Department of Applied IT, University of Gothenburg
As machine learning models continue to permeate critical domains such as medicine, the criminal justice system, and financial markets, a significant concern arises due to these models' inherent lack of interpretability. This limitation hinders human understanding and challenges ensuring accountability, fairness, and trust in decision-making. In this lecture, we explore the field of explainable recommender systems as an essential branch of AI research. By providing a broad overview of the concepts, methods, and evaluation approaches, we aim to empower audiences with different backgrounds to tackle the challenges associated with explainability in recommender systems.
Throughout the lecture, we introduce concepts and explore different approaches and techniques for enhancing the explainability of recommender systems. Additionally, we examine the emerging field of LLM-based (Language Model-based) recommenders, which leverage the power of advanced language models to provide more explainable and contextual recommendations/explanations.
Finally, we shed light on various evaluation approaches specifically tailored to measure the impact of explanations across different goals. By understanding the metrics and benchmarks used for evaluation, researchers and practitioners can accurately assess the performance and interpretability of their recommender systems.
Bio: Leandro Balby Marinho
Leandro Balby Marinho (http://leandro.lsd.ufcg.edu.br/) is an associate professor at the Department of Computer Science of the Federal University of Campina Grande (UFCG), Brazil. He has a distinguished research background in Recommender Systems and Machine Learning in diverse domains and a proven track record of published papers in several major academic venues. In particular, he received the best paper award at ACM RecSys in 2015 and served as one of the general co-chairs of RecSys in 2020. His current interests in Recommender Systems include equipping complex recommender models with meaningful and user-friendly explanations and pre-training strategies for improving model performance and training speed.