Skip to content

Latest commit

 

History

History
34 lines (24 loc) · 2.16 KB

README.md

File metadata and controls

34 lines (24 loc) · 2.16 KB

NeuralIR

Content

Content of the directories:

  • code: this folder contains the Matlab code for running the analyses and reproducing the experimental evaluation. It requires the MATTERS library. This folder also contains the source code of NVSM (Python) and DRMM (Python).
  • data: the runs, pools and measure files produced by the experimental evaluation (Matlab) and all the plots used in the paper.
  • figure: the plots used in the paper (pdf + Omnigraffle sources and .mat figures).
  • DRMM_repro: contains the code to prepare the input for DRMM algorithm.

NVSM and DRMM source code and settings

NVSM:

NVSM requires the following directory tree with the following folder names (example with Robust04 collection): - root: nvsm.py, nvsm_ny.py, indexing.py, utils.py, utils_ny.py, indri_stopwords.txt, stoplist-##.txt - corpus: robust04 - robust04: robust04, splits, queries, qrels - robust04: contains document collection - splits: contains validation.txt and test.txt - which refer to queries ids used for validation and test, respectively - queries: contains queries for target collection - qrels: contains qrels for target collection

	- trec_eval: folder containing trec_eval from NIST TREC

DRMM:

The code to compute the input data for DRMM algorithm is contained in the DRMM_repro directory. It can be run on the included data (cranfield collection) in the DRMM_repro/data directory after extracting to the same folder the data in the processed_corpus.zip archive.

References

  1. Stefano Marchesin, Alberto Purpura, Gianmaria Silvello: Focal Elements of Neural Information Retrieval Models. An Outlook through a Reproducibility Study. IP&M 2019. In press (https://doi.org/10.1016/j.ipm.2019.102109)
  2. Christophe Van Gysel, Maarten de Rijke, Evangelos Kanoulas: Neural Vector Spaces for Unsupervised Information Retrieval. ACM Trans. Inf. Syst. 36(4): 38:1-38:25 (2018). (https://dl.acm.org/citation.cfm?doid=3211967.3196826)
  3. Jiafeng Guo, Yixing Fan, Qingyao Ai, W. Bruce Croft: A Deep Relevance Matching Model for Ad-hoc Retrieval. CIKM 2016: 55-64. (https://dl.acm.org/citation.cfm?doid=2983323.2983769)