Agent4Rec: Revolutionizing Recommender Systems with Generative Agents

Introduction

Agent4Rec is a cutting-edge recommender system simulator that leverages the power of Large Language Models (LLMs) to create 1,000 generative agents. These agents, initialized from the MovieLens-1M dataset, simulate realistic user interactions with movie recommendations, providing unprecedented insights into human behavior in recommendation environments.

Key Features

• 1,000 LLM-Empowered Agents: Each agent embodies unique social traits and preferences, mimicking real-world diversity.
• Realistic Interactions: Agents engage with personalized movie recommendations in a page-by-page manner.
• Diverse Actions: Simulated users can watch, rate, evaluate, exit, and even conduct interviews about recommended content.
• Flexible Configuration: Supports various recommender systems and simulation settings.

How It Works

Agent4Rec simulates a dynamic environment where AI-powered agents interact with movie recommendations. The system allows researchers and developers to explore the potential of LLM-empowered generative agents in replicating genuine, independent human behavior within recommendation contexts.

Supported Recommender Systems

• Random: Randomly recommends items to users.
• Pop: Recommends popular items based on overall ratings.
• MF: Pretrained Matrix Factorization model with BPR loss.
• MultVAE: Pretrained Variational Autoencoder for collaborative filtering.
• LightGCN: Pretrained Graph Convolutional Network model.

Getting Started

Prerequisites

• Python 3.9.12 (Python 3.10+ may cause issues with the 'reckit' package)
• PyTorch 1.13.1+cu117

Installation

1. Set up a virtual environment and install PyTorch manually.
2. Install dependencies:

   pip install -r requirements.txt
   

3. Set up necessary environments:

   python setup.py build_ext --inplace
   

Running a Simulation

1. Export your OpenAI API key:

   export OPENAI_API_KEY=your_api_key_here
   

2. Run a quick start simulation:

   python main.py
   

3. For more advanced configurations:

   python main.py --simulation_name MyExp --modeltype MF --n_avatars 10 --max_pages 5 --items_per_page 4 --execution_mode parallel
   

Benefits and Applications

• Research Tool: Ideal for studying user behavior in recommendation systems.
• Algorithm Testing: Test and refine recommendation algorithms with realistic user simulations.
• User Experience Optimization: Gain insights to improve recommendation interfaces and strategies.
• Scalable Testing: Simulate large-scale user interactions without the need for real user studies.

Conclusion

Agent4Rec opens new avenues for understanding and improving recommender systems. By simulating realistic user interactions at scale, it provides valuable insights that can lead to more effective, personalized, and engaging recommendation experiences in real-world applications.