Session-Based Recommendation
Recommendations based on the current user session rather than historical profiles – ideal for anonymous visitors.
Session-based recommendation predicts the next relevant item based on the current click sequence – without historical user profile.
Explanation
Session-based RecSys uses RNNs, Transformers, or GNNs to predict the next relevant item from the click sequence of a session. No user profile needed.
Marketing Relevance
For e-commerce with a high share of anonymous visitors, session-based recommendation is the most important personalization approach.
Example
A visitor clicks on running shoes → sports shorts → fitness tracker. The system recommends sportswear based on the session sequence.
Common Pitfalls
Sessions can be very short (1-2 clicks). Multi-intent sessions (browsing + purchase intent) are hard to model.
Origin & History
GRU4Rec (Hidasi et al., 2016) was the first deep learning model for session-based recommendation. SR-GNN (2019) used graph neural networks. SASRec (Kang & McAuley, 2018) introduced self-attention.
Comparisons & Differences
Session-Based Recommendation vs. Collaborative Filtering
CF needs user history; session-based works with anonymous visitors and the current session alone.