ChatShop

Implementation of the ChatShop environment and search agents for the paper:

ChatShop: Interactive Information Seeking with Language Agents
Sanxing Chen, Sam Wiseman, Bhuwan Dhingra

If you find this work useful in your research, please cite:

@misc{chen2024chatshop,
      title={ChatShop: Interactive Information Seeking with Language Agents}, 
      author={Sanxing Chen and Sam Wiseman and Bhuwan Dhingra},
      year={2024},
      eprint={2404.09911},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2404.09911}, 
}

Setup

0. Clone the repository
1. Create a new conda environment with Python

   > conda create -n chatshop python=3.8.13
   > conda activate chatshop

2. Run setup_ws.sh -d all to download product data from WebShop
3. Run ./run_server.sh to start the ChatShop server
   • you can test it by running the following command:

     curl "localhost:3000/api/search_results?keywords=protein%20shake&gid=1"

4. Test agents with the ChatShop environment

   usage: ipofn.py [-h] [--model {gpt-3.5-turbo,gpt-3.5-turbo-1106,gpt-4-1106-preview}] [--strategy {random,allq,interleave}] [--cot] [--game_mode {full,part,subj}]
                   [--qmode {none,open,compare,mix}] [--num_question NUM_QUESTION]
                   exp_name
   
   positional arguments:
   exp_name              experiment name
   
   optional arguments:
   -h, --help            show this help message and exit
   --model {gpt-3.5-turbo,gpt-3.5-turbo-1106,gpt-4-1106-preview}
                           Agent: select chat model
   --strategy {random,allq,interleave}
                           Agent: action strategy
   --cot                 Agent: ReAct style prompting
   --game_mode {full,part,subj}
                           Env: task ambiguity mode, full is the original WebShop instructions, part is the instructions without salient attributes, subj is the
                           instructions without any attributes.
   --qmode {none,open,compare,mix}
                           Env: communication channel, none means no communication between agent and shopper, open enables open-ended communication, compare is the
                           instance-based communication.
   --num_question NUM_QUESTION
                           Env: maximum number of interactions with the simulated shopper.
   
   > python -m web_agent_site.models.ipofn test --game_mode subj --qmode open --num_question 1

Acknowledgements

This repository sourced data and code from WebShop.