Transcript Slide 1
First year History students searching for Napoleon T I L E Also borrowed … They downloaded … They rated this resource as … They also recommended ... D David Kay Sero Consulting [email protected] Sitting on a goldmine - the value of attention and activity data Slides & ideas stolen from • David Jennings, author ‘Net blogs & rock ’n’ roll’ • Dave Pattern, Huddersfield • Mark van Harmelen, Hedtek • Paul Walk, UKOLN • Caroline Williams, Mimas 2 Supermarkets “Supermarkets gain valuable insights into user behaviour by data mining purchases and uncovering usage trends. Further insights are gained by analysing purchasing histories, facilitated by the use of store loyalty cards.” Dave Pattern, University of Huddersfield – December 2008 Libraries? “Libraries could gain valuable insights into user behaviour by data mining borrowing and uncovering usage trends. Further insights are gained by analysing borrowing histories, facilitated by the use of library cards.” Dave Pattern, University of Huddersfield – December 2008 Distributed … Content & finding aids anywhere & any type Concentration of … Context data Catalysing contribution What’s recommended in the VLE? Did anyone highly rate this textbook? Across … An Institution A Consortium A national system Global communities What’s did last year’s students download most? What’s the economics take on this topic? What do undergraduates elsewhere read? Resources other people are using (CERLIM research for MOSAIC) • 90% of students said they would like to be able to find out what other people are using Finding out what resources other people are using 50 40 30 20 10 0 45 42 32 People taking the same People taking a similar course as you at course as you: another university: Lecturers teaching on your course: Finding out what resources other people are using 30 25 20 15 10 5 0 27 21 14 Lecturers teaching a similar course at another university: People taking the same People taking a similar course as you in a course at a different different year of study: university and a different year of study: California State University 2008 What can be done in the library? Dave Pattern - Huddersfield • Raw popularity enables recommendation • Lending paths enable sequential recommendation What people borrow next • Dewey matches enable recommendation from new book lists at course level based on Dewey-based patterns of borrowing • Search histories yield Search clouds, Search combinations that work (law try criminal law, civil law), Search terms that ‘discover’ books suggestions based on circ data “people who borrowed this…” getting personal! suggestions for what to borrow next the impact on borrowing average number of books borrowed 17 16 15 14 13 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 average number of books borrowed per active borrower per calendar year (2009 predicted) usage data for staff loans per academic school Motivation Attention Contribution Work: PLF – 18 May, 2008 - 6 versions Activity (Downloads Served) & Contributions V1 – 1192 DL – 30 comments V2 – 1389 DL V3 – 4517 DL – 5 reviews V4 – 1338 DL V5 – 285 DL – 1 review V6 – 2702 DL – 1 review ALL – 11423 DL – 37 contributions 15 Aggregate niche to deliver network ? Motivation Attention Motivation Attention Motivation Attention Contribution Contribution Contribution Motivation Attention Motivation Attention Motivation Attention Contribution Contribution Contribution Is it desirable and feasible to combine the outputs of niche communities into a large network? ‘… and the more we track, the better we can adapt our service without your intervention.’ The Institution Teaching & Learning The Library Research The Network Who do you serve? Niche / specialist networks • Automated recommendation systems may be of interest at undergraduate level • But, in academia, beyond undergraduate it’s long tail all the way! • Small networks based on people actually knowing each other • Networks provide the opportunity for other people, known to me, to intervene on my behalf, thus reducing ‘filter-failure’ 19 Some Questions • Does activity data need to be aggregated above the institutional level to achieve web scale and deliver network effect? • Amazon tells you that ‘people who did this also did that’. Can academic libraries offer something more significant (‘people LIKE YOU who did this also did that’) because they know the user’s context (typically their course and institution)? • Precision in such as metadata and even citation is subject to personal judgements and motivations. Are these less reliable than the pointers derived from a mass of contextualised activity data? • As proxies for real activity, are lists – formal reading lists, informal student lists – a form of attention data which can be highly weighted? 15-07-2010 Workshop Some Next Steps • Joint work on standards and specs with publishers and systems vendors • Solutions for the key requirement of context • Reliable subject (course / module) identifiers, which may or may not be UCAS or JACS codes • Look beyond just books and just use data