A Survey on Session-Based Recommendation

PhD Qualifying Examination


Title: "A Survey on Session-Based Recommendation"

by

Mr. Hong SU


Abstract:

Session-based recommendation (SBR) aims to predict a user's next interaction
from a short, anonymous sequence of items. Unlike sequential recommendation
with persistent profiles, SBR must infer intent rapidly from sparse,
ephemeral signals. This survey traces the methodological landscape of SBR
from early heuristics and Markov models to contemporary neural approaches
based on recurrent networks, convolutions, graph neural networks, and
Transformers. It formalizes the classic problem and a generalized variant
with auxiliary information such as structured attributes and multimodal
content, and reviews widely used public datasets and evaluation practices.
Beyond architectures, it highlights the role of self-supervised and
contrastive objectives in combating sparsity, and examines system-level
challenges—explainability, the opportunities and pitfalls of LLM-assisted
reranking and explanation, and fairness for users, providers, and items.
The presentation synthesizes design trade-offs across model families and
outlines open directions toward robust, interpretable, and responsible SBR
systems.


Date:                   Monday, 6 October 2025

Time:                   4:00pm - 6:00pm

Venue:                  Room 5501
                        Lifts 25/26

Committee Members:      Prof. Raymond Wong (Supervisor)
                        Prof. Shing-Chi Cheung (Chairperson)
                        Dr. Junxian He