Social Sciences, Social Data and the Semantic Web (S3W)

Recent years have seen phenomenal growth in quantity and range of digital data that might be used for social research. The ESRC has already invested in harnessing administrative and business data as well as 'new and emerging forms of data' (e.g. social media and sensor data) for the social sciences. Now a new opportunity arises. 'Semantic linked data' (SLD) offers a new method for structuring and organizing digital data, which promises to have a profound effect on research capacity for data linkage and analysis across multiple, heterogeneous sources, at hitherto unimaginable speed and scale. Indeed, within the Computer Sciences, the proponents of SLD argue that if data are published following shared standards and protocols the Web will be transformed from a library of documents into a single linked data base, described as the 'semantic web'.

The value of data linkage is already well established in the social sciences, but existing methods are labour intensive, involve the retrospective matching of records, across small numbers of data sets, and are done to address particular pre-determined questions with the linkage made for that specific purpose. In contrast, SLD techniques focus on the prospective production of data to allow the on-going matching and accumulation of information about people, places, businesses, artefacts and even conceptual categories and like 'race' or 'class' to be drawn together, however the subsequent user determines, at the scale of the World Wide Web (Halford, Pope and Weal 2012).

However, whilst SLD offers great promise to the social sciences there is - to date - negligible use of SLD by social scientists. The agenda for SLD is being driven by computer scientists and demonstrations are based on relatively straightforward examples such as transport timetables or estates data. Whilst these work well technically, they offer no substantial investigation of how appropriate the techniques might be in addressing more complex social science questions. At a time of financial constraint, when funding for major new data collection is uncertain, it is essential that we explore these opportunities.

The research proposed here will be the detailed investigation into if and how SLD might be harnessed for social science research. To achieve this we have drawn together a strong team of social and computational scientists, with a well-established track record of collaboration. This team will be supported by an outstanding Advisory Group of experts, who have already agreed to participate in this project (see Impact Summary below).

We will explore three research questions:

(i) What are the implications of using SLD methods to describe social data?
(ii) What does SLD contribute to our capacity to understand health inequalities across the life course?
(iii) What are the implications of SLD for data archiving and re-use?

To answer these questions we will:

(i) Carry out a detailed study of the processes involved in converting existing data into SLD. This will be undertaken by the research team, with the participation of experts from our Advisory Group. Specifically, we will work with the English Longitudinal Survey of Ageing and the Great British Class Survey and other related data already in the 'linked data cloud'
(ii) Develop a 'demonstrator' of SLD (using the data sets developed at (i) above) to examine the specific question of health inequalities across the life-course
(iii) Collaborate with the UK Data Service and the GESIS-Leibniz Institute in Germany (which provides a similar data infrastructure to UKDS) to explore the opportunities for data archiving.

In this way, we seek to engage social science in the ongoing development of SLD and the emerging Semantic Web; and to explore the implications of SLD for building next generation data infrastructures in the social sciences.

Three specific non-academic user groups will benefit from this research.

i) Data archive professionals and policy makers: the value of secondary data analysis is now increasingly recognised, both in terms of securing maximum returns from initial financial investment and in terms of generating knowledge and understanding. This has been a policy priority for Research Councils UK and for the UK data archives, now consolidated as the UK Data Service. The rapid development of open and semantically linked data (SLD) promise a step-change in the find-ability and reusability of existing data, allowing analysis across multiple data sets using computational tools to permit a step-change in the scale and range of data linking that is possible. To date, there has been very little application of SLD techniques to social science data. Meanwhile the development of these techniques in the computational sciences continues apace but with little reference to the social sciences, or the particular challenges that our data may present for these techniques. This project will harness current computational advances to the social sciences, and work closely with our key members of our Advisory Group to ensure that these reach non-academic users directly. In particular we will work with the UK Data Service, who have time costed into this bid, to explore how SLD might support their current shift towards 'data as a platform'. The Office for National Statistics will also be closely involved, as the project offers insights to several of their teams including the Big Data team, ONS Geography, and the Health Inequalities team as well as to those concerned with data infrastructure.

ii) Semantic Web developers: seeking to further the development of tools and inference techniques to encompass a broader range of quantitative and qualitative datasets. By having a better understanding of the way open data providers produce data the community will be able to improve tool support for the creation, publishing and analysis of linked data. This study will provide a case study in modelling qualitative data, such as interview transcripts, which is widely ignored by the Semantic Web community currently. The longitudinal and aggregate nature of the datasets being considered in these studies presents a particular challenge in moving beyond the current consensus of identifier use on the Semantic Web.

Both groups have been involved in the preparation of this research proposal and will be integrally involved in the project as it unfolds. This will provide firm grounding for impact with these users, on which we can build to develop wider policy, skills and public impact. The mechanisms that we will use to achieve this are described in the attached 'Pathways to Impact' document.