to the current motivations driving RS adoption.

Evaluation plays a vital role in recommender systems—in research and practice—whether for confirming algorithmic concepts or assessing the operational validity of designs and applications. It may span the evaluation of early ideas and approaches up to elaborate implementations of systems integrated into everyday product settings; it may target a wide spectrum of different factors being evaluated. In this special issue, we explore recommender systems evaluation—theory and practice—while considering a diverse set of perspectives. These include recommender systems purposes, stakeholders, methodological approaches, and consequences. The collection of articles in this special issue offers insightful analyses of current recommender system evaluation practices, acknowledging their limitations, and setting out future research directions. As recommender systems evolve, the need for adequate evaluation methods and approaches increases. This special issue sheds light on areas undergoing development or requiring added attention from the research and practitioner communities in recommender systems. The compilation serves as a call to the recommender systems research community, motivating continued research and exploration of evaluation metrics, methods, and strategies.

Recommender systems research and practice are fast-developing topics with growing adoption in a wide variety of information access scenarios. In this article, we present an overview of research specifically focused on the evaluation of recommender systems. We perform a systematic literature review, in which we analyze 57 papers spanning six years (2017–2022). Focusing on the processes surrounding evaluation, we dial in on the methods applied, the datasets utilized, and the metrics used. Our study shows that the predominant experiment type in research on the evaluation of recommender systems is offline experimentation and that online evaluations are primarily used in combination with other experimentation methods, e.g., an offline experiment. Furthermore, we find that only a few datasets (MovieLens, Amazon review dataset) are widely used, while many datasets are used in only a few papers each. We observe a similar scenario when analyzing the employed performance metrics—a few metrics are widely used (precision, normalized Discounted Cumulative Gain, and Recall), while many others are used in only a few papers. Overall, our review indicates that beyond-accuracy qualities are rarely assessed. Our analysis shows that the research community working on evaluation has focused on the development of evaluation in a rather narrow scope, with the majority of experiments focusing on a few metrics, datasets, and methods.

Session-based recommendation aims to predict the next item based on a set of anonymous sessions. Capturing user intent from a short interaction sequence imposes a variety of challenges since no user profiles are available and interaction data is naturally sparse. Recent approaches relying on graph neural networks (GNNs) for session-based recommendation use global item relations to explore collaborative information from different sessions. These methods capture the topological structure of the graph and rely on multi-hop information aggregation in GNNs to exchange information along edges. Consequently, graph-based models suffer from noisy item relations in the training data and introduce high complexity for large item catalogs. We propose to explicitly model the multi-hop information aggregation mechanism over multiple layers via shortest-path edges based on knowledge from the sequential recommendation domain. Our approach does not require multiple layers to exchange information and ignores unreliable item-item relations. Furthermore, to address inherent data sparsity, we are the first to apply supervised contrastive learning by mining data-driven positive and hard negative item samples from the training data. Extensive experiments on three different datasets show that the proposed approach outperforms almost all of the state-of-the-art methods.

Evaluation is important when developing and deploying recommender systems. The PERSPECTIVES workshop sheds light on the different, potentially diverging or contradictory perspectives on the evaluation of recommender systems. Building on the discussions and outcomes of the PERSPECTIVES workshops held at RecSys 2021 and 2022, the third edition of the PERSPECTIVES workshop held at RecSys 2023 brought together researchers and practitioners from academia and industry to reflect on the evaluation of recommender systems critically. The workshop featured a keynote and focused on the interactive part with discussions in small groups and the plenum. We discussed problems and lessons learned, encouraged the exchange of the many perspectives on evaluation, and aimed to move the discourse forward within the community.

When we appreciate a piece of music, it is most naturally because of its content, including rhythmic, tonal, and timbral elements as well as its lyrics and semantics. This suggests that the human affinity for music is inherently content-driven. This kind of information is, however, still frequently neglected by mainstream recommendation models based on collaborative filtering that rely solely on user-item interactions to recommend items to users. A major reason for this neglect is the lack of standardized datasets that provide both collaborative and content information. The work at hand addresses this shortcoming by introducing Music4All-Onion, a large-scale, multi-modal music dataset. The dataset expands the Music4All dataset by including 26 additional audio, video, and metadata characteristics for 109,269 music pieces. In addition, it provides a set of 252,984,396 listening records of 119,140 users, extracted from the online music platform Last.fm, which allows leveraging user-item interactions as well. We organize distinct item content features in an onion model according to their semantics, and perform a comprehensive examination of the impact of different layers of this model (e.g., audio features, user-generated content, and derivative content) on content-driven music recommendation, demonstrating how various content features influence accuracy, novelty, and fairness of music recommendation systems. In summary, with Music4All-Onion, we seek to bridge the gap between collaborative filtering music recommender systems and content-centric music recommendation requirements.

Evaluation of recommender systems is a central activity when developing recommender systems, both in industry and academia. The second edition of the PERSPECTIVES workshop held at RecSys 2022 brought together academia and industry to critically reflect on the evaluation of recommender systems. In the 2022 edition of PERSPECTIVES, we discussed problems and lessons learned, encouraged the exchange of the various perspectives on evaluation, and aimed to move the discourse forward within the community. We deliberately solicited papers reporting a reflection on problems regarding recommender systems evaluation and lessons learned. The workshop featured interactive parts with discussions in small groups as well as in the plenum, both on-site and online, and an industry keynote.

We present the Height Optimized Trie (HOT), a fast and space-efficient in-memory index structure. The core algorithmic idea of HOT is to dynamically vary the number of bits considered at each node, which enables a consistently high fanout and thereby good cache efficiency. For a fixed maximum node fanout, the overall tree height is minimal and its structure is deterministically defined. Multiple carefully engineered node implementations using SIMD instructions or lightweight compression schemes provide compactness and fast search and optimize HOT structures for different usage scenarios. Our experiments, which use a wide variety of workloads and data sets, show that HOT outperforms other state-of-the-art index structures for string keys both in terms of search performance and memory footprint, while being competitive for integer keys.

Evaluation is a cornerstone in the process of developing and deploying recommender systems. The PERSPECTIVES workshop brought together academia and industry to critically reflect on the evaluation of recommender systems. Particularly, the workshop aimed to shed light on the different, and maybe even diverging or contradictory perspectives on the evaluation of recommender systems. Papers reporting a reflection on problems regarding recommender systems evaluation and lessons learned were solicited. The workshop combined flash presentations of accepted papers, a keynote from industry, and an interactive part with discussions in break-out rooms as well as in the plenum. The workshop complemented the program of the main conference as it emphasized problems and lessons learned, fostered exchange integrating various perspectives on evaluation, and sought to move the recommender systems community forward as an outcome of the workshop.

Recommender systems, like other tools that make use of machine learning, are known to create or increase certain biases. Earlier work has already unveiled different performance of recommender systems for different user groups, depending on gender, age, country, and consumption behavior. In this work, we study user bias in terms of another aspect, i.e., users’ personality. We investigate to which extent state-of-the-art recommendation algorithms yield different accuracy scores depending on the users’ personality traits. We focus on the music domain and create a dataset of Twitter users’ music consumption behavior and personality traits, measuring the latter in terms of the OCEAN model. Investigating [email protected] and [email protected] of the recommendation algorithms SLIM, embarrassingly shallow autoencoders for sparse data (EASE), and variational autoencoders for collaborative filtering (Mult-VAE) on this dataset, we find several significant differences in performance between user groups scoring high vs. groups scoring low on several personality traits.

We present the Height Optimized Trie (HOT), a fast and space-efficient in-memory index structure. The core algorithmic idea of HOT is to dynamically vary the number of bits considered at each node, which enables a consistently high fanout and thereby good cache efficiency. The layout of each node is carefully engineered for compactness and fast search using SIMD instructions. Our experimental results, which use a wide variety of workloads and data sets, show that HOT outperforms other state-of-the-art index structures for string keys both in terms of search performance and memory footprint, while being competitive for integer keys. We believe that these properties make HOT highly useful as a general-purpose index structure for main-memory databases.

Over the last years, music consumption has changed fundamentally: people switch from private, mostly limited music collections to huge public music collections provided by music streaming platforms. Thus, the amount of available music has increased dramatically and music streaming platforms heavily rely on recommender systems to assist users in discovering music they like. Incorporating the context of users has been shown to improve the quality of recommendations. Previous approaches based on pre-filtering suffered from a split dataset. In this work, we present a context-aware recommender system based on factorization machines that extracts information about the user's context from the names of the user's playlists. Based on a dataset comprising 15,000 users and 1.8 million tracks we show that our proposed approach outperforms the pre-filtering approach substantially in terms of accuracy of the computed recommendations.

Semistructured data provides the users of a community-based information system with the flexibility to store information without having to adhere to any predefined, rigid schema. However, such flexibility needs to be used with caution as it can lead to a very heterogeneous data structure and is therefore not feasible in terms of unified data access and search functionality. We present an approach which avoids such proliferation of substructures and provides the inserting user with recommendations, which are responsible for the creation of a commonly used structure. The presented recommendation algorithm adapts the recommendations to the stored information and its structure created by the community.