metadata

language: en
tags:
  - PROP
  - Pretrain4IR
license: apache-2.0
datasets:
  - msmarco

PROP-marco-step400k

PROP, Pre-training with Representative wOrds Prediction, is a new pre-training method tailored for ad-hoc retrieval. PROP is inspired by the classical statistical language model for IR, specifically the query likelihood model, which assumes that the query is generated as the piece of text representative of the “ideal” document. Based on this idea, we construct the representative words prediction (ROP) task for pre-training. The full paper can be found here.

This model is pre-trained with more steps than PROP-marco on MS MARCO document corpus, and used at the MS MARCO Document Ranking Leaderboard where we reached 1st place.

Citation

If you find our work useful, please consider citing our paper:

@inproceedings{DBLP:conf/wsdm/MaGZFJC21,
  author    = {Xinyu Ma and
               Jiafeng Guo and
               Ruqing Zhang and
               Yixing Fan and
               Xiang Ji and
               Xueqi Cheng},
  editor    = {Liane Lewin{-}Eytan and
               David Carmel and
               Elad Yom{-}Tov and
               Eugene Agichtein and
               Evgeniy Gabrilovich},
  title     = {{PROP:} Pre-training with Representative Words Prediction for Ad-hoc
               Retrieval},
  booktitle = {{WSDM} '21, The Fourteenth {ACM} International Conference on Web Search
               and Data Mining, Virtual Event, Israel, March 8-12, 2021},
  pages     = {283--291},
  publisher = {{ACM}},
  year      = {2021},
  url       = {https://doi.org/10.1145/3437963.3441777},
  doi       = {10.1145/3437963.3441777},
  timestamp = {Wed, 07 Apr 2021 16:17:44 +0200},
  biburl    = {https://dblp.org/rec/conf/wsdm/MaGZFJC21.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}