User-centered semantic dataset retrieval

Finding relevant research data is an increasingly important but time-consuming task in daily research practice. Several studies report on difficulties in dataset search, e.g., scholars retrieve only partial pertinent data, and important information can not be displayed in the user interface. Overcoming these problems has motivated a number of research efforts in computer science, such as text mining and semantic search. In particular, the emergence of the Semantic Web opens a variety of novel research perspectives. Motivated by these challenges, the overall aim of this work is to analyze the current obstacles in dataset search and to propose and develop a novel semantic dataset search. The studied domain is biodiversity research, a domain that explores the diversity of life, habitats and ecosystems. This thesis has three main contributions: (1) We evaluate the current situation in dataset search in a user study, and we compare a semantic search with a classical keyword search to explore the suitability of semantic web technologies for dataset search. (2) We generate a question corpus and develop an information model to figure out on what scientific topics scholars in biodiversity research are interested in. Moreover, we also analyze the gap between current metadata and scholarly search interests, and we explore whether metadata and user interests match. (3) We propose and develop an improved dataset search based on three components: (A) a text mining pipeline, enriching metadata and queries with semantic categories and URIs, (B) a retrieval component with a semantic index over categories and URIs and (C) a user interface that enables a search within categories and a search including further hierarchical relations. Following user centered design principles, we ensure user involvement in various user studies during the development process.