Compositionality in knowledge and language—the ability to represent complex concepts as a combination of simpler ones—is a hallmark of human cognition and communication. Despite recent advances, deep neural networks still struggle to acquire this property reliably. Neural models for emergent communication look to endow artificial agents with compositional language by simulating the pressures that form human language. In this work, we introduce CELEBI (Compressive-Expressive Language Emergence through a discrete Bottleneck and Iterated learning), a novel self-supervised framework for inducing compositional representations through a reconstruction-based communication game between a sender and a receiver. Building on theories of language emergence and the iterated learning framework, we integrate three mechanisms that jointly promote compressibility, expressivity, and efficiency in the emergent language. First, Progressive Decoding incentivizes intermediate reasoning by requiring the receiver to produce partial reconstructions after each symbol. Second, Final-State Imitation trains successive generations of agents to imitate reconstructions rather than messages, enforcing a tighter communication bottleneck. Third, Pairwise Distance Maximization regularizes message diversity by encouraging high distances between messages, with formal links to entropy maximization. Our method significantly improves both the efficiency and compositionality of the learned messages on the Shapes3D and MPI3D datasets, surpassing prior discrete communication frameworks in both reconstruction accuracy and topographic similarity. This work provides new theoretical and empirical evidence for the emergence of structured, generalizable communication protocols from simplicity-based inductive biases.

In recent years, Reinforcement Learning (RL) has shown great promise in session-based recommendation. Sequential models that use RL have reached state-of-the-art performance for the Next-item Prediction (NIP) task. This result is intriguing, as the NIP task only evaluates how well the system can correctly recommend the next item to the user, while the goal of RL is to find a policy that optimizes rewards in the long term – sometimes at the expense of suboptimal short-term performance. Then, how can RL improve the system’s performance on short-term metrics? This article investigates this question by exploring proxy learning objectives, which we identify as goals RL models might be following, and thus could explain the performance boost. We found that RL – when used as an auxiliary loss – promotes the learning of embeddings that capture information about the user’s previously interacted items. Subsequently, we replaced the RL objective with a straightforward auxiliary loss designed to predict the number of items the user interacted with. This substitution results in performance gains comparable to RL. These findings pave the way to improve performance and understanding of RL methods for recommender systems.

Generative adversarial training has recently raised significant interest in recommender systems. Adversarial pairwise learning, in particular, has led to methods to select and create unobserved training samples that generalize user preferences, increasing the accuracy and robustness of collaborative filtering (CF) models. Despite this success, only some authors have analyzed adversarial sampling’s ability to recommend long-tail and cold-start items, as well as their ability to promote novelty and diversity. These concerns are crucial for modern recommender systems.This paper investigates adversarial pairwise learning in data sparsity scenarios in which most items are consumed by only a few users (long-tail items), and there is a substantial proportion of items without interactions (cold-start items). We found that adversarial sampling increases the bias of CF models toward popular items, resulting in under-recommendation of relevant but less popular long-tail items, poor cold start performance, and low aggregate diversity, i.e., unfair coverage of items among recommendation lists. To address these problems, we propose a multi-modal extension of adversarial pairwise learning that incorporates text and visual information about the items in addition to user-item interaction data. As in the original model, our approach relies on a minimax game. A generative model proposes items for a user considering her visual and textual preferences. Then, a collaborative critic discriminates the suggested items from those already consumed by the user.We conduct experiments on three challenging datasets from the online retail domain in which more than 99.99% of the user-item interactions are unknown, and around 2/3 of the items have less than 5 interactions. We evaluate the advantages of our approach to adversarial and non-adversarial methods, achieving state of the art results in the most complex scenario: the recommendation of new items. Furthermore, we found that the proposed adversarial framework successfully leverages content to make more diverse and novel recommendations.Our code is publicly available on GitHub https://anonymous.4open.science/r/M-APL-D7B7/.

Journalism has become more data-driven and inherently visual in recent years. Photographs, illustrations, infographics, data visualizations, and general images help convey complex topics to a wide audience. The way that visual artifacts influence how readers form an opinion beyond the text is an important issue to research, but there are few works about this topic. In this context, we research the persuasive, emotional and memorable dimensions of data visualizations and illustrations in journalistic storytelling for long-form articles. We conducted a user study and compared the effects which data visualizations and illustrations have on changing attitude towards a presented topic. While visual representations are usually studied along one dimension, in this experimental study, we explore the effects on readers' attitudes along three: persuasion, emotion, and information retention. By comparing different versions of the same article, we observe how attitudes differ based on the visual stimuli present, and how they are perceived when combined. Results indicate that the narrative using only data visualization elicits a stronger emotional impact than illustration-only visual support, as well as a significant change in the initial attitude about the topic. Our findings contribute to a growing body of literature on how visual artifacts may be used to inform and influence public opinion and debate. We present ideas for future work to generalize the results beyond the domain studied, the water crisis.

Pre-training is essential for effective representation learning models, especially in natural language processing and computer vision-related tasks. The core idea is to learn representations, usually through unsupervised or self-supervised approaches on large and generic source datasets, and use those pre-trained representations (aka embeddings) as initial parameter values during training on the target dataset. Seminal works in this area show that pre-training can act as a regularization mechanism placing the model parameters in regions of the optimization landscape closer to better local minima than random parameter initialization. However, no systematic studies evaluate the effectiveness of pre-training strategies on model-based collaborative filtering. This paper conducts a broad set of experiments to evaluate different pre-training strategies for collaborative filtering using Matrix Factorization (MF) as the base model. We show that such models equipped with pre-training in a transfer learning setting can vastly improve the prediction quality compared to the standard random parameter initialization baseline, reaching state-of-the-art results in standard recommender systems benchmarks. We also present alternatives for the out-of-vocabulary item problem (i.e., items present in target but not in source datasets) and show that pre-training in the context of MF acts as a regularizer, explaining the improvement in model generalization.

Abstract We present a methodology for making sense of the communicative role of data visualizations in journalistic storytelling and share findings from surveying water‐related data stories. Data stories are a genre of long‐form journalism that integrate text, data visualization, and other visual expressions (e.g., photographs, illustrations, videos) for the purpose of data‐driven storytelling. In the last decade, a considerable number of data stories about a wide range of topics have been published worldwide. Authors use a variety of techniques to make complex phenomena comprehensible and use visualizations as communicative devices that shape the understanding of a given topic. Despite the popularity of data stories, we, as scholars, still lack a methodological framework for assessing the communicative role of visualizations in data stories. To this extent, we draw from data journalism, visual culture, and multimodality studies to propose an interpretative framework in six stages. The process begins with the analysis of content blocks and framing elements and ends with the identification of dimensions, patterns, and relationships between textual and visual elements. The framework is put to the test by analyzing 17 data stories about water‐related issues. Our observations from the survey illustrate how data visualizations can shape the framing of complex topics.

Embeddings are core components of modern model-based Collaborative Filtering (CF) methods, such as Matrix Factorization (MF) and Deep Learning variations. In essence, embeddings are mappings of the original sparse representation of categorical features (e.g., user and items) to dense low-dimensional representations. A well-known limitation of such methods is that the learned embeddings are opaque and hard to explain to the users. On the other hand, a key feature of simpler KNN-based CF models (aka user/item-based CF) is that they naturally yield similarity-based explanations, i.e., similar users/items as evidence to support model recommendations. Unlike related works that try to attribute explicit meaning (via metadata) to the learned embeddings, in this paper, we propose to equip the learned embeddings of MF with meaningful similarity-based explanations. First, we show that the learned user/item embeddings of MF do not preserve the distances between users (or items) in the original rating matrix. Next, we propose a novel approach that initializes Stochastic Gradient Descent (SGD) with user/item embeddings that preserve the structural properties of the original input data. We conduct a broad set of experiments and show that our method enables explanations, very similar to the ones provided by KNN-based approaches, without harming the prediction performance. Moreover, we show that fine-tuning the structure-preserving embeddings may unlock better local minima in the optimization space, leading simple vanilla MF to reach competitive performances with the best-known models for the rating prediction task.

In recent years, instructional design has become even more challenging for teaching staff members in higher education institutions. If instructional design causes student overload, it could lead to superficial learning and decreased student well-being. A strategy to avoid overload is reflecting upon the effectiveness of teaching practices in terms of time-on-task. This article presents a Work-In-Progress conducted to provide teachers with a dashboard to visualize student self-reports of time-on-task regarding subject activities. A questionnaire was applied to 15 instructors during a set trial period to evaluate the perceived usability and usefulness of the dashboard. Preliminary findings reveal that the dashboard helped instructors became aware about the number of hours spent outside of class time. Furthermore, data visualizations of time-on-task evidence enabled them to redesign subject activities. Currently, the dashboard has been adopted by 106 engineering instructors. Future work involves the development of a framework to incorporate user-based improvements.

The video game industry has adopted recommendation systems to boost users interest with a focus on game sales. Other exciting applications within video games are those that help the player make decisions that would maximize their playing experience, which is a desirable feature in real-time strategy video games such as Multiplayer Online Battle Arena (MOBA) like as DotA and LoL. Among these tasks, the recommendation of items is challenging, given both the contextual nature of the game and how it exposes the dependence on the formation of each team. Existing works on this topic do not take advantage of all the available contextual match data and dismiss potentially valuable information. To address this problem we develop TTIR, a contextual recommender model derived from the Transformer neural architecture that suggests a set of items to every team member, based on the contexts of teams and roles that describe the match. TTIR outperforms several approaches and provides interpretable recommendations through visualization of attention weights. Our evaluation indicates that both the Transformer architecture and the contextual information are essential to get the best results for this item recommendation task. Furthermore, a preliminary user survey indicates the usefulness of attention weights for explaining recommendations as well as ideas for future work. The code and dataset are available at https://github.com/ojedaf/IC-TIR-Lol .

Explaining suggestions made by recommendation systems is key to make users trust and accept these systems. This is specially critical in areas such as art image recommendation. Traditionally, artworks are sold in galleries where people can see them physically, and artists have the chance to persuade the people into buying them. On the other side, online art stores only offer the user the action of navigating through the catalog, but nobody plays the persuading role of the artist. Moreover, few works in recommendation systems provide a perspective of the many variables involved in the user perception of several aspects of the system such as domain knowledge, relevance, explainability, and trust. In this article, we aim to fill this gap by studying several aspects of the user experience with a recommender system of artistic images, from algorithmic and HCI perspectives. We conducted two user studies in Amazon Mechanical Turk to evaluate different levels of explainability, combined with different algorithms. While in study 1 we focus only on a desktop interface, in study 2 we attempt to understand the effect of explanations in mobile devices. In general, our experiments confirm that explanations of recommendations in the image domain are useful and increase user satisfaction, perception of explainability and relevance. In the first study, our results show that the observed effects are dependent on the underlying recommendation algorithm used. In the second study, our results show that these effects are also dependent of the device used in the study but with a smaller effect. Finally, using the framework by Knijnenburg et al., we provide a comprehensive model, for each study, which synthesizes the effects between different variables involved in the user experience with explainable visual recommender systems of artistic images.

There are very few works about explaining content-based recommendations of images in the artistic domain. Current works do not provide a perspective of the many variables involved in the user perception of several aspects of the system such as domain knowledge, relevance, explainability, and trust. In this paper, we aim to fill this gap by studying three interfaces, with different levels of explainability, for artistic image recommendation. Our experiments with N=121 users confirm that explanations of recommendations in the image domain are useful and increase user satisfaction, perception of explainability and relevance. Furthermore, our results show that the observed effects are also dependent on the underlying recommendation algorithm used. We tested two algorithms: Deep Neural Networks (DNN), which has high accuracy, and Attractiveness Visual Features (AVF) with high transparency but lower accuracy. Our results indicate that algorithms should not be studied in isolation, but rather in conjunction with interfaces, since both play a significant role in the perception of explainability and trust for image recommendation. Finally, using the framework by Knijnenburg et al., we provide a comprehensive model which synthesizes the effects between different variables involved in the user experience with explainable visual recommender systems of artistic images.

E-Sports has been positioned as an important activity within MOBA (Multiplayer Online Battle Arena) games in recent years. There is existing research on recommender systems in this topic, but most of it focuses on the character recommendation problem. However, the recommendation of items is also challenging because of its contextual nature, depending on the other characters. We have developed a framework that suggests items for a character based on the match context. The system aims to help players who have recently started the game as well as frequent players to take strategic advantage during a match and to improve their purchasing decision making. By analyzing a dataset of ranked matches through data mining techniques, we can capture purchase dynamic of experienced players to use it to generate recommendations. The results show that our proposed solution yields up to 80% of mAP, suggesting that the method leverages context information successfully. These results, together with open issues we mention in the paper, call for further research in the area.

The first workshop on Interactive Data Mining is held in Melbourne, Australia, on February 15, 2019 and is co-located with 12th ACM International Conference on Web Search and Data Mining (WSDM 2019). The goal of this workshop is to share and discuss research and projects that focus on interaction with and interactivity of data mining systems. The program includes invited speaker, presentation of research papers, and a discussion session.

We design, implement and evaluate EpistAid, an interactive relevance feedback system to support physicians towards a more efficient citation screening process for Evidence Based Health Care (EBHC). The system combines a relevance feedback algorithm with an interactive interface inspired by Tinder-like swipe interaction. To evaluate its efficiency and effectiveness in the citation screening process we conducted a user study with real users (senior medicine students) using a large EBHC dataset (Epistemonikos), with around 400,000 documents. We compared two relevance feedback algorithms, Rocchio and BM25-based. The combination of Rocchio relevance feedback with the document visualization yielded the best recall and F-1 scores, which are the most important metrics for EBHC document screening. In terms of cognitive demand and effort, BM25 relevance feedback without visualization was perceived as needing more physical and cognitive effort. EpistAid has the potential of improving the process for answering clinical questions by reducing the time needed to classify documents, as well as promoting user interaction. Our results can inform the development of intelligent user interfaces for screening research articles in the clinical domain and beyond.

With the increase of data collected and computation power available, modern recommender systems are ever facing new challenges. While complex models are developed in academia, industry practice seems to focus on relatively simple techniques that can deal with the magnitude of data and the need to distribute the computation. The workshop on large-scale recommender systems (LSRS) is a meeting place for industry and academia to discuss the current and future challenges of applied large-scale recommender systems.

With the increase of data collected and computation power available, modern recommender systems are ever facing new challenges. While complex models are developed in academia, industry practice seems to focus on relatively simple techniques that can deal with the magnitude of data and the need to distribute the computation. The workshop on large-scale recommender systems (LSRS) is a meeting place for industry and academia to discuss the current and future challenges of applied large-scale recommender systems.

A large body of research in recommender systems focuses on optimizing prediction and ranking. However, recent work has highlighted the importance of other aspects of the recommendations, including transparency, control and user experience in general. Building on these aspects, we introduce MoodPlay, a hybrid recommender system music which integrates content and mood-based filtering in an interactive interface. We show how MoodPlay allows the user to explore a music collection by latent affective dimensions, and we explain how to integrate user input at recommendation time with predictions based on a pre-existing user profile. Results of a user study (N=240) are discussed, with four conditions being evaluated with varying degrees of visualization, interaction and control. Results show that visualization and interaction in a latent space improve acceptance and understanding of both metadata and item recommendations. However, too much of either can result in cognitive overload and a negative impact on user experience.

Several approaches have been researched to help people deal with abundance of information. An important feature pioneered by social tagging systems and later used in other kinds of social systems is the ability to explore different community relevance prospects by examining items bookmarked by a specific user or items associated by various users with a specific tag . A ranked list of recommended items offered by a specific recommender engine can be considered as another relevance prospect. The problem that we address is that existing personalized social systems do not allow their users to explore and combine multiple relevance prospects. Only one prospect can be explored at any given time—a list of recommended items, a list of items bookmarked by a specific user, or a list of items marked with a specific tag. In this article, we explore the notion of combining multiple relevance prospects as a way to increase effectiveness and trust. We used a visual approach to recommend articles at a conference by explicitly presenting multiple dimensions of relevance. Suggestions offered by different recommendation techniques were embodied as recommender agents to put them on the same ground as users and tags. The results of two user studies performed at academic conferences allowed us to obtain interesting insights to enhance user interfaces of personalized social systems. More specifically, effectiveness and probability of item selection increase when users are able to explore and interrelate prospects of items relevance—that is, items bookmarked by users, recommendations and tags. Nevertheless, a less-technical audience may require guidance to understand the rationale of such intersections.

In this paper, we briefly describe the goal and activities of the LAK15 workshop on Visual Aspects of Learning analytics.

In this paper, we present work-in-progress of a recently started project that aims at studying the effect of time in recommender systems in the context of social tagging. Despite the existence of previous work in this area, no research has yet made an extensive evaluation and comparison of time-aware recommendation methods. With this motivation, this paper presents results of a study where we focused on understanding (i) "when" to use the temporal information into traditional collaborative filtering (CF) algorithms, and (ii) "how" to weight the similarity between users and items by exploring the effect of different time-decay functions. As the results of our extensive evaluation conducted over five social tagging systems (Delicious, BibSonomy, CiteULike, MovieLens, and Last.fm) suggest, the step (when) in which time is incorporated in the CF algorithm has substantial effect on accuracy, and the type of time-decay function (how) plays a role on accuracy and coverage mostly under pre-filtering on user-based CF, while item-based shows stronger stability over the experimental conditions.

Large scale virtual worlds such as massive multiplayer online games or 3D worlds gained tremendous popularity over the past few years. With the large and ever increasing amount of content available, virtual world users face the information overload problem. To tackle this issue, game-designers usually deploy recommendation services with the aim of making the virtual world a more joyful environment to be connected at. In this context, we present in this paper the results of a project that aims at understanding the mobility patterns of virtual world users in order to derive place recommenders for helping them to explore content more efficiently. Our study focus on the virtual world SecondLife, one of the largest and most prominent in recent years. Since SecondLife is comparable to real-world Location-based Social Networks (LBSNs), i.e., users can both check-in and share visited virtual places, a natural approach is to assume that place recommenders that are known to work well on real-world LBSNs will also work well on SecondLife. We have put this assumption to the test and found out that (i) while collaborative filtering algorithms have compatible performances in both environments, (ii) existing place recommenders based on geographic metadata are not useful in SecondLife.

No abstract available.

Research in recommender systems has traditionally focused on improving the predictive accuracy of recommendations by developing new algorithms or by incorporating new sources of data. However, several studies have shown that accuracy does not always correlate with a better user experience, leading to recent research that puts emphasis on Human-Computer Interaction in order to investigate aspects of the interface and user characteristics that influence the user experience on recommender systems. Following this new research this paper presents SetFusion, a visual user-controllable interface for hybrid recommender system. Our approach enables users to manually fuse and control the importance of recommender strategies and to inspect the fusion results using an interactive Venn diagram visualization. We analyze the results of two field studies in the context of a conference talk recommendation system, performed to investigate the effect of user controllability in a hybrid recommender. Behavioral analysis and subjective evaluation indicate that the proposed controllable interface had a positive effect on the user experience.

Research on recommender systems has traditionally focused on the development of algorithms to improve accuracy of recommendations. So far, little research has been done to enable user interaction with such systems as a basis to support exploration and control by end users. In this paper, we present our research on the use of information visualization techniques to interact with recommender systems. We investigated how information visualization can improve user understanding of the typically black-box rationale behind recommendations in order to increase their perceived relevance and meaning and to support exploration and user involvement in the recommendation process. Our study has been performed using TalkExplorer, an interactive visualization tool developed for attendees of academic conferences. The results of user studies performed at two conferences allowed us to obtain interesting insights to enhance user interfaces that integrate recommendation technology. More specifically, effectiveness and probability of item selection both increase when users are able to explore and interrelate multiple entities -- i.e. items bookmarked by users, recommendations and tags.

Social learning has confirmed its value in enhancing the learning outcomes across a wide spectrum. To support social learning, a visual approach is a common technique to represent and organize multiple students' data in an informative way. This paper presents a design of comparative social visualization for E-learning, which encourages information discovery and social comparisons. Classroom studies confirmed the motivational impact of personalized social guidance provided by the visualization in the target context. The visualization encouraged students to do some work ahead of the course schedule. Moreover, class leaders provided an implicit social guidance for the rest of the class and successfully led the way to discover the most relevant resources creating good trails for the rest of the class. We summarized the evidence of students' engagement and performance through the social visualization interface.

Recommendation Systems have been studied from several perspectives over the last twenty years --prediction accuracy, algorithmic scalability, knowledge sources, types of recommended items and tasks, evaluation methods, etc.-- but one area that has not been deeply investigated is the effect of different visualizations and their interaction with personal traits on users' evaluation of the recommended items. In this paper, I survey visual approaches that go beyond presenting the recommended items as a textual list or as annotations in context. I also review related literature from recommendations' explanations. In this thesis, I aim to understand how different visualizations and some personal traits might influence users' assessment of recommended items, particularly in domains where multidimensional data or contextual constraints are involved. I present the prototype of 2 recommendation visualizations and then briefly propose the research approach of this investigation.

Social navigation and social tagging technologies enable user communities to assemble the collective wisdom, and use it to help community members in finding the right information. However, it takes a significantly-sized community to make a social system truly useful. The question addressed in this p

Social tagging systems pose new challenges to developers of recommender systems. As observed by recent research, traditional implementations of classic recommender approaches, such as collaborative filtering, are not working well in this new context. To address these challenges, a number of research groups worldwide work on adapting these approaches to the specific nature of social tagging systems. In joining this stream of research, we have developed and evaluated two enhancements of user-based collaborative filtering algorithms to provide recommendations of articles on Cite ULike, a social tagging service for scientific articles. The result obtained after two phases of evaluation suggests that both enhancements are beneficial. Incorporating the number of raters into the algorithms, as we do in our NwCF approach, leads to an improvement of precision, while tag-based BM25 similarity measure, an alternative to Pearson correlation for calculating the similarity between users and their neighbors, increases the coverage of the recommendation process.

Collaborative tagging systems pose new challenges to the developers of recommender systems. As observed by recent research, traditional implementations of classic recommender approaches, such as collaborative filtering, are not working well in this new context. To address these challenges, a number of research groups worldwide work on adapting these approaches to the specific nature of collaborative tagging systems. In joining this stream of research, we have developed and compared three variants of user-based collaborative filtering algorithms to provide recommendations of articles on CiteULike. The first approach, Classic Collaborative filtering (CCF) uses Pearson correlation to calculate similarity between users and a classic adjusted ratings formula to rank the recommendations. The second approach, Neighbor-weighted Collaborative Filtering, takes into account the number of raters in the ranking formula of the recommendations. The third approach explores an innovative way to form the user neighborhood based on a modified version of the Okapi BM25 model over users' tags. Our results suggest that both alterations of CCF are beneficial. Incorporating the number of raters into the algorithms leads to an improvement of precision, while tag-based BM25 can be considered as an alternative to Pearson correlation to calculate the similarity between users and their neighbors.