Models: various bag-of-words approaches with (vanilla) doc2query expansions
This page documents regression experiments for the TREC 2017 Complex Answer Retrieval (CAR) section-level passage retrieval task (v2.0), with doc2query expansions, as proposed in the following paper:
Rodrigo Nogueira, Wei Yang, Jimmy Lin, Kyunghyun Cho. Document Expansion by Query Prediction. arxiv:1904.08375
These experiments are integrated into Anserini's regression testing framework. For more complete instructions on how to run end-to-end experiments, refer to this 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 car17v2.0-doc2query
Typical indexing command:
bin/run.sh io.anserini.index.IndexCollection \
-threads 30 \
-collection JsonCollection \
-input /path/to/car-paragraphCorpus.v2.0-doc2query \
-generator DefaultLuceneDocumentGenerator \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-storePositions -storeDocvectors -storeRaw \
>& logs/log.car-paragraphCorpus.v2.0-doc2query &
The directory /path/to/car17v2.0-doc2query
should be the root directory of Complex Answer Retrieval (CAR) paragraph corpus (v2.0) that has been augmented with the doc2query expansions, i.e., collection_jsonl_expanded_topk10/
as described in this page.
For additional details, see explanation of common indexing options.
The "benchmarkY1-test" topics and qrels (v2.0) are stored here, which is linked to the Anserini repo as a submodule. They are downloaded from the CAR website:
Specifically, this is the section-level passage retrieval task with automatic ground truth.
After indexing has completed, you should be able to perform retrieval as follows:
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-topics tools/topics-and-qrels/topics.car17v2.0.benchmarkY1test.txt \
-topicReader Car \
-output runs/run.car-paragraphCorpus.v2.0-doc2query.bm25.topics.car17v2.0.benchmarkY1test.txt \
-bm25 &
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-topics tools/topics-and-qrels/topics.car17v2.0.benchmarkY1test.txt \
-topicReader Car \
-output runs/run.car-paragraphCorpus.v2.0-doc2query.bm25+rm3.topics.car17v2.0.benchmarkY1test.txt \
-bm25 -rm3 &
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-topics tools/topics-and-qrels/topics.car17v2.0.benchmarkY1test.txt \
-topicReader Car \
-output runs/run.car-paragraphCorpus.v2.0-doc2query.bm25+ax.topics.car17v2.0.benchmarkY1test.txt \
-bm25 -axiom -axiom.deterministic -rerankCutoff 20 &
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-topics tools/topics-and-qrels/topics.car17v2.0.benchmarkY1test.txt \
-topicReader Car \
-output runs/run.car-paragraphCorpus.v2.0-doc2query.ql.topics.car17v2.0.benchmarkY1test.txt \
-qld &
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-topics tools/topics-and-qrels/topics.car17v2.0.benchmarkY1test.txt \
-topicReader Car \
-output runs/run.car-paragraphCorpus.v2.0-doc2query.ql+rm3.topics.car17v2.0.benchmarkY1test.txt \
-qld -rm3 &
bin/run.sh io.anserini.search.SearchCollection \
-index indexes/lucene-index.car-paragraphCorpus.v2.0-doc2query/ \
-topics tools/topics-and-qrels/topics.car17v2.0.benchmarkY1test.txt \
-topicReader Car \
-output runs/run.car-paragraphCorpus.v2.0-doc2query.ql+ax.topics.car17v2.0.benchmarkY1test.txt \
-qld -axiom -axiom.deterministic -rerankCutoff 20 &
Evaluation can be performed using trec_eval
:
bin/trec_eval -c -m map -c -m recip_rank tools/topics-and-qrels/qrels.car17v2.0.benchmarkY1test.txt runs/run.car-paragraphCorpus.v2.0-doc2query.bm25.topics.car17v2.0.benchmarkY1test.txt
bin/trec_eval -c -m map -c -m recip_rank tools/topics-and-qrels/qrels.car17v2.0.benchmarkY1test.txt runs/run.car-paragraphCorpus.v2.0-doc2query.bm25+rm3.topics.car17v2.0.benchmarkY1test.txt
bin/trec_eval -c -m map -c -m recip_rank tools/topics-and-qrels/qrels.car17v2.0.benchmarkY1test.txt runs/run.car-paragraphCorpus.v2.0-doc2query.bm25+ax.topics.car17v2.0.benchmarkY1test.txt
bin/trec_eval -c -m map -c -m recip_rank tools/topics-and-qrels/qrels.car17v2.0.benchmarkY1test.txt runs/run.car-paragraphCorpus.v2.0-doc2query.ql.topics.car17v2.0.benchmarkY1test.txt
bin/trec_eval -c -m map -c -m recip_rank tools/topics-and-qrels/qrels.car17v2.0.benchmarkY1test.txt runs/run.car-paragraphCorpus.v2.0-doc2query.ql+rm3.topics.car17v2.0.benchmarkY1test.txt
bin/trec_eval -c -m map -c -m recip_rank tools/topics-and-qrels/qrels.car17v2.0.benchmarkY1test.txt runs/run.car-paragraphCorpus.v2.0-doc2query.ql+ax.topics.car17v2.0.benchmarkY1test.txt
With the above commands, you should be able to reproduce the following results:
MAP | BM25 | +RM3 | +Ax | QL | +RM3 | +Ax |
---|---|---|---|---|---|---|
TREC 2017 CAR: benchmarkY1test (v2.0) | 0.1807 | 0.1529 | 0.1470 | 0.1752 | 0.1447 | 0.1339 |
MRR | BM25 | +RM3 | +Ax | QL | +RM3 | +Ax |
TREC 2017 CAR: benchmarkY1test (v2.0) | 0.2750 | 0.2289 | 0.2186 | 0.2653 | 0.2144 | 0.1981 |