README.md

metadata

annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
language:
  - en
license:
  - apache-2.0
multilinguality:
  - monolingual
size_categories:
  - 10K<n<100K
source_datasets:
  - extended|other-MS^2
  - extended|other-Cochrane
task_categories:
  - summarization
  - text2text-generation
task_ids:
  - summarization-other-query-based-summarization
  - summarization-other-query-based-multi-document-summarization
  - summarization-other-scientific-documents-summarization
paperswithcode_id: multi-document-summarization
pretty_name: MSLR Shared Task

This is a copy of the MS^2 dataset, except the input source documents of its validation split have been replaced by a sparse retriever. The retrieval pipeline used:

• query: The background field of each example
• corpus: The union of all documents in the train, validation and test splits. A document is the concatenation of the title and abstract.
• retriever: BM25 via PyTerrier with default settings
• top-k strategy: "oracle", i.e. the number of documents retrieved, k, is set as the original number of input documents for each example

Retrieval results on the test set:

ndcg	recall@100	recall@1000	Rprec
0.4012	0.3780	0.6601	0.1833

Note: The abstract field of the validation split contains both the title and the abstract. Accordingly, the title field contains empty strings. This decision was made in order to simplify the retrieval pipeline.