RecBole-RBSR

Review-based Sequential Recommendation (RBSR) with RecBole

Recbole-RBSR implements review-based sequential recommendation on top of RecBole library.

• Shares the unified API(e.g., models) and input files with RecBole library.
• Reflects RecBole v1.2.1. (Latest version)

List of Modified and Added Files/Folders

RecBole-RBSR/
├── recbole/
│   ├── data/
│   │   └── dataset/
│   │       └── customized_dataset.py
│   ├── model/
│   │   ├── layers.py
│   │   └── sequential_recommender_review_based/
│   ├── utils/
│   │   └── utils.py
│   └── properties/
│       └── model/
│           ├── CCA.yaml
│           ├── IntentRec.yaml
│           ├── PIRSP.yaml
│           └── RNS.yaml
├── run_rbsr.py

Implemented Models

Model	Paper
RNS	A Review-Driven Neural Model for Sequential Recommendation (2019, IJCAI) [code]
PIRSP	Integrates review-based user-item interactions into sequential modeling (2021, EAAI)
CCA	Cascaded Cross Attention for Review-based Sequential Recommendation (2023, ICDM) [code]
IntentRec	IntentRec: Incorporating latent user intent via contrastive alignment for sequential recommendation (2025, ECRA)

• (Note)
  • The origin code of RNS and CCA was revised for the RecBole library.
  • PIRSP extends RNS with an item sequence encoder.
  • IntentRec is the contributor's own paper and implementation.

Dataset

Amazon Review 5-core dataset

• Provided subsets:

  • Musical Instruments (MI)
  • Automotive (AM)

• Data conversion scripts based on RecSysDatasets will be update soon.

Example Usage

python run_rbsr.py --model IntentRec --dataset MI