Successfully defended my undergraduate thesis 🎓
I successfully defended my undergraduate thesis, “Leveraging Knowledge Graphs and Graph-Based Deep Learning Models for Recommendation Systems,” at Ho Chi Minh City University of Technology (HCMUT), under the supervision of Assoc. Prof. Dr. Thoai Nam. The graduation ceremony will be held in November 2026.
đź“‚ Thesis materials (report, slides): Google Drive
Abstract
Knowledge graph-based recommenders add semantic relations beyond user–item interactions, but most rely on fixed embeddings that fail to generalize to unseen items without retraining. Subgraph-based reasoning helps by recommending from local evidence around a user, yet full multi-hop expansion is costly and pulls in irrelevant nodes that hurt quality.
We propose Knowledge-aware Intent-guided Subgraph Sampling for Recommendation, which builds compact user-centric subgraphs by aligning node sampling with the user’s latent interests. Semantic attributes of the user’s interacted items are encoded into multiple personalized intents that guide adaptive sampling during multi-hop propagation and drive intent-aware item prediction, preserving preference-aligned evidence while cutting redundant expansion.
On the Last-FM and Amazon-Book benchmarks, under both traditional and new-item settings, the method matches strong subgraph-based baselines and outperforms collaborative filtering, embedding, GNN, and intent-agnostic sampling methods—while substantially reducing propagated messages, inference time, and GPU memory versus full subgraph reasoning. This shows intent-guided sampling balances accuracy and efficiency for inductive knowledge graph-based recommendation.