MILL

The code implementation of paper MILL: Mutual Verification with Large Language Models for Zero-Shot Query Expansion

Prepare the Environment

conda create -n terrier python=3.8
conda activate terrier
pip install python-terrier openai retrying torch transformers pandarallel

The java home maybe need to be set when you use pyterrier

export JAVA_HOME= your_java_home

Before Run the Code

please set the api_key and base_url in both run.py and utils.py

Run the code

cd terrier
python run.py --dataset={} --qe={} --pos_lis={} --N={} --n={} --K={} --k={}

The description for the parameters:

• dataset: the dataset you want use. Candidate values: refer to pyterrier doc
• qe: the query expansion method name.
• pos_lis: the (start,end) queries that you want to use in the dataset
• N: the number of generated candidate documents $N$
• n: the number of selected generated documents $n$
• K: the number of PRF candidate documents $K$
• k: the number of selected PRF documents $k$