Transformers4Rec is a flexible and efficient library for sequential and session-based recommendation and works with PyTorch.

[AAAI 2019] Source code and datasets for "Session-based Recommendation with Graph Neural Networks"

Recommender Systems Paperlist that I am interested in

Session-based Recommendation

[SIGIR 2020] Python implementation for "TAGNN: Target Attentive Graph Neural Networks for Session-based Recommendation"

PyTorch Implementation of Introducing Self-Attention to Target Attentive Graph Neural Networks (AISP '22)

[ECIR 2024] Official repository for the paper titled "Self Contrastive Learning for Session-based Recommendation"

A session-based recommendation system to recommend baby products on Amazon using 4 models namely ITEMKNN, POP, GRU4Rec, and STAMP, STAMP performs the best in all accuracy metrics followed by GRU4Rec. We also did result analysis, including ranking accuracy, coverage, popularity.

Comparative study of Graph Attention Networks for graph-based session recommendation.

Code for 2022 Applied Science Special Issue "Logit Averaging: Capturing Global Relation for Session-based Recommendation"

Data and feedback-loop half of a session-based music recommender on Navidrome — live scrobbler events, versioned datasets for GRU4Rec/SessionKNN, and drift-triggered retraining.

Two-stage session-based recommender (LightGBM LambdaRank) for an Inditex hackathon. Cold-start focused, 93% of sessions have no user history. NDCG@5 = 0.377, Hit Rate@5 = 76%.

This repository contains my summaries of various academic papers I have read.

Real-time traveler representation learning for hotel recommendations, benchmarking recency-weighted baselines, GRUs, and Transformers with temporal evaluation and streaming inference.

Amazon KDD Cup '23: Multilingual Recommendation Challenge

Research-oriented e-commerce recommender system for session intent, candidate generation, funnel-aware ranking, cold start, catalog exposure, and serving tradeoffs.

Controlled experiments on when chronological interaction order improves next-item recommendation beyond recent-history content and simple transition models.

OTTO session-based recommendation — co-visitation baseline + SASRec multi-task transformer