Skip to content

Latest commit

 

History

History
70 lines (51 loc) · 3.89 KB

regressions-trec02-ar.md

File metadata and controls

70 lines (51 loc) · 3.89 KB
Raw

Anserini Regressions: TREC 2002 Monolingual Arabic

This page documents BM25 regression experiments for monolingual Arabic document retrieval as part of the TREC 2002 CLIR Track. The description of the document collection can be found on the TREC data page.

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 trec02-ar

Indexing

Typical indexing command:

target/appassembler/bin/IndexCollection \
  -collection CleanTrecCollection \
  -input /path/to/trec02-ar \
  -index indexes/lucene-index.trec02-ar/ \
  -generator DefaultLuceneDocumentGenerator \
  -threads 16 -storePositions -storeDocvectors -storeRaw -language ar \
  >& logs/log.trec02-ar &

The collection comprises Agence France Presse (AFP) Arabic newswire, from LDC2001T55 (Arabic Newswire Part 1). Inside the LDC2007T38 distribution, there should be a directory named transcripts, which contains 2,337 gzipped files in 7 directories, 1994 ... 2000. The path above /path/to/trec02-ar/ should point to this transcripts/ directory. The collection contains 383,872 documents.

For additional details, see explanation of common indexing options.

Retrieval

Topics and qrels are stored in src/main/resources/topics-and-qrels/, downloaded from NIST's page for non-English topics and non-English relevance judgments:

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

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.trec02-ar/ \
  -topics src/main/resources/topics-and-qrels/topics.trec02ar-ar.txt \
  -topicreader Trec \
  -output runs/run.trec02-ar.bm25.topics.trec02ar-ar.txt \
  -bm25 -language ar &

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 src/main/resources/topics-and-qrels/qrels.trec02ar.txt runs/run.trec02-ar.bm25.topics.trec02ar-ar.txt

Effectiveness

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

MAP BM25
TREC 2002 (Monolingual Arabic) 0.2932
P20 BM25
TREC 2002 (Monolingual Arabic) 0.3610
nDCG@20 BM25
TREC 2002 (Monolingual Arabic) 0.4056