Skip to content

Latest commit

 

History

History
114 lines (80 loc) · 6.86 KB

regressions-msmarco-doc-docTTTTTquery-per-passage.md

File metadata and controls

114 lines (80 loc) · 6.86 KB
Raw

Anserini: Regressions for MS MARCO Document Ranking

This page documents regression experiments for the MS MARCO document ranking task, which is integrated into Anserini's regression testing framework. Note that there are four different regression conditions for this task, and this page describes the following:

  • Indexing Condition: each MS MARCO document is first segmented into passages, each passage is treated as a unit of indexing
  • Expansion Condition: doc2query-T5

In the passage indexing condition, we select the score of the highest-scoring passage from a document as the score for that document to produce a document ranking; this is known as the MaxP technique. All four conditions are described in detail here, in the context of doc2query-T5.

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.

Indexing

Typical indexing command:

nohup sh target/appassembler/bin/IndexCollection -collection JsonCollection \
 -input /path/to/msmarco-doc-docTTTTTquery-per-passage \
 -index indexes/lucene-index.msmarco-doc-docTTTTTquery-per-passage.pos+docvectors+raw \
 -generator DefaultLuceneDocumentGenerator \
 -threads 1 -storePositions -storeDocvectors -storeRaw \
  >& logs/log.msmarco-doc-docTTTTTquery-per-passage &

The directory /path/to/msmarco-doc-docTTTTTquery-per-passage/ should be a directory containing the expanded document collection; see this link for how to prepare this collection.

For additional details, see explanation of common indexing options.

Retrieval

Topics and qrels are stored in src/main/resources/topics-and-qrels/. The regression experiments here evaluate on the 5193 dev set questions.

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

nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-doc-docTTTTTquery-per-passage.pos+docvectors+raw \
 -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-doc.dev.txt \
 -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-default.topics.msmarco-doc.dev.txt \
 -bm25 -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 1000 &

nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-doc-docTTTTTquery-per-passage.pos+docvectors+raw \
 -topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-doc.dev.txt \
 -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.topics.msmarco-doc.dev.txt \
 -bm25 -bm25.k1 2.56 -bm25.b 0.59 -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 1000 &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -m map -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.msmarco-doc.dev.txt runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-default.topics.msmarco-doc.dev.txt

tools/eval/trec_eval.9.0.4/trec_eval -m map -c -m recall.100 -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.msmarco-doc.dev.txt runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.topics.msmarco-doc.dev.txt

Effectiveness

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

MAP BM25 (default) BM25 (tuned)
MS MARCO Doc: Dev 0.3182 0.3211
R@100 BM25 (default) BM25 (tuned)
MS MARCO Doc: Dev 0.8481 0.8627
R@1000 BM25 (default) BM25 (tuned)
MS MARCO Doc: Dev 0.9490 0.9530

Explanation of settings:

  • The setting "default" refers the default BM25 settings of k1=0.9, b=0.4.
  • The setting "tuned" refers to k1=2.56, b=0.59, tuned to optimize for recall@100 (i.e., for first-stage retrieval) on 2019/12.

In these runs, we are retrieving the top 1000 hits for each query and using trec_eval to evaluate all 1000 hits. Since we're in the passage condition, we fetch the 10000 passages and select the top 1000 documents using MaxP. This lets us measure R@100 and R@1000; the latter is particularly important when these runs are used as first-stage retrieval. Beware, an official MS MARCO document ranking task leaderboard submission comprises only 100 hits per query. See this page for details on Anserini baseline runs that were submitted to the official leaderboard.

The MaxP passage retrieval functionality is only available in SearchCollection; we use a simple script to convert the output back into the MS MARCO format for evaluation.

To generate an MS MARCO submission with the BM25 tuned parameters, corresponding to "BM25 (tuned)" above:

$ target/appassembler/bin/SearchCollection -topicreader TsvString \
   -topics src/main/resources/topics-and-qrels/topics.msmarco-doc.dev.txt \
   -index indexes/lucene-index.msmarco-doc-docTTTTTquery-per-passage.pos+docvectors+raw \
   -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.trec \
   -bm25 -bm25.k1 2.56 -bm25.b 0.59 -hits 1000 \
   -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 100

$ python tools/scripts/msmarco/convert_trec_to_msmarco_run.py \
   --input runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.trec \
   --output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.txt

$ python tools/scripts/msmarco/msmarco_doc_eval.py \
   --judgments src/main/resources/topics-and-qrels/qrels.msmarco-doc.dev.txt \
   --run runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.txt

#####################
MRR @100: 0.32081861579183746
QueriesRanked: 5193
#####################

This run corresponds to the MS MARCO document ranking leaderboard entry "Anserini's BM25 + doc2query-T5 expansion (per passage), parameters tuned for recall@100 (k1=2.56, b=0.59)" dated 2020/12/11, and is reported in the Lin et al. (SIGIR 2021) Pyserini paper.