This project provides VirtualTaobao, a simulator for online retail recommendation systems, designed for researchers in recommendation systems and reinforcement learning. It offers a realistic, data-driven environment to train and evaluate recommendation algorithms by simulating customer interactions and item selection processes.

How It Works

VirtualTaobao models customers with static and dynamic attributes, and items with 27-dimensional features. The simulation involves a customer query, item retrieval, a learned model assigning weights to items, and presenting the top 10 items. Customer actions (click, browse next, leave) drive the simulation, providing feedback for model training. This approach offers a "live" environment mirroring real-world Taobao interactions for RL agents.

Quick Start & Requirements

• Primary install: pip install -e .
• Prerequisites: Python, PyTorch (for supervised learning example).
• Links: Supervised Learning Example, Reinforcement Learning Example

Highlighted Details

• Trained on real-world, anonymized Taobao data.
• Provides OpenAI Gym interface for RL integration.
• Includes example implementations for both supervised and reinforcement learning.
• Simulates customer attributes (age, gender, history) and item attributes (price, sales, CTR).

Maintenance & Community

• Outcome of joint work between Nanjing University and Alibaba Group.
• No explicit community links (Discord/Slack) or roadmap provided in the README.

Licensing & Compatibility

• License: Not explicitly stated in the README.
• Usage: Free for research use, requires proper credit to the AAAI'19 paper. Commercial use implications are unclear without a stated license.

Limitations & Caveats

The current V0 model is based on a "middle-scaled" dataset, with larger models planned. The specific itemset content is not exposed due to confidence issues. The license for commercial use is not specified.