Fair TREC Tools

Public tools for working with the Fair TREC data.

Environment & Compilation

The provided Conda environment spec will install all required dependencies, except for the AWS CLI tools needed for downloading the Open Corpus:

conda env create -f environment.yml
conda activate fairtrec

High-throughput data processing tools, such as the subsetter, are implemented in Rust (installed in the Conda environment); to build, run:

cargo build --release

Downloading Data

You need two sets of data:

1. The released data files from the Fair TREC web site, stored in data/ai2-trec-release

2. The Open Corpus, downloaded to data/corpus with:

   aws s3 cp --no-sign-request --recursive s3://ai2-s2-research-public/open-corpus/2020-05-27/ data/corpus
   

Subsetting the Corpus

To re-generate the OpenCorpus subet containing all files in the paper metadata file, run:

./target/release/subset-corpus -M data/ai2-trec-release/paper_metadata.csv \
    -o data/corpus-subset-for-meta.gz data/corpus

To generate a subset based on the candidate sets from query records, run:

./target/release/subset-corpus -Q data/TREC-Competition-training-sample.json \
    -o data/corpus-subset-for-queries.jsonl.gz data/corpus

The subset command also produces metadata CSV alongside the compressed JSON output.

The --help option works and will produce usage help.