Information Technology Reference
Benchmark Datasets for Learning to Rank
In this part, we introduce the publicly available benchmark datasets for learning to
rank, which provide good experimental environments for researchers to study differ-
ent kinds of learning to rank algorithms. We first make a detailed introduction to the
LETOR datasets, the first benchmark datasets for the research on learning to rank.
Then we present the experimental results for typical learning to rank algorithms
provided with the LETOR datasets and give some discussions on these algorithms.
After that, we briefly mention two newly released benchmark datasets, which are
much larger than LETOR and can support more types of learning to rank research.
After reading this part, the readers are expected to be familiar with existing
benchmark datasets, and know how to test their own algorithms on these datasets.