README.md

metadata

license: mit
task_categories:
  - text-retrieval

Simulated Query and Document Streams of LoTTE Forum Dataset

This dataset contains the simulated streams for the paper "MURR: Model Updating with Regularized Replay for Searching a Document Stream". Please refer to the paper for the detailed sampling process.

The code for sampling the queries, qrels, and documents are documents in sampling.py. It is not intended to be ran but for recording purposes.

The random numbers for sampling are also recorded in the files for futrue references.

Dataset Structure

root
├───sampling.py
│
├───docs
│   └───test_D{1,2,3}_{0,1,2,3,4}.parquet
│   
├───queries
│   └───test_D{1,2,3}_{0,1,2,3,4}.jsonl
│   
└───qrels
    └───test_D{1,2,3}_{0,1,2,3,4}.qrels

D{1,2,3} indicates different streams and the subsequent {0,1,2,3,4} are the sessions ids.

Citation

@inproceedings{ecir2025murr,
    author = {Eugene Yang, Nicola Tonellotto, Dawn Lawrie, Sean MacAvaney, James Mayfield, Douglas Oard and Scott Miller},
    title = {MURR: Model Updating with Regularized Replay for Searching a Document Stream},
    booktitle = {Proceedings of the 47th European Conference on Information Retrieval (ECIR)},
    year = {2025},
}