metadata-automation

Super pleased to announce that we are kicking off a project to explore ways to uplift and improve the quality of metadata in the social and population sciences using machine learning. The project website is at metadata-automation.org and we'll be tweeting from here.

Congratulations to my PhD student Zeqiang Wang & @yuqiw_ for their work on #longitudinal semantic shifts getting accepted @emnlpmeeting Findings. Co-supervised w/ WeiWang, @nishanthsastry & @spuddybike @CLOSER_UK. @MetadataUplift

Pleased to be in Manchester for @NCRMUK #MethodsCon where Jon Johnson @spuddybike is presenting our @ESRC-funded #FutureDataServices MetacurateML @MetadataUplift project. @CLOSER_UK @ScotCen

🆕Standardised scales (SS) added to Discovery, our research tool for longitudinal population studies! This update will make it easier to determine if questions belong to a SS & identify where they are used across our studies & waves of data collection. closer.ac.uk/news-opinion/n…

Job Alert! We are looking for a PDRA in #NLP and #ML, as part of our @ESRC-funded #FutureDataServices grant, at @ComputingSurrey. The successful candidates will collaborate with @CLOSER_UK @ucl, @UKDataService and @ScotCen. Apply by: 22 July 2024 jobs.surrey.ac.uk/031024

Excited to kick-off our #METACURATEML project, with @CLOSER_UK @ucl, @UKDataService & @ScotCen to generate #ML-driven #metadata and #data solutions which are #FAIR ready, using #GrowingupinScotland study as a pilot dataset. Grateful for funding from @ESRC #FutureDataService.

But if the advance from chatbots to ChatGPT is in automating the answers, can we automate the questions as well? Can we automate the prompt engineering? Excellent points on creativity, enquiry and why it ain’t intelligence ben-evans.com/benedictevans/…

@supdey @ComputingSurrey @spuddybike @CLOSER_UK @NllpWorkshop @emnlpmeeting @ucl_arc @DiRAC_HPC

Our #NLProc paper "Privacy Pitfalls of Online Service Terms and Conditions: a Hybrid Approach for Classification and Summarization" w/ Emilia Lukose @ComputingSurrey and @spuddybike @CLOSER_UK, UCL has been accepted @NllpWorkshop @emnlpmeeting! @MetadataUplift 🧵

Our final report of the @DiRAC_HPC funded Activity on 'Understanding the multiple dimensions of prediction of concepts in social and biomedical science questionnaires', jointly led by @spuddybike, is now out in dirac.ac.uk/wp-content/upl… @MetadataUplift

@iassistdata @SupDe @ESRC @DiRAC_HPC We used @CLOSER_UK Discovery as our training data to develop an ML agnostic pipeline to explore different models and approaches.

Our first paper on metadata uplift using machine learning is out in @iassistdata doi.org/10.29173/iq1023 led by @supde. Funded by @ESRC and @DiRAC_HPC.

Encouraging feedback from the DDI-L community on our @ESRC funded ML pipeline approach for data extraction from longitudinal questionnaires. See slides from the talk here: doi.org/10.5281/zenodo… #eddi2021 @MetadataUplift

#JobFairy UCL have a range of permanent posts for RSEs. It’s a really great team (we work with them!). More info over at lnkd.in/dnJqHKEe

I'm really glad folks are looking into this and grateful to @strwbilly for this reporting highlighting the negative impacts of #AIhype technologyreview.com/2021/07/30/103…

@jamespjh Social science @ucl is reaching out to CS at metadata-automation.org

DDI-CDI is a @DDIAlliance candidate standard that is aimed at allowing cross domain integration. The paper raises some useful ways in which standards can work together for machine actionable data management

#longitudinal #emoji

@Letxuga007 Of course one of big problems of location based assignment of measures is how heterogeneous many areas are, especially in urban areas even if you have my postcode it will cover everything from <10 to >90 income bracket.

Really useful review of deep learning text classification methods dl.acm.org/doi/pdf/10.114…

@HarryAEvans @Letxuga007 Is eight questions too much to get a robust measure? #9" target="_blank" rel="nofollow noopener">webarchive.nationalarchives.gov.uk/20160106042025…

@Letxuga007 Suspect in addition to other notes it's related to the cost of obtaining "social class" - requires surveys and is an individual level measure, which then requires linking with good individual level data