Transcript Rob_Procter
Text Mining for e-Social Science Rob Procter National Centre for e-Social Science [email protected] http://www.ncess.ac.uk 21st March, 2005 NaCTeM 1 Overview National Centre for e-Social Science Why do we need e-Social Science? Applications of text mining 21st March, 2005 NaCTeM 2 NCeSS goals Explore opportunities for take up of eResearch within the social sciences: – Substantive social science research problems – Enhance existing research methods – Facilitate new research methods Social shaping of e-Research: – Socio-technical factors in the design, uptake and use of eResearch infrastructures and tools – Implications for research practice and knowledge production – Policy and socio-economic impacts 21st March, 2005 NaCTeM 3 NCeSS structure Co-ordinating hub at 21st March, 2005 NaCTeM Manchester Four research ‘nodes’ already commissioned, more to be announced later this year ‘Small’ research projects 4 Why do we need e-Social Science? As social scientists become concerned with addressing complex, multi-faceted research problems, they require the creation and use of increasingly multi-level, multi-textured data resources e-Social Science offers improved methods and tools for data description, discovery and curation e-Social Science offers new ways for the study of complex social phenomena 21st March, 2005 NaCTeM 5 Text mining for e-Social Science e-Social Science is often associated with quantitative research methods: – Statistical analysis, modelling and simulation e-Social Science applications in qualitative research have received less attention: – New tools are needed to address challenges of analysing increasing volumes of qualitative data 21st March, 2005 NaCTeM 6 Applications of text mining I Content analysis of textual data is an established method in qualitative social science research There are many variations on but its essence involves searches for keywords, their cooccurrences and interpretation in context Simple computer-based tools (QDAS) have been used for some time to support content analysis New text mining techniques promise improvement in power and sophistication 21st March, 2005 NaCTeM 7 Content analysis example Systematic reviews of the scientific literature are an important tool for researchers and policy makers Finding relevant studies calls for the search of very large literature databases Records are not well coded, so this usually results in a very large number of titles and abstracts (often > 10,000) being retrieved which need to be sorted manually Text mining techniques such auto-summarising, annotating etc. can help to make systematic reviews easier to conduct and improve their quality 21st March, 2005 NaCTeM 8 Applications of text mining II The solution of complex, multifaceted social science problems calls for new ways of doing research Quantitative and qualitative research methods reflect a fundamental divide in social sciences Researchers could benefit from tools which would enable them to combine these different methodological traditions 21st March, 2005 NaCTeM 9 ‘Mixed methods’ example FINGrid is a demonstrator for analysing financial information in form of quantitative data (time series) and qualitative data (financial/political news) FINGrid works by text mining financial news (Reuters news feed) for ‘market sentiment’ and then attempts to correlate this data with time series data of market price movements The ultimate aim is to understand better the relationship between price movements and market sentiment Prof K Ahmad, University of Surrey 21st March, 2005 NaCTeM 10 21st March, 2005 NaCTeM 11 Applications of text mining III Data about social phenomena is being generated on increasing scale as by-product of the everyday activities of social actors: – Bulletin boards, weblogs, news feeds, etc. These resources could provide researchers a much richer picture of social phenomena than is available via more conventional data gathering techniques But researchers lack adequate tools to exploit them 21st March, 2005 NaCTeM 12 A Virtual Observatory for the Social Sciences? Bulletin boards, weblogs, news feeds, etc. raise different requirements for description and discovery from conventional social science datasets: – They are dynamic, their content growing and their relevance for researchers changing continuously Making such data sources useful for research would need text mining tools for: – – – – Filtering Annotating Summarising Information extraction such as sentiment analysis Using text mining in these ways, we could create a ‘Virtual Observatory for the Social Sciences’ 21st March, 2005 NaCTeM 13 Getting involved in NCeSS Small projects programme is open until July 31st First International Conference on e-Social Science 22nd-24th June, Manchester http://www.ncess.ac.uk/events/conference/index.shtml 21st March, 2005 NaCTeM 14