A-LLMRec: A PyTorch Implementation and Evaluation

This repository contains a PyTorch implementation of the paper:

"Large Language Models meet Collaborative Filtering: An Efficient All-round LLM-based Recommender System"
Kim et al., KDD 2024

---

📌 Objectives

This work focuses on two key goals:

• ✅ Reproduction: A correct and faithful implementation of the A-LLMRec architecture, including its two-stage training process.
• 📊 Evaluation: Performance testing on standard datasets and a comparative analysis with the results from the original authors.

This implementation was developed and evaluated on the Beauty_5 and Video_Game_5 Amazon review datasets.

---

🚀 Key Features

While remaining faithful to the original methodology, this implementation improves usability and clarity with:

• 🔁 Single, Self-Contained Script
  Entire experimental pipeline (data loading ➝ training ➝ evaluation) in one clean, well-documented script.

• 📦 Modern Data Handling
  Efficient use of pandas and PyTorch’s Dataset and DataLoader abstractions.

• 🧩 Modular, Clear Code
  Functions and classes (e.g., AlignmentModule) are clearly named after the components in the original paper.

• 💾 Resumable Training
  Automatic checkpointing enables you to resume training without losing progress.

• 📈 Comprehensive Reporting
  Logs results to TensorBoard, saves plots, and writes a summary of final metrics to .csv.

---

⚙️ How to Run

🔧 Prerequisites

• Python 3.8+
• PyTorch 1.12+
• Transformers
• Sentence-Transformers
• Pandas, NumPy, TQDM, Matplotlib

🛠️ Setup

Clone this repository:

git clone [your-repo-url]
cd [your-repo-name]

Install required packages:

pip install torch transformers sentence-transformers pandas numpy tqdm matplotlib

Prepare your data: Place your dataset file (e.g., Beauty_5.json) in a known location (e.g., Datasets folder in Google Drive).

▶️ Execution Open the a_llmrec_full_featured_implementation.py script and update the main block at the bottom:

if __name__ == '__main__':
    # ==========================================================================
    # 1. IMPORTANT: Update this path to the location of your .json dataset file
    # ==========================================================================
    dataset_path = "/content/drive/MyDrive/Datasets/Beauty_5.json"
    dataset_name = "Beauty_5"
    
    # 2. Set your desired K values for evaluation
    validation_k = 5  # K for validation during Stage 1 training
    test_k = 10       # K for final test set evaluation

    # 3. Run the experiment
    run_experiment(dataset_name, dataset_path, validation_k=validation_k, test_k=test_k)

📄 Citation If you use this implementation in your work, please cite the original paper:

@inproceedings{kim2024large,
  title={Large language models meet collaborative filtering: An efficient all-round llm-based recommender system},
  author={Kim, Sein and Kang, Hongseok and Choi, Seungyoon and Kim, Donghyun and Yang, Minchul and Park, Chanyoung},
  booktitle={Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
  year={2024}
}