Stavroula Eleftherakis

Modern recommender systems are data-intensive pipelines, so fairness depends not only on the recommendation algorithm but also on the training data it learns from. We study this data-level vulnerability through fairness-targeted shilling attacks, where adversaries inject fake user profiles to manipulate fairness in collaborative filtering recommenders. We model the attack as an attempt to alter consumer-side group fairness between privileged and unprivileged users, using disparate impact over relevance- and exposure-based benefits. We instantiate this setting with interpretable profile-injection strategies adapted from classic shilling heuristics and evaluate them across five datasets, multiple attack budgets, 13 recommendation algorithms, and both conventional and fairness-aware settings. Our results show that profile injection is a concrete threat to fairness in recommendation, that effective attacks must exploit group-specific behavioral patterns, and that fairness-aware models may become more vulnerable under such interventions.

For over five decades, enabling non-technical users to access databases without SQL has been a major research focus. Despite recent advances in Text-to-SQL, the primary focus remains on accuracy, often overlooking real-world deployment challenges. This demo presents DataDazzle, a natural language data exploration system. It allows users to query data, review and verify answers, and benefit from recommendations, offering a comprehensive data exploration experience.

In this demonstration, we present DatAgent, an intelligent data assistant system that allows users to ask queries in natural language, and can respond in natural language as well. Moreover, the system actively guides the user using different types of recommendations and hints, and learns from user actions. We will demonstrate different exploration scenarios that show how the system and the user engage in a human-like interaction inspired by the interaction paradigm of chatbots and virtual assistants.

Database interaction is often characterized as a non-trivial and time consuming process due to user’s inexperience with the data or the query language. Therefore, there is a need for the databases to be able to "talk back" in order to assist the users during data exploration and eventually lead them to the desired results. In this paper, we tackle the problem of SQL-to-NL by extending the graph-based model of Logos. Our novel extensions include improvements in terms of the system’s translation capabilities and the fluency of the generated explanations. Finally, we report several challenges, highlighted by experiments on different user cases, i.e, astronomy and policy making.

We address demographic bias in neighborhood-learning models for collaborative filtering recommendations. Despite their superior ranking performance, these methods can learn neighborhoods that inadvertently foster discriminatory patterns. Little work exists in this area, highlighting an important research gap. A notable yet solitary effort, Balanced Neighborhood Sparse LInear Method (BNSLIM) aims at balancing neighborhood influence across different demographic groups. Yet, BNSLIM is hampered by computational inefficiency, and its rigid balancing approach often impacts accuracy. In that vein, we introduce two novel algorithms. The first, an enhancement of BNSLIM, incorporates the Alternating Direction Method of Multipliers (ADMM) to optimize all similarities concurrently, greatly reducing training time. The second, Fairly Sparse Linear Regression (FSLR), induces controlled sparsity in neighborhoods to reveal correlations among different demographic groups, achieving comparable efficiency while being more accurate. Their performance is evaluated using standard exposure metrics alongside a new metric for user coverage disparities. Our experiments cover various applications, including a novel exploration of bias in course recommendations by teachers’ country development status. Our results show the effectiveness of our algorithms in imposing fairness compared to BNSLIM and other well-known fairness approaches.