Dataset search in biodiversity research: Do metadata in data repositories reflect scholarly information needs?

Challenges to be addressed in computer science Besides domain-specific metadata standards, the usage of controlled vocabularies is the second important building block for successful dataset retrieval. According to the RDA Data Maturity Model [66], controlled vocabularies ensure interoperability (I2-01M). From computer science perspective, aligned terminologies that ideally provide URIs to concepts in ontologies allow further technical improvements. Computer science research can contribute to improvements for dataset search by develop- ing methods and software tools that facilitate standard-compliant metadata provision ideally at the time of data collection, thus ensuring metadata standards to be actually used by data pro- viders. Such tools for metadata creation should provide the opportunity to use controlled vocabularies for the data description. In a best case scenario, even the raw data contain aligned terminologies facilitating automatic metadata creation. In the Life Sciences, a variety of con- trolled terminologies and terminology providers already exist. We provide a list of ontology and semantic service providers in our GitHub repository (https://github.com/fusion-jena/ QuestionsMetadataBiodiv/blob/master/biodivTerminologyServices.md). For legacy data, data structure changes are less feasable and therefore, other computer sci- ence techniques are necessary to extract additional information and to improve metadata. The text mining community has already developed various taggers and pipelines to extract organ-

isms [64], chemistry items [62] or genes [72] from textual resources. Recently, a tagger has been developed for biodiversity metadata [65] identifying environmental terms, data parame- ters, processes and materials. These annotations could support automatic facet or category cre- ation. In addition, some of these taggers already provide a linkage to the Linked Open Data (LOD) (https://www.lod-cloud.net/) Cloud and controlled vocabularies. Our results of the content analysis reveals that text mining can help to overcome the shortcomings of current metadata and support the enrichment of metadata with additional relevant information for data discovery. However, more investigations are needed to determine how many false positive annotations occur and if there is a need to further adapt existing pipelines for metadata enrichment. Semantic enrichment also allows the integration of schema.org (https://schema.org) or bioschemas.org (https://bioschemas.org/). Driven by Google and the RDA Data Discovery Par- adigm Interest Group, this community initiative aims to provide structured information for HTML web pages such as unique identifiers, persons, locations or time information. That sup- ports external search engines or data providers to crawl the landing pages of search applica- tions provided by the data repositories per dataset. Finally, with controlled vocabularies and ontological enrichment full semantic search sys- tems are possible. First approaches were mainly developed for publications in bio-medicine

and health sciences, e.g., [73–77]. Semantic search systems need controlled vocabularies for proper entity recognition and disambiguation. That enables developers of search systems in data portals to expand a search query on semantically related terms such as synonyms or more specific or broader terms [76, 77] or faceted search [74, 75, 77]. Due to these numerous studies in bio-medicine, large data repositories such as EBI [78] and NCBI [79] already provide semantic techniques in their search systems. As the timelines in Subsection “Results and Met- rics” reveal, Dryad and Figshare also started using semantic formats for metadata but so far do not offer a semantic search. To the best of our knowledge, there are few semantic search approaches focusing on dataset retrieval in biodiversity research [80, 81]. In addition, there are only very few systems that consider semantically related terms or categories in scoring [77]. Hence, as semantic search approaches are still under research in computer science, integration in current data portals has only just started. PLOS ONE Dataset search in biodiversity research PLOS ONE | https://doi.org/10.1371/journal.pone.0246099 March 24, 2021 31 / 36