About Me
Welcome! I am Felice Antonio Merra. I am working as a Senior Data Science Manager at Cognism.com.
Before joining Cognism, I have been working for more than 3 years as an Applied Scientist at Amazon.com working on Generative AI (Rufus and AmazonQ), Information Retrieval (Amazon Search), Recommender Systems and Natural Language Processing.
I graduated as Ph.D. cum laude defending a Ph.D. Thesis on Adversarial Machine Learning in Recommender Systems at Department of Electrical Engineering and Information Technology, Polytechnic University of Bari @PolibaOfficial supervised by Prof. Tommaso Di Noia.
I did a Summer Internship as Applied Scientist at Amazon.com in the Amazon Search and Personalization Team, and a research Visiting at Knowledge Media Institute under the supervision of Prof. Enrico Motta.
Email: merrafelice@gmail.com
Important News
- [Highlight] I’ll be servicing RecSys 2026 as Industrial co-chair
- [Highlight] Best Short Paper: Do LLMs Memorize Recommendation Datasets? A Preliminary Study on MovieLens-1M at SIGIR 2025
- [Highlight] ICML’25 accepted paper on Bayesian optimized prompt engineering. Hyperband-based Bayesian Optimization for Black-box Prompt Selection
- [Highlight] Best Short Paper - Runner Up: A Formal Analysis of Recommendation Quality of Adversarially-trained Recommenders at CIKM 2021
- [Highlight] MIT-IBM Watson AI Lab best paper award: Understanding the Effects of Adversarial Personalized Ranking Optimization Method on Recommendation Quality at AdvML@KDD 2021
Research Interest
My research activities mainly focus on artificial intelligence (AI). My investigation is devoted to novel approaches and applying machine learning (ML) algorithms, particularly to Trustworthy AI and Generative AI. In particular, I devote my attention to recommender system (RS) applications to study the robustness of modern ML recommender models affected by adversarial threats.
After having assessed the state-of-the-art of AML techniques in RS, I am invetsigating three main areas of study:
- the robustness of recommender models when affected by hand-engineered shilling attacks,
- the formal study of the effects of AML training strategies on the beyond-accuracy effects of recommenders, i.e., bias disparity, fairness, novelty
- the proposal of adversarial attacks against multimedia retrieval models.
- the Trustworthiness of ML (security, explainability, privacy, fairness, bias)
In the future, I plan to extend the previous line of study and continue to investigate AML approaches on other ML tasks, e.g., computer vision and reinforcement learning, with the aim to bridge the final users’ at the core of my research to verify how much they can Trust an ML system.
For more information you can look at my Curriculum Vitae.
Publications
Check my Google Scholar profile: https://scholar.google.com/citations?user=O4bDuG4AAAAJ&hl=en