Skip to content

Latest commit

 

History

History
86 lines (62 loc) · 4.57 KB

regressions-backgroundlinking20.md

File metadata and controls

86 lines (62 loc) · 4.57 KB
Raw

Anserini Regressions: TREC 2020 News Background Linking

Models: various bag-of-words approaches

This page describes regressions for the background linking task in the TREC 2020 News Track. 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 backgroundlinking20

Indexing

Typical indexing command:

target/appassembler/bin/IndexCollection \
  -collection WashingtonPostCollection \
  -input /path/to/wapo.v3 \
  -index indexes/lucene-index.wapo.v3/ \
  -generator WashingtonPostGenerator \
  -threads 1 -storePositions -storeDocvectors -storeRaw \
  >& logs/log.wapo.v3 &

The directory /path/to/core18/ should be the root directory of the TREC Washington Post Corpus v3, i.e., ls /path/to/core18/ should bring up a single JSON file.

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. They are downloaded from NIST:

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

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.wapo.v3/ \
  -topics tools/topics-and-qrels/topics.backgroundlinking20.txt \
  -topicreader BackgroundLinking \
  -output runs/run.wapo.v3.bm25.topics.backgroundlinking20.txt \
  -backgroundlinking -backgroundlinking.k 100 -bm25 -hits 100 &

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.wapo.v3/ \
  -topics tools/topics-and-qrels/topics.backgroundlinking20.txt \
  -topicreader BackgroundLinking \
  -output runs/run.wapo.v3.bm25+rm3.topics.backgroundlinking20.txt \
  -backgroundlinking -backgroundlinking.k 100 -bm25 -rm3 -hits 100 &

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.wapo.v3/ \
  -topics tools/topics-and-qrels/topics.backgroundlinking20.txt \
  -topicreader BackgroundLinking \
  -output runs/run.wapo.v3.bm25+rm3+df.topics.backgroundlinking20.txt \
  -backgroundlinking -backgroundlinking.datefilter -backgroundlinking.k 100 -bm25 -rm3 -hits 100 &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -c -M1000 -m map -c -M1000 -m ndcg_cut.5 tools/topics-and-qrels/qrels.backgroundlinking20.txt runs/run.wapo.v3.bm25.topics.backgroundlinking20.txt

tools/eval/trec_eval.9.0.4/trec_eval -c -M1000 -m map -c -M1000 -m ndcg_cut.5 tools/topics-and-qrels/qrels.backgroundlinking20.txt runs/run.wapo.v3.bm25+rm3.topics.backgroundlinking20.txt

tools/eval/trec_eval.9.0.4/trec_eval -c -M1000 -m map -c -M1000 -m ndcg_cut.5 tools/topics-and-qrels/qrels.backgroundlinking20.txt runs/run.wapo.v3.bm25+rm3+df.topics.backgroundlinking20.txt

Effectiveness

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

MAP BM25 +RM3 +RM3+DF
TREC 2020 Topics 0.3286 0.4528 0.3438
nDCG@5 BM25 +RM3 +RM3+DF
TREC 2020 Topics 0.5231 0.5696 0.5304