tallesl/trec-corpus

TREC corpus

A corpus from Text REtrieval Conference (TREC), featuring 528k documents from the following sources:

• Foreign Broadcast Information Service (FBIS)
• Federal Register (FR94)
• Financial Times (FT)
• LA Times (LATIMES)

Using the dataset

Thanks to the ir_datasets package, you can load and use it in very few steps.

1. Download the dataset.
2. Extract the zip file and place the four extracted folders (FBIS, FR94, FT, and LATIMES) into ~/.ir_datasets/disks45/corpus/.
3. Load and iterate it with ir_datasets:

from ir_datasets import load  # pip install ir_datasets

print('Press Enter to load the next document.\n')

for doc in load('disks45/nocr').docs_iter():
    input()
    print(f'ID: {doc.doc_id}\n')
    print(f'Title: {doc.title}\n')
    print(f'Body: {doc.body[:500]}...\n')
    print('=' * 50)

TREC-7 topics

You can find 20 selected topics from TREC-7 here. They may be used (together with the dataset) on benchmarking information retrieval applications.