Learning to Search for Datasets

Abstract

Over the years, search engines have developed to return a broad range of retrievable items, from documents to answers, people, locations, and products. Research datasets are increasingly being turned in retrievable items too. This raises a number of interesting challenges. Starting from the user end (What do users want from datasets) to increasing the retrievability of datasets (What kind of contextual information is available to enrich datasets so as to make the more easily retrieval) to optimizing rankers for datasets in the absence of large volumes of interaction data (How can we train learning to rank datasets algorithms in weakly supervised ways).