W3C

Chat log from User Modeling for Accessibility Online Symposium

15 Jul 2013


Introduction

http://www.w3.org/WAI/RD/2013/user-modeling/#agenda

http://www.w3.org/WAI/RD/2013/user-modeling/#proceedings

<Stephan_Roth> @Silvia: Is the profile locally stored?

<andy> In education there are many kinds of data that need privacy - ti would be not sensible to develop a privacy solution separately for accessibility preferences

<andy> Opinion:Different architectures for learning systems may need different privacy solutions - preferences can piggy back on whatever the solution appropriate to the architecture in use is

<Whitney_Q> Opinion: Maybe the most important thing is that privacy is thought about from the start, so it's an integral part of the technical and interface design

<Pradipta_Biswas> A paper on user profile clustering from UniCamb http://link.springer.com/chapter/10.1007%2F978-1-4471-2867-0_20

<Nick_Kaklanis> Our virtual user models repository can be found here: http://veritas-project.iti.gr/VirtualUserModels.html

<Christos_Kourou> thanks for sharing

<Whitney_Q> I think that the idea of storytelling

<Whitney_Q> is good for understanding variation and context

<Whitney_Q> It is information to feed into

<Whitney_Q> technnical development and dynamic profiles.

<Whitney_Q> IT makes sure that we have good assumptoins

<Whitney_Q> about what users will need, and how they can interact

<Whitney_Q> with the systems well

<Whitney_Q> One of the biggest problems I see in many preference systems is that even if they have the right preferences, they are too hard to use. Users do not understand the implications of each setting. Or the interaction is "clunky" and therefore not as useful.

<Pradipta_Biswas> I told 'Story Telling or Automatic Adaptation is NOT a bad idea', though it printed otherwise in live caption

Session 2

<andy> I've jotted down some thoughts/quesitons about some of these things - is it useful to post them here ? I'd be interested to here the responses of others on them

<andy> 1. Rhetorical Question/Issue/Statement to all (doesn't need me to explain it verbally but I hope others will express opinions too): How can we get some homogeneity across these approaches so as to begin to exploit advantages of interoperability at scale. This is particularly pertinent in view of the increasing proliferation and diversification of technologies ?

<andy> 2. With respect to dynamic profiles - much data starts out as static structures but as the power we have increases they move towards being dynamically-generated info or parts of them - there is no difference except in terms of what one has the power to do at runtime - the nearer to runtime stuff is done the better

<andy> 3. How can these approaches move into practice, what is the progression path for good work that is published in this group - is there one ?

<Yehya_Mohamad> @Andy: we have at the end half an hour, wehere you can bring your thoughts and questions.

<Whitney_Q> andy (2): perhaps one of the questions for user research/personas is how can we create a good set of starting points, so as to require less need for additional personalization. It will never be zero, but a way to provide a consistent set of starting points would be useful.

<Yehya_Mohamad> @Andy: If you have a direct comment to the authors, we can place it in between

<andy> Matthew - yes absolutely correct

<Matt_Bell> Andy/Yehya - you may be reasking your question in a later discussion, for now I would quickly approach your scaling problem via localisation.

<Yehya_Mohamad> what do mean?

<Yehya_Mohamad> what do you mean?

<Matt_Bell> would you prefer discussion to be held to the end, or is this a channel for parallel discussions?

<andy> I'm just making general points - some authors are answering them for me - for example the abstract model that can represent device aspects such as what browser can do what (sorry, don't remember who said it)

<PhilipA> That was me

<Yehya_Mohamad> you can enter your thoughts or questions here. But it would be difficult to discuss in parallel. We can take immediate questions and postbone general to the end

<Pradipta_Biswas> Hi Chairs, I shall be back in five minutes

<andy> Whitney - do you have a feeling for how we can prevent the misuse of personas as stereotypes ? For example its my experience that *we* know its a model and not accurate to an individual - but in the commercial world there is much pressure to do stuff for the cheapest possible cost and avoid an idealised view ?

<andy> Love that approach Whitney

<andy> :-) nice answer

<Whitney_Q> Yes, personas as a starting point for developing a technical virtual user model - with good mapping between them.

<Whitney_Q> It's the mapping that's important: that we can trace the models back to user data; and see how user stories travel forward into models and technical implementation.

<Whitney_Q> Audio is breaking up!

<andy> please repeate

<Whitney_Q> Matthew, agree: after the complete analysis of full sample, there is often more similarity. It's the difference between open coding and identifying categories that join them.

<Stephan_Roth> Are the device profiles shared across users?

<PhilipA> Is it a question to me? If yes: no, they are not shared between the users, because they are very individual and contain information about hardware and installed/activated software.

<Whitney_Q> PhillipA: Can information about clusters of preferences be used as ongoing data collection?

<Matt_Bell> Whitney, that is the advantage of a semantic approach... if you define a person's profile in terms of one ontology at a low level of abstraction, it can be transformed for use in another ontology, if/when it is created

<Whitney_Q> Thanks, that's helpful

<Whitney_Q> Random question: where is the boundary between "the perfect" and "the good" (and the "not good enough")?

<Whitney_Q> Another random question: is there a difference between "need" and "preferences"? We tend to conflate them, but in thinking about removing barriers, there may be a base level of "needs" that can be enhanced? Or are there - especially in vision - so many variations, that needs are more effectively expressed as preferences?

<Matt_Bell> That depends on the individual person and application. *IMHO* Bad data is preferable to no data - provided you know that the quality is low.

<Whitney_Q> In sampling, we say "recruit loosely, and grade on a curve" (from Steve Krug)

<PhilipA> Whitney_Q: Can you describe in more detail what you mean by "Can information about clusters of preferences be used as ongoing data collection?"

<Whitney_Q> If a cluster analysis shows that groups of prefs tend to co-exist, that is valuable data, even if we do not know (yet) why this is true? It's like the predictions of what products you might like, when based on overlap of your other preferences and choices.

<Matt_Bell> Whitney - that is another good point... you cannot describe a person using a single datapoint. Even if the context remains static a user may have a range of values that are valid for a single preference.

<Whitney_Q> Yes - just as we see ability as a range, preference can also be a range.

<Whitney_Q> This has been a really interesting and provocative conversation! Thank you everyone. I also want to thank the organizers for a good symposium.

<Whitney_Q> Is the medical (characteristics) model helpful in static design elements, but a prefences model more useful in dynamic media?

<Whitney_Q> Once, I was reading an ebook, while I was on a train. As the day faded and it got darker, I found I changed the text size, making it larger and larger." How do we use a story like this to make dynamic models.

<Markus_Modzelewski> I think it depends on the question what you want. The medical model is much more suitable if you have questions about success / failure. If you focus on scale or fuzzy data, a preference model is more suitable.

<Whitney_Q> Good point.

<andy> i agree Pradipta

<Whitney_Q> Agree - mapping between models is important

<Whitney_Q> We are just echoing the speakers

<andy> agree with Matthew on context

<Matt_Bell> Context is a detailed version of metadata... the more metadata attached to data the better

<Matt_Bell> Thanks to the captioning, that was very useful... Thanks to the organisers and chairs as well.

<andy> Thankyou for an interesting symposium

<deborah> Thank you so much, everyone.

<Silvia_Mirri> Thank you Chairs, thank you Shadi, thank all of you!