Skip to content

Latest commit

 

History

History
110 lines (73 loc) · 7.48 KB

regressions-dl20-doc-docTTTTTquery-per-passage.md

File metadata and controls

110 lines (73 loc) · 7.48 KB
Raw

Anserini: Regressions for DL20 (Doc) w/ per-passage docTTTTTquery

This page describes experiments, integrated into Anserini's regression testing framework, for the TREC 2020 Deep Learning Track (Document Ranking Task) on the MS MARCO document collection using relevance judgments from NIST.

Note that the NIST relevance judgments provide far more relevant documents per topic, unlike the "sparse" judgments provided by Microsoft (these are sometimes called "dense" judgments to emphasize this contrast). For additional instructions on working with MS MARCO document collection, refer to this page.

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 45 topics for which NIST has provided judgments as part of the TREC 2020 Deep Learning Track. The original data can be found here.

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.dl20.txt \
 -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-default.topics.dl20.txt \
 -bm25 -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 100 &

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.dl20.txt \
 -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-default+rm3.topics.dl20.txt \
 -bm25 -rm3 -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 100 &

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.dl20.txt \
 -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.topics.dl20.txt \
 -bm25 -bm25.k1 2.56 -bm25.b 0.59 -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 100 &

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.dl20.txt \
 -output runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned+rm3.topics.dl20.txt \
 -bm25 -bm25.k1 2.56 -bm25.b 0.59 -rm3 -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 100 &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c src/main/resources/topics-and-qrels/qrels.dl20-doc.txt runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-default.topics.dl20.txt

tools/eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c src/main/resources/topics-and-qrels/qrels.dl20-doc.txt runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-default+rm3.topics.dl20.txt

tools/eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c src/main/resources/topics-and-qrels/qrels.dl20-doc.txt runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned.topics.dl20.txt

tools/eval/trec_eval.9.0.4/trec_eval -m map -c -m ndcg_cut.10 -c -m recip_rank -c -m recall.100 -c src/main/resources/topics-and-qrels/qrels.dl20-doc.txt runs/run.msmarco-doc-docTTTTTquery-per-passage.bm25-tuned+rm3.topics.dl20.txt

Effectiveness

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

MAP BM25 (default) +RM3 BM25 (tuned) +RM3
DL20 (Doc) 0.4150 0.4269 0.4042 0.4023
NDCG@10 BM25 (default) +RM3 BM25 (tuned) +RM3
DL20 (Doc) 0.5957 0.5848 0.5931 0.5723
RR BM25 (default) +RM3 BM25 (tuned) +RM3
DL20 (Doc) 0.9361 0.8944 0.9469 0.9150
R@100 BM25 (default) +RM3 BM25 (tuned) +RM3
DL20 (Doc) 0.6201 0.6443 0.6192 0.6392

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 using the MS MARCO document sparse judgments to optimize for recall@100 (i.e., for first-stage retrieval) on 2019/12.

Settings tuned on the MS MARCO document sparse judgments may not work well on the TREC dense judgments.

Note that retrieval metrics are computed to depth 100 hits per query (as opposed to 1000 hits per query for DL20 passage ranking). Also, remember that we keep qrels of all relevance grades, unlike the case for DL20 passage ranking, where relevance grade 1 needs to be discarded when computing certain metrics.