Topic Modeling for Search and Exploration in Multivariate Research Data Repositories

Abstract

Huge amounts of multivariate research data are produced and made publicly available in digital libraries. Little research focused on similarity functions that take multivariate data documents as a whole into account. Such similarity functions are highly beneficial for users, by enabling them to browse and query large collections of multivariate data using nearest-neighbor indexing. In this paper we tackle this challenge and propose a novel similarity function for multivariate data documents based on topic-modeling. Based on a previously developed bag-of-words approach for multivariate data, we can then learn a topic model for a collection of multivariate data documents and represent each document as a mixture of topics. This representation is very suitable for efficient nearest-neighbor indexing and clustering according to the topic distribution of a document. We present a use-case where we apply this approach to retrieval of multivariate data in the field of climate research.