Disentangling Preferences, Entangling Lives

On 28 May 2025, Tran Nhu Thuat successfully defended his dissertation, titled "Disentangling User Preferences Towards Self-Interpretable Recommender Systems."

Thuat outlining the structure of his thesis at the beginning of the defense

The main thesis of his dissertation is that disentangling the multiple latent factors underlying user preferences, through both behavioral interactions and side information, enables more accurate and interpretable recommendations.

The first part focuses on disentangling user preferences from item interactions by modeling personalized, compositional, and shared interests across users.

The second part addresses behavioral data sparsity by aligning preference factors derived from ratings and textual information, using compositional de-attention and optimal transport to improve recommendation accuracy and interpretability.

Dinner at Ellenborough Market Cafe in December 2020 while Thuat was a first-year PhD candidate (Thuat was third from left)

Thuat first joined Preferred.AI in 2020 as a PhD candidate and our first Vingroup Young Talent Scholar. Over the ensuing five years, he has built an exemplary record of research achievements and accolades, adding the Dean's List, SMU Presidential Doctoral Fellowship, and SDSC Dissertation Fellowship to his list of distinctions. Through his dedication, intellectual curiosity, and pursuit of excellence, he has set a high standard for the younger scholars and students in our group to aspire to.

Post-defense celebration at Chimichanga (Thuat was fourth from right)

Since completing his PhD, Thuat has also celebrated two significant milestones in his personal and professional life: getting married and embarking on his career with Huawei. We warmly congratulate him on these wonderful new beginnings and wish him every success and happiness in the years ahead!