Skip to content

Latest commit

 

History

History
69 lines (50 loc) · 4.04 KB

regressions-clef06-fr.md

File metadata and controls

69 lines (50 loc) · 4.04 KB
Raw

Anserini Regressions: CLEF2006 Monolingual French

This page documents BM25 regression experiments for monolingual French document retrieval as part of the CLEF 2006 Multilingual Document Retrieval (Ad Hoc) Track. Associated data can be found on the CLEF test suites pages.

The exact configurations for these regressions are stored in this YAML file. Note that this page is automatically generated from this template as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead.

From one of our Waterloo servers (e.g., orca), the following command will perform the complete regression, end to end:

python src/main/python/run_regression.py --index --verify --search --regression clef06-fr

Indexing

Typical indexing command:

target/appassembler/bin/IndexCollection \
  -collection JsonCollection \
  -input /path/to/clef06-fr \
  -index indexes/lucene-index.clef06-fr/ \
  -generator DefaultLuceneDocumentGenerator \
  -threads 16 -storePositions -storeDocvectors -storeRaw -language fr \
  >& logs/log.clef06-fr &

The collection comprises news articles from ATS (SDA) and Le Monde totaling 177,452 documents. Since the original distribution is in a format that's slightly different from standard TREC collections, we used a preprocessing script to convert the collection into Anserini's JSON line format (we also applied a bit of light data cleaning using a script that has been lost; if you have problems reproducing our results, get in touch directly). The directory /path/to/clef06-fr/ should point to the location of the processed collection.

For additional details, see explanation of common indexing options.

Retrieval

Topics and qrels are stored here, which is linked to the Anserini repo as a submodule.

After indexing has completed, you should be able to perform retrieval as follows:

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.clef06-fr/ \
  -topics tools/topics-and-qrels/topics.clef06fr.mono.fr.txt \
  -topicreader TsvString \
  -output runs/run.clef06-fr.bm25.topics.clef06fr.mono.fr.txt \
  -bm25 -language fr &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -m map -m P.20 -m ndcg_cut.20 tools/topics-and-qrels/qrels.clef06fr.txt runs/run.clef06-fr.bm25.topics.clef06fr.mono.fr.txt

Effectiveness

With the above commands, you should be able to reproduce the following results:

MAP BM25
CLEF 2006 (Monolingual French) 0.3115
P20 BM25
CLEF 2006 (Monolingual French) 0.3184
nDCG@20 BM25
CLEF 2006 (Monolingual French) 0.4457