Last Thursday members from the D2D team met up with people from the DPIE 1 and DPIE 2 projects, as well as Paul Walk (representing the IE demonstrator project). The aim was to talk ‘personalisation’ developments for the JISC IE. It’s impossible to cover the entire scope of discussion here (we were at it most of the day). As you might predict, it was a day of heated but engaging debate around a topic that is technically and socially complex. As we think about the strategic future of services and cross-service development, and there are serious questions marks over which direction we’re headed in terms of personalisation (and, of course, if it’s even possible to talk of ‘a’ direction).
The key practical aim of the meeting was to share the personalisation aspects of D2D project work, and also to discuss the recommendations of the two DPIE reports. The D2D work includes some development of personalisation components for Copac, components we are referring to cautiously as ‘lightweight’ for now. One way in which we plan to ‘personalise’ the service for users is by offering a ‘My Local Library’ cross-search, achieved (we hope — we’re very much in early phases here) via a combination of authentication and IP recognition used to identify users’ geographical location, and then a cross-search of local institutional library holdings data via Z39.50 targets.
In addition, by next the middle of next year, Copac users will be able to save marked lists of records and export them into other 2.0 environments via an atom feed (I’ll let Ashley write the more technical post on that development). Further down the line (i.e. beyond the next six months) we are interested in providing tools for users to annotate, bookmark and tag these records, but we also want to make sure that any such developments are not made in isolation and are ‘Copac’-centric — there’s a lot to explore here, obviously.
In and of themselves, these developments are not especially complex — the latter is an example of personalisation via ‘customisation’ (to use JISC definitions) where users explicitly customise content according to their own preferences. What I am especially interested in, however, is how saved lists (‘My Bibliography?’) could be used to potentially support adaptive personalisation (this is what Max Hammond, co-author of the DPIE 2 report wryly referred to at the meeting as ‘spooky’).
Dave Pattern’s experiment with using circulation data to ‘recommend’ items to University of Huddersfield library users is well known, and I hope the first step towards some potentially very interesting UK developments. On this end, we’re interested in knowing if there is anything similar to be gleaned from saved personal lists — ‘users with this item in their saved lists also have…’ (or something along those lines). This terrain is very much untested, and one of the critical issues, of course, is uptake. Amazon’s recommender function is effective due to the sheer number of users (effective *some* of the time, that is — we all have ‘off-base’ Amazon recommendations stories to tell, I admit). And this is just one small example of how adaptive personalisation of a service like Copac (or other JISC IE services) might work — there are also opportunities around capturing attention data, for instance.
The DPIE 2 report urges extreme caution in this regard. It raises some very pointed questions about how JISC and its services should approach adaptive personalisation. Too often, the authors warn, ‘personalisation’ is established as a specific goal, with the assumption that ‘personalisation’ is intrinsically valuable. In this context, change is technology rather than user driven, which is fine for experimental and demonstrator work, but high-risk for established services with a strong likelihood for failure. They question how helpful definitions of Personalisation put forth by JISC are in carrying forward a development agenda (Customisation; Adaptive Personalisation based on Data held elsewhere (APOD); Adaptive Personalisation based on User Activity (APUA). This definition “provides a mix of concepts from data capture to functionality, rather than setting out the logical link between a source of data about a user, a model of that user, and an approach to providing the user with a personal service” (17). Also missing is a robust benefits mapping process — “there is little analysis of the benefits of personalisation, beyond an assertion that it improves the user experience” (20). The report concludes:
Complex developments of â€œpersonalisation servicesâ€ and similar should not be a current
priority for JISC. It seems unlikely that an external personalisation service will be able to
generate a user model which is detailed enough to be of genuine use in personalising
content; user preferences are probably not broadly applicable beyond the specific resource
for which they are set, and user behaviour is difficult to understand without deep
understanding of the resource being used. Attempting to develop user models which are
sufficiently generic to be of use to several services, but sufficiently detailed to facilitate useful
functionality is likely to be a challenging undertaking, with a high risk of failure. (26)
These are somewhat sobering thoughts, especially in a climate of personalisation and 2.0 fervour, but overall the report is useful in considering how to tread the next steps in development activity. Key for us is this issue of the user model — can we (Copac? SUNCAT? JISC?) develop one that is likely to be of use to several services? My hunch right now is ‘no.’ We know very little concrete about researchers’ behaviour and how they might benefit from such tools (interestingly, both DPIE 2 reports focused on benefits for undergraduate students, when most of the services in question are largely used by researchers). About humanities researchers, we know even less (much of the interesting work around online ‘Collaboratories’ centres on the STEM disciplines). Apparently JISC is about to commission some investigative work with researcher-users, and here at Mimas a team is about to undertake some focus group work with humanities researchers to determine how personalisation tools for services like Copac, Intute and the Archives Hub could (or could not) deliver specific benefits to their work. I’m sure this research will prove very useful.
We’re urged to ‘proceed with caution,’ but we proceed nonetheless. At Copac we’re taking a long hard look at what a personalised service might look like, and accepted that some risk-taking is likely forecast for the future. I’m very interested to know other’s opinions on a possible recommender function for Copac — at what level could such a tool prove useful, and when might it possibly be obstructive? Personally, I have used the ‘People who have bought also bought’ feature in Amazon quite extensively as a useful search tool. I am less likely to take up the direct recommendations that Amazon pitches at me through my ‘personalised’ Amazon home page, however. (This comes, in part, from making purchases for a six year old boy. If only I could toggle between ‘mummy’ and ‘professional’ profiles…. now there’s a radical thought).