“You looked at The Complete Essays by Montaigne; you might also consider The Renaissance in Europe: A Reader edited by Whitlock.” Most of us are familiar with Amazon’s gently pushy way of suggesting further purchases. If you’re a music fan, you may have tried “scrobbling” [1] each song you listen to into the massive Last.fm database of listener behaviour. In return for this gift of your data, you get to explore the habits of others who share some of your tastes, and you get a series of recommendations for other music you might enjoy.

If it works for retail and leisure, might this same approach also be applicable for libraries and learning? In introducing this workshop, organised by the Association for Learning Technology (ALT) with the JISC-funded TILE Project (Towards Implementation of Library 2.0 and the e-Framework), Ken Chad noted that, amid a rash of “crowd-sourcing” ventures, the higher education sector has done relatively little so far to exploit data from learners. Yet valuable data about learner behaviour certainly exists in library systems. So could we, Ken asked, be approaching a tipping point where data, methods and technologies combine to revolutionise the learning potential of library services? Or, as the workshop title put it, are we sitting on a goldmine?

If we could map the paths that learners take between sets of resources, we would get new ways of viewing and understanding the links between the items that make up a library. Of course, the road well travelled is not always the most efficient or creative way of learning – and certainly not the most original. But by bringing these patterns into the open, we could enable a new level of reflection, judgement and guidance about how to get the most out of a library.

The TILE workshop explored practical (could we?) and policy (should we?) issues which relate to pursuing some aspects of this broader approach. It explored the area at the intersection between library and learning technology, virtual learning environments (VLEs) and library systems. Navigating it successfully will require careful judgements that balance the virtues of “top down” disciplines of traditional information management with the “bottom up” folk wisdom of Web 2.0 techniques.

Scale and ambition
The model that the TILE project has developed recognises two contrasting perspectives on the library domain: a narrow definition concerned just with the management of a library’s own assets; and a wider definition of the total set of processes required to help people interact with learning “stuff” (content, metadata, reading lists, profiles). While stressing the strengths of each perspective, David Kay, the TILE Project Manager, left little doubt that the ambitions he harboured were on a larger “web-scale” stage.

Along with Joy Palmer (Mimas’s Manager of Library and Archival Services)in a later session, David cited Lorcan Dempsey’s call for a model of library use suited to the web environment. “We need to think about library services in the context of the full web of user experience,” writes Dempsey (2005). The size, and more particularly the reach, of services like Google and Amazon gives them a gravitational pull that draws learners towards them. As long as library resources remain fragmented, they will never exert the same pull. David told us how California State University’s library holdings now include user-generated tags. With over a million students spread across the university’s many sites, these tags constitute a genuine web-scale service. However, he was at pains to point out, examples such as Amazon’s analysis of users’ clickstreams, show that you do not require users to generate content explicitly in order to capture their context; you can get value from the choices that are implicit in the tracks they leave.

Joy Palmer envisaged a range of personalisation tools and APIs (application programming interfaces) that might build on library usage data. The Copac catalogue – funded by JISC, with 54 contributing libraries – that Joy works on will soon include a “my bibliography” feature, with a feed to allow it to be shared; additional features might include automated recommendations and tagging facilities. Again, services like Amazon and Last.fm provide further examples through the way they enable third parties to build services on top of their open APIs. Students could conceivably come to share reading lists on their social network profiles the same way they share music playlists now.

Implementation and examples

David Kay outlined options for building services on top of learners’ library data: as well as providing APIs; these include building applications and liberating the data. The workshop touched on several examples of these.

Mendeley (www.mendeley.com) is a free social software application for managing and sharing research papers. By aggregating the data from its users, it is also a Web 2.0 social network for discovering research trends and connecting to like-minded academics. At present it does not draw any data from academic institutions, but its growth plans unquestionably overlap with the wider definition of the library domain.

Calling in by video link, Dave Pattern, Library Systems Manager at the University of Huddersfield, told how his university had experimented with mining their borrowing data. They started by generating borrowing suggestions for students to see if these would be useful. They were also able to generate keyword suggestions – a potentially very valuable link between the vocabularies used by librarians and by learners – and also to identify keywords that people enter most frequently that get zero results (e.g. “newspapermen”, “ligament”).

On the day of the workshop, Dave Pattern published usage data for two million transactions on 80,000 library titles, broken down by year, academic school, and academic course (with relevant UCAS codes), under a common data licence at library.hud.ac.uk/data/usagedata/. The rights issues with this data are complex, but Dave was able to side-step the complexities by using a very open licence, which also served the purpose of allowing the data be distributed, shared and used as widely as possible. Thus he hopes to provide an open resource that anyone can use as one of the foundations of further innovation.

Joy Palmer explained how Mimas is interested in exploring the potential of the Huddersfield work in a national context with services like Copac. Presently Copac has data on two million searches per month, and 800,000 user sessions – but this does not tell them a lot about learner behaviour. While Joy could see a clear need for deep log analysis of attention data, circulation data provides an instant snapshot of learner behaviour that could be profoundly useful for system-wide services such as Copac. As well as supporting developments around adaptive personalisation such as recommender functions, it could provide rich opportunities for text-mining and improved search.

Mark van Harmelen from the TILE Project, described a web-service-based architecture to realise the project’s approach. The architecture is in two parts: to gather and aggregate anonymised library user behaviour data from educational institutions, and to search a catalogue that is enhanced with this data, thus providing searches that are enhanced with recommended results. This architecture can be seen in Figure 1.

Figure 1: Web-service-based architecture

Mark mentioned promising results emerging from another JISC-funded project in which he is involved (the EIE Project) for search performance using the Lucene search engine. He also demonstrated how library records can be treated as social objects, in the same way as Flickr treats photographs as social objects that enable communication and discussion between users, and how those objects can be integrated into a learning environment.

Questions and open issues
Mark Toole, Director of Information Services at the University of Stirling, raised the concern that data samples from different places might not “fit together” – meaning that any analysis would be comparing apples and pears – because different institutions organise their modules in very different ways. Does this mean that their data cannot comfortably coexist in the same set? Flickr has shown how it is possible to tease out clusters of semantically different data – distinguishing, for example, “jaguar” the animal, the car and the Apple operating system – by comparing the contextual tags. So all may not be lost.

Clearly capturing a learner’s context adds another dimension to the data. Are you borrowing this book about Napoleon to study the psychology of dictatorship or French history? Often there may be a trade-off between collecting extra data and keeping the collection process – for learners and library staff alike – as simple and integrated into the workflow as possible.

Library specialists will have to get to grips with some of the standard problems that affect recommendations based on “collaborative filtering” of data. One such is the “Harry Potter Effect”: because people with many different interests and dispositions have read a Harry Potter book, the automated recommendations have a propensity to tell you, “People who borrowed books on Renaissance literature also borrowed Harry Potter.” The statement is statistically correct, but not finely-tuned or useful. Another is the “cold start problem”, which affects new additions to a library: as no one has ever borrowed them at this point, they will not be recommended by the system unless there is some editorial intervention in the data links.

Joy Palmer reminded us of the challenges of semantic context and “ontological drift” when user-generated commentary on contentious subjects becomes too rich to be easily assimilated – for example, consider the multiple sparring entries relating to the state of Israel on Wikipedia. She questioned whether the library OPAC (Online Public Access Catalogue) was too generic a system to support contextually and academically meaningful personalisation, and this point was carried over to the break-out discussions about whether users would be motivated to contribute content to institutional OPACs.

User data always brings with it concerns about privacy, rights, and ownership of data, together with the relative merits of opt-in versus opt-out schemes. Where data is in aggregated and anonymous form, there should in principle be no risk to personal data, but security procedures need to be monitored carefully to ensure that anonymity cannot be compromised.

Conclusions
Perhaps unsurprisingly, the workshop participants were optimistic and eager to roll up their sleeves and experiment with pilot projects (any sceptics in the library profession probably stayed away). One comment was that, “The idea is so different from what we do now that we just have to try it, boldly.” And there was consensus there is not just one opportunity in this area, but many.

The application of Web 2.0 ideas to libraries in education is not a matter of trying to ape the features of Facebook and MySpace, and it is emphatically not about stopping people from using existing social networks, blogs or other services. Web 2.0 is not like that; it is more likely to involve, say, the provision of library profile “widgets” that learners can embed in their blogs – and their coursework.

The complexity when applying these ideas in education stems from the need to retain some kinds of value judgement that do not apply to Amazon’s retailing or Last.fm’s music discovery. Perhaps analysis will reveal which resources the successful students use and how these compare with those the poor students use. Does students’ selection of resources influence their capability, or vice-versa? And many library professionals will be wary of the herd tendency in basing recommendations on behaviour of other students. Just because learners do not follow the official reading lists to the letter does not mean that they should not be encouraged to do so. Professionals will need to bring all their experience to bear in order to judge how to moderate, and when to intervene in, the emergent behaviours of learners as captured by usage data.

But one way or another, through HE institutions or entrepreneurial social software such as Mendeley, learners will access a whole new perspective on the universe of resources – including which of them are used most often, most highly rated and which seem to connect and work best together. The best way to influence this new approach is to be part of making it happen.

The TILE Workshop “Sitting on a gold mine” was organised by the Association for Learning Technology (ALT) and held at JISC's offices in London on 12 December 2008. Further information is available from www.sero.co.uk/jisc-tile.html

David Jennings is author of Net, Blogs and Rock’n’Roll: How Digital Discovery Works and What it Means for Consumers, Creators and Culture (Nicholas Brealey Publishing, 2007). David’s report was assisted by contributions from Joy Palmer (Manager of Library and Archival Services at Mimas), Phil Barker (Learning Technology Adviser at Heriot Watt’s Institute for Computer Based Learning) and Mark van Harmelen.

David Jennings
DJ Alchemi Ltd.

References
Dempsey, L. (2005) The User Interface that Isn’t. Lorcan Demspey’s Weblog, orweblog.oclc.org/archives/000667.html [accessed 22.12.2008]