SPINRec: Stochastic Path Integration for Neural Recommender Explanations
SPINRec: Stochastic Path Integration for Neural Recommender Explanations
Yahlly Schein, M.Sc. student at the department of Industrial Advisor: Dr. Noam Koenigstein
Abstract:
The growing reliance on recommender systems in everyday decision-making has heightened the need for explainability to enhance transparency, trust, and user engagement. While much prior research has focused on user-centric aspects, such as interpretability and satisfaction, fidelity - ensuring explanations accurately reflect a recommender's decision-making process - remains underexplored. In our work, we introduce SpinRec, the first path-integration-based algorithm designed specifically for explainability in recommender systems. To address the challenges posed by data sparsity and the implicit, binary, and diverse signals inherent to this domain, SpinRec employs a novel dynamic baseline sampling technique over the distribution of users' personal data and achieves state-of-the-art fidelity, significantly outperforming existing explanation methods.
Bio:
Yahlly Schein, M.Sc. student at the department of Industrial Engineering department, he conducts his research at the DELTA Lab under the supervision of Dr. Noam Koenigstein. His work focuses on explainable recommender systems, aiming to develop novel algorithms that provide transparent and effective recommendations while designing techniques to evaluate explainers, ensuring their quality and relevance. By leveraging cutting-edge deep learning methods, his research seeks to achieve state-of-the-art results and enhance user trust and understanding in recommendation systems.
