Project

Nowadays there is a growing trend in many scientific disciplines to support researchers by providing enhanced information access through linking of publications and underlying datasets. Open Science encourages scientific practices in which all research data is interlinked and contextualized to enhance reproducibility and reusability of research results. Ideally, publications that report on a result of an empirical study should thus contain a direct link to the cited dataset and lead the reader directly to the research data that underlies the publication.

However, in practice, standards for referencing between primary text and the cited data and its variables are often missing. A recent user study conducted by GESIS reveals that researchers would considerably benefit from increased linking and semantic annotation of scientific publications. In addition, researchers also demand that data citations should include information at the right granular level of the cited data, thus facilitating the identification and verification of the part of data that actually supports a specific claim. Improving access to scientific publications along the FAIR best practices also requires semantic indexing of texts with salient entities and specific variables that make up the focus of the study – requirements which are rarely addressed today.

The key vision behind VADIS is to allow for searching und using survey variables in context and thereby help to increase the reproducibility of research results. We achieve this by combining text mining techniques and semantic web technologies that identify and exploit links between publications, their topics, and the specific variables that are covered in the surveys. These semantic links in scientific texts build the basis for the development of applications to give users better access to scientific literature such as passage search, summarization, and information retrieval.

To achieve this, we will analyze and link variables in context by identifying references to survey variables within the full text of research literature, creating semantic links based on these references and making the resulting data available as Linked Open Data. Next, we will develop data-driven profiles of survey variable on the basis of both context-independent and context-dependent metrics. Finally, we will improve the access to survey and literature by providing information on variables from surveys, the developed metrics as well as textual summaries of linked literature. As a result of this, our project will be able to provide improved access to research literature in the social sciences based on the seamless integration within existing infrastructures. To quantify the effectiveness of our framework we design several use case scenarios for a target group of researchers that will be implemented as interfaces for exploration and research. The improvements on information access from experts will be thoroughly investigated in a user study.

• A Shared Task for Survey Variable Identification in Social Science Publications is carried out at the 3rd Workshop on Scholarly Document Processing (sdproc.org) to compare different approaches on the same evaluation data set. It will take place at the https://coling2022.org/ in Korea in October 2022
• More information is available at VAriable Detection, Interlinking and Summarization project | vadis-project.github.io