Trustworthy and Explainable Recommender Systems

RecSys IL has a new LinkedIn community! Click here to join >>

Join us at the upcoming RecSys IL meetup on May 26, where we'll dive into how explainability, trust, and privacy are reshaping the next generation of recommender systems.

**Talks are in English.

Date: May 26, 17:30
Location: Lusha office - 132 Menachem Begin, Azrieli Center, Triangle Building, 44th floor, Tel Aviv

Agenda:

17:30 - 18:00 Registration, Mingling, Snacks & Drinks
18:00 - 18:30 Explaining Recommendations at Scale (Guy Lissak)
18:30 - 19:00 Towards Faithful and Interpretable Recommender Systems: Recent Advances in Explainable AI (Prof. Noam Koenigstein)
19:00 - 19:30 Privacy-Preserving Context-Aware Recommendations: A Federated Learning Approach (Dr. Moshe Unger)

Details:

Talk 1

Title: Explaining Recommendations at Scale

Speaker: Guy Lissak, Insights Team Lead at Lusha

Bio:
Guy is a Data Engineering Team Lead at Lusha, where he builds large-scale data pipelines and real-time recommender systems. His work focuses on turning data into useful, explainable insights that help users make better decisions.
He holds a B.Sc. in Electronics and Computer Engineering from the Holon Institute of Technology, and pursuing a Data Science MSc these days.

Abstract:
What if your recommendation system didn't just suggest, but also clearly explained why?
Join us as we reveal how Lusha is enhancing recommendation engines by building high-performance, explainable, content-based systems that empower smarter, faster sales decisions at scale.
Creating a recommendation system is one challenge, building one that operates in real-time, handles enormous data volumes, and transparently explains its reasoning is an entirely different achievement.
In this session, we'll take you under the hood of Lusha's architecture to explore how it achieves ultra-low latency and scalability in its recommendation engines. Discover how cutting-edge techniques like embedding generation, vector compression, and real-time indexing in advanced vector databases come together to deliver recommendations users can trust, all at lightning speed.

We'll also dive into how we evaluate recommendation performance at scale, ensuring accuracy and relevance. Moreover, learn how we've embedded transparency deep into our system—from CRM insights and behavioral analytics to company-level indicators—surfaced through an intuitive, action-oriented user interface.

Talk 2

Title: Towards Faithful and Interpretable Recommender Systems: Recent Advances in Explainable AI

Speaker: Prof. Noam Koenigstein, School of Industrial Engineering and Inteligent Systems at Tel Aviv University

Bio:
Prof. Noam Koenigstein received his B.Sc. degree in computer science (cum laude) from the Technion – Israel Institute of Technology, Haifa, Israel, in 2007, and his M.Sc. degree in electrical engineering from Tel Aviv University, Tel Aviv, Israel, in 2009. He obtained his Ph.D. from the School of Electrical Engineering at Tel Aviv University in 2013. In 2011, Prof. Koenigstein joined Microsoft's Xbox Machine Learning research team, developing recommendation algorithms for millions of users globally, and later managed the recommendations research team for the Microsoft Store. In 2017, he became Senior Vice President and Head of Data Science at Citi Bank's Israeli Innovation Lab, leading data science activities at their Israeli research center. Prof. Koenigstein joined Tel Aviv University in 2018, where he currently serves as an Associate Professor at the School of Industrial Engineering and Intelligent Systems. He heads the DELTA Lab for applied deep-learning technologies, guiding students in the development and application of machine learning algorithms to a variety of real-world problems.

Abstract:
Explainable AI (XAI) plays a crucial role in recommender systems by helping users understand personalized recommendations and fostering trust in these increasingly complex models. Effective explanations must accurately reflect the true reasoning behind recommendations, going beyond simple plausibility.

In this talk, Prof. Noam Koenigstein from Tel Aviv University will share recent advances from his research group, DeltaLab@TAU, focusing on methods to ensure recommendations are both interpretable and faithful to their underlying logic. He will introduce innovative approaches for generating accurate explanations, evaluating their fidelity rigorously, and uncovering interpretable concepts from complex recommender models.
Join us to learn about cutting-edge developments in explainable recommender systems, and discover how transparency and trust can be significantly enhanced in personalized recommendation experiences.

Talk 3

Title: Privacy-Preserving Context-Aware Recommendations: A Federated Learning Approach

Speaker: Dr. Moshe Unger, Coller School of Management at Tel Aviv University

Bio:
Dr. Moshe Unger serves as an Assistant Professor at the Coller School of Management, Tel Aviv University. He completed his B.Sc., M.Sc., and Ph.D. in Software and Information Systems Engineering at Ben-Gurion University of the Negev, Israel. After earning his doctorate, he was a postdoctoral research scientist at NYU Stern School of Business. His research focuses on user modeling and personalization, data mining, machine learning, and recommender systems. He is currently an Amazon visiting academic at AWS, working on time-series personalization and forecasting. Previously, he collaborated with companies like Dell EMC, Deutsche Telekom, and Spotify.

Abstract:
Context-aware recommender systems (CARS) leverage dynamic contextual information, such as location and user behavior, to enhance recommendation relevance. However, incorporating such data introduces significant privacy risks, as sensitive user information can be exposed to misuse. Federated recommendation systems (FR) offer a promising solution by decentralizing data storage, yet they often overlook the impact of contextual information on performance. In this talk, I will introduce PRISM (Privacy and Recommender Systems Integration and Management), a novel approach that utilizes federated learning and differential privacy to deliver secure, context-aware recommendations. By integrating rich contextual data while preserving user privacy, PRISM achieves high accuracy and beyond-accuracy metrics, including diversity and novelty. I will discuss the framework’s design and implementation, share results from evaluations on multiple datasets, and examine the trade-offs between privacy and recommendation quality. This talk demonstrates how PRISM establishes a new standard for privacy-preserving CARS, offering a balanced approach to secure, personalized recommendations.