Computer Science researcher working at the intersection of machine learning and the systems that put it to use: recommender systems, knowledge-graph reasoning, and LLM evaluation and serving.
I am a recent Computer Science graduate of Ho Chi Minh University of Technology – VNUHCM (HCMUT). Most recently I have studied the accuracy–diversity trade-off to tackle popularity bias in knowledge-graph conversational recommenders, which advanced into a peer-reviewed, co-first-author paper. I also build and evaluate large-scale AI and LLM systems in production as a software engineer.
I am applying to graduate research programs and am most excited by problems involving data-intensive systems, information retrieval, and integration with language models.
Academic Incentive Scholarship for 7 consecutive semesters (top 5% of cohort by GPA).
View transcriptRelevant coursework: Probability & Statistics (10), Big Data (9.9), Data Mining (9.0), Natural Language Processing (8.8), Linear Algebra (8.8), Machine Learning (8.2).
Reviewed sequential and session-based recommendation system, evaluated how a sequential recommender (Bert4Rec) performs when user history is a short session rather than a long history.
Studied transfer learning in conversational recommendation: compared learning item representations from scratch against reusing pretrained ones. Found that pretraining improves accuracy at the top ranks but weakens it deeper in the recommendation list.
Reviewed knowledge-graph conversational recommenders across the two main types: attribute-based and dialogue-based. Compared how each handles the accuracy–diversity trade-off and found that diversity is largely unaddressed in dialogue-based models — the gap my thesis targets.
Diversity measures — catalog coverage, novelty, intra-list diversity — cannot be optimised directly by standard training, so models learn to ignore them. I proposed two methods that bring diversity into training, tested on six knowledge-graph CRS models across two benchmarks:
A lazy-loading approach to training with large-scale interaction data
A controlled comparison across the ML spectrum
Awarded for seven consecutive semesters to students in the top 5% of the cohort by GPA.
Built data-driven shipping-load insights for Busan Ship Transport Co. in a Korean–Vietnamese team.
Contributing to big-data and high-performance-computing research projects.