Skip to content

Latest commit

 

History

History
70 lines (51 loc) · 3.88 KB

regressions-ntcir8-zh.md

File metadata and controls

70 lines (51 loc) · 3.88 KB
Raw

Anserini Regressions: NTCIR-8 Monolingual Chinese

This page documents BM25 regression experiments for NTCIR-8 ACLIA (IR4QA subtask), monolingual Chinese topics. The description of the document collection can be found in the NTCIR-8 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 ntcir8-zh

Indexing

Typical indexing command:

target/appassembler/bin/IndexCollection \
  -collection CleanTrecCollection \
  -input /path/to/ntcir8-zh \
  -index indexes/lucene-index.ntcir8-zh/ \
  -generator DefaultLuceneDocumentGenerator \
  -threads 16 -storePositions -storeDocvectors -storeRaw -language zh -uniqueDocid -optimize \
  >& logs/log.ntcir8-zh &

The collection comprises Xinhua articles from 2002-2005, totaling 308,845 documents, from LDC2007T38: Chinese Gigaword Third Edition. We build the index directly from the raw LDC data: the directory /path/to/ntcir8-zh/ should point to the directory data/xin_cmn/ from LDC2007T38. In that directory, there should be 48 gzipped files matching the pattern xin_cmn_200[2-5]*.

For additional details, see explanation of common indexing options.

Retrieval

Topics and qrels are stored in src/main/resources/topics-and-qrels/, downloaded from the NTCIR Test Collection page:

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

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.ntcir8-zh/ \
  -topics src/main/resources/topics-and-qrels/topics.ntcir8zh.eval.txt \
  -topicreader TsvString \
  -output runs/run.ntcir8-zh.bm25.topics.ntcir8zh.eval.txt \
  -bm25 -language zh &

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.ntcir8.eval.txt runs/run.ntcir8-zh.bm25.topics.ntcir8zh.eval.txt

Effectiveness

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

MAP BM25
NTCIR-8 ACLIA (IR4QA subtask, Monolingual Chinese) 0.4014
P20 BM25
NTCIR-8 ACLIA (IR4QA subtask, Monolingual Chinese) 0.3849
nDCG@20 BM25
NTCIR-8 ACLIA (IR4QA subtask, Monolingual Chinese) 0.4757