Duricic Tomislav, Kowald Dominik, Emanuel Lacic, Lex Elisabeth
2023
By providing personalized suggestions to users, recommender systems have become essential to numerous online platforms. Collaborative filtering, particularly graph-based approaches using Graph Neural Networks (GNNs), have demonstrated great results in terms of recommendation accuracy. However, accuracy may not always be the most important criterion for evaluating recommender systems' performance, since beyond accuracy aspects such as recommendation diversity, serendipity, and fairness can strongly influence user engagement and satisfaction. This review paper focuses on addressing these dimensions in GNN-based recommender systems, going beyond the conventional accuracy-centric perspective. We begin by reviewing recent developments in approaches that improve not only the accuracy-diversity trade-off, but also promote serendipity and fairness in GNN-based recommender systems. We discuss different stages of model development including data preprocessing, graph construction, embedding initialization, propagation layers, embedding fusion, score computation, and training methodologies. Furthermore, we present a look into the practical difficulties encountered in assuring diversity, serendipity, and fairness, while retaining high accuracy. Finally, we discuss potential future research directions for developing more robust GNN-based recommender systems that go beyond the unidimensional perspective of focusing solely on accuracy. This review aims to provide researchers and practitioners with an in-depth understanding of the multifaceted issues that arise when designing GNN-based recommender systems.
Oana Inel, Duricic Tomislav, Harmanpreet Kaur, Lex Elisabeth, Nava Tintarev
2021
Online videos have become a prevalent means for people to acquire information. Videos, however, are often polarized, misleading, or contain topics on which people have different, contradictory views. In this work, we introduce natural language explanations to stimulate more deliberate reasoning about videos and raise users’ awareness of potentially deceiving or biased information. With these explanations, we aim to support users in actively deciding and reflecting on the usefulness of the videos. We generate the explanations through an end-to-end pipeline that extracts reflection triggers so users receive additional information to the video based on its source, covered topics, communicated emotions, and sentiment. In a between-subjects user study, we examine the effect of showing the explanations for videos on three controversial topics. Besides, we assess the users’ alignment with the video’s message and how strong their belief is about the topic. Our results indicate that respondents’ alignment with the video’s message is critical to evaluate the video’s usefulness. Overall, the explanations were found to be useful and of high quality. While the explanations do not influence the perceived usefulness of the videos compared to only seeing the video, people with an extreme negative alignment with a video’s message perceived it as less useful (with or without explanations) and felt more confident in their assessment. We relate our findings to cognitive dissonance since users seem to be less receptive to explanations when the video’s message strongly challenges their beliefs. Given these findings, we provide a set of design implications for explanations grounded in theories on reducing cognitive dissonance in light of raising awareness about online deception.
Lovric Mario, Duricic Tomislav, Tran Thi Ngoc Han, Hussain Hussain, Lacic Emanuel, Morten A. Rasmussen, Kern Roman
2021
Methods for dimensionality reduction are showing significant contributions to knowledge generation in high-dimensional modeling scenarios throughout many disciplines. By achieving a lower dimensional representation (also called embedding), fewer computing resources are needed in downstream machine learning tasks, thus leading to a faster training time, lower complexity, and statistical flexibility. In this work, we investigate the utility of three prominent unsupervised embedding techniques (principal component analysis—PCA, uniform manifold approximation and projection—UMAP, and variational autoencoders—VAEs) for solving classification tasks in the domain of toxicology. To this end, we compare these embedding techniques against a set of molecular fingerprint-based models that do not utilize additional pre-preprocessing of features. Inspired by the success of transfer learning in several fields, we further study the performance of embedders when trained on an external dataset of chemical compounds. To gain a better understanding of their characteristics, we evaluate the embedders with different embedding dimensionalities, and with different sizes of the external dataset. Our findings show that the recently popularized UMAP approach can be utilized alongside known techniques such as PCA and VAE as a pre-compression technique in the toxicology domain. Nevertheless, the generative model of VAE shows an advantage in pre-compressing the data with respect to classification accuracy.
Lacic Emanuel, Traub Matthias, Duricic Tomislav, Haslauer Eva, Lex Elisabeth
2018
A challenge for importers in the automobile industry is adjusting to rapidly changing market demands. In this work, we describe a practical study of car import planning based on the monthly car registrations in Austria. We model the task as a data driven forecasting problem and we implement four different prediction approaches. One utilizes a seasonal ARIMA model, while the other is based on LSTM-RNN and both compared to a linear and seasonal baselines. In our experiments, we evaluate the 33 different brands by predicting the number of registrations for the next month and for the year to come.