NPL collection

The NPL (also known as the VASWANI) collection is a collection of around 10,000 document titles. It has a bit of a reputation of messing up people's experiments.

The bits

• doc-text - The documents in text form
• query-text - The queries in text form
• doc-vecs - The documents represented by term IDs
• query-vecs - The queries represented by term IDs
• term-vocab - Table of terms with corresponding IDs
• rlv-ass - Relevance assesments
• term-vecs - A file of numbers, don't know what they do
• term-vocab - Another file of numbers

• npl.tar.gz - All the bits put together