Visualizing and ﬁltering social 8es in SocialLearn by topic and type Visualising Social Learning in the SocialLearn Environment. Bieke Schreurs and Maarten de Laat (Open University, The Netherlands), Chris Teplovs (ProblemshiB Inc. and University of Windsor), Rebecca Ferguson and Simon Buckingham Shum (Open University UK), SoLAR Storm webinar, Open University UK. hGp://bit.ly/LearningAnaly8csOU
Disposi8onal Learning Analy8cs for C21/LLL Ques8oning and Diﬀerent social challenging network paGerns as behaviours as proxies for Learning proxies for CriKcal RelaKonships Curiosity Cross-‐contextual Persevering behaviours as proxies behaviours as proxies for Meaning Making for Resilience Shaofu Huang: Prototyping Learning Power Modelling in SocialLearnhttp://www.open.ac.uk/blogs/SocialLearnResearch/2012/06/20/social-learning-analytics-symposium
Learning analytics focus group projectsPerformance Support Data Assurance & Transparency Agency BI strategycurrently looking at feedback - text analysis of existing feedback from training, develop examples - ratings & recommendations for procedures – useful, accurate, up-to-dateevaluation report January 2013“Well done youve used really nice language in that email” “you seem to have been working on this report for 7 years” “8 out of 10 assessors said they prefer…”
Exponen8al Random Graph Models A dFirst Experiments with Mutuality a m Transi8vity C o o p Homophily er (JI S C
e.g. JISC and CETIS Teams • Showing our colours? ● Main eﬀect ● Homophily ● Mixing edges + sender(base=c(-‐4,-‐21,-‐29,-‐31)) + receiver(base=c(-‐14,-‐19,-‐23,-‐28)) + nodematch("team", diﬀ=TRUE, keep=c(1,3,4)) + mutual All images and text CC-‐By: Adam Cooper, 2012
Exploring Learning AMore paossibili8es of naly8cs the wareness Enthusiasm! Lessons Learned Vak voor Vak User Needs UvAnaly-‐ 8cs PinPoint MAIS ProF Curri Analy8cs M hGp://youtu.be/Xs3MsGPVivg Seven tangible examples to refer to Community of various Areas of work to be experts done…
Unlikely Very unlikely Neither Likely or Unlikely Very unlikely 7% 2% Unlikely 0% 2% Neither Likely or 5% Unlikely 11% Before Very Likely How likely AEer 32% are you to Likely 29% use this Very Likely feedback? 64% Likely 48% Clearer sense of where they sit in comparison to their cohort which mo8vates them to want to do more to improve Shining aGen8on to important areas that they tend to neglect Mo8va8ng high achieving students Seeing a bigger picture For some this is emo8onally challenging and sensi8ve but for others it’s not
Social learning analy-cs: discourse Challenge: Locate the exploratory dialogue Manual analysis identifies indicatorsCategory Indicator Challenge But if, have to respond, my view Cri8que However, I’m not sure, maybe Discussion of resources Have you read, more links Evalua8on Good example, good point Explana8on Means that, our goals Explicit reasoning Next step, relates to, that’s why Jus8ﬁca8on I mean, we learned, we observed Reﬂec8on of perspec8ves of others Agree, here is another, take your point 23
Self-‐training framework for automa-c exploratory discourse detec-on • Framework uses cue phrases to make use of discourse features for classiﬁca8on • Uses a k-‐nearest neighbours instance selec8on approach to draw on topical features
1. Uniview -‐ Oracle-‐based data warehouse / BI repor8ng since 2009 2. Used R randomForest for learning tech review & NSS analysis since 2010 3. Consistent student sa8sfac8on data collec8on, 10,770 respondents 2011 4. Star8ng major Analy8cs project (SQL Server, SSAS, SSRS, SP2010) A League table rankings Marke)ng & Recruitment Reputa)on Processes C B Learning, Teaching, Assessment Student Intake Student Reten)on & Personal Development (Aspira)ons, A8tude Success & Processes, Facili)es & Abili)es) Sa)sfac)on & Resources Resource alloca)on All Year Numbers A Recruit to target B Improve sa8sfac8on, reten8on & success C Inform decision-‐makers Prof Mark Stubbs | Head of Learning & Research Tech | email@example.com | twiGer.com/thestubbs
students Data sources VLE TMA Demographic Other.. Who is struggling? RETAIN predic8ve models Why are they Dashboard visualisa8ons struggling?
BUILDING THE PREDICTIVE MODELS Developed and tested on 3 historic data sets Compared: decision trees and SVM’s. Compared: VLE only, TMA and combined MAIN FINDINGS • No overall clicking measure correlated with pass/fail: focus on change in student behaviour instead • High precision can be achieved in predic8ng both performance drop and ﬁnal outcome (pass/fail) for all 3 modules, using combined VLE and TMA data • Demographic data can improve performance, but in early stages the VLE ac8vity is the most informa8ve data source. • Successfully applied 2010 model to 2011 data. Even some success across modules.
Labs www.triballabs.net Learning Analy8cs R&D Project • Partnership with a university to develop a Learning Analy8cs PoC: – Predic8ve model which can predict student success – Combine data from mul8ple administra8ve and ac8vity sources – Test how support staﬀ can interact with the model and correctly interpret predic8ons – Bring together visualisa8on and ac8on – onen a missing element @chrisaballard
Labs www.triballabs.net Mapping Success Factors Academic Integra-on Engagement Circumstances Grades VLE Ac8vity Social Background Library Ac8vity Proximity Finance Social Integra-on Prepara-on for HE Forum interac8on Demographics Qualiﬁca8ons @chrisaballard