Give Users the Wheel

What if you could simply tell a recommendation system what you want instead of relying on likes, dislikes, and watch history? Kyle Polich talks with Fuyuan Lyu about the DPR framework, which combines large language models and traditional recommender systems to give users direct control over recommendations through natural language. Together they explore how conversational interfaces could transform platforms like YouTube, TikTok, and news feeds while preserving the strengths of modern recommendation algorithms.

Guest

Fuyuan Lyu: I’m currently a Ph.D. candidate at McGill University and Mila. I am fortunate to be advised by Prof. Jin L.C. Guo and Prof. Xue Liu. Before joining McGill, I earned a bachelor's degree in Computer Science and the Zhiyuan Honour Degree in Engineering from Shanghai Jiao Tong University (SJTU). I was advised by Prof. Li Jiang for my bachelor's thesis and worked with Prof. Xiaokang Yang. I also worked with Prof. Weichen Liu as a visiting research student at Nanyang Technological University (NTU), who inspired me to pursue high-quality research. I am passionate about building data-centric AI. Specifically, I study (i) how to model and select features and (ii) how to provide high-quality labels for both deep learning and foundation models fully automatically. I am eager to free humankind from repetitive, tedious work and enable us to focus on more creative tasks. I am actively working on applications in a variety of domains, including advertising, finance, and software engineering.