DUC 2005

The DUC 2005 data set is a dataset for summarization which consists of 50 document collections of 25 documents each; each document collection includes a human-written query. Each document collection additionally has five human-written “reference” summaries (250 words long, each) that serve as the gold standard

Source: Search-based Structured Prediction Image Source: https://duc.nist.gov/duc2005/tasks.html