Knowledge Profiling

[About] [Objective] [Example scenarios] [Publication]

One obstacle to personalization in e-learning is the question of how to obtain a learner profile for a learner who is new to the e-learning system. Although it is possible for the system to derive the learner's knowledge profile over time or by posing a series of test questions, the learner may be unwilling to spend a lot of effort on this procedure. One potential solution to this problem lies in the social Web. If a learner is active on the social Web, a considerable amount of information about her is available online (blog entries, tweets, bookmarks, etc.).

In the line of work followed here, we investigate the specific case of Twitter and research the following question: is it possible to estimate a learner's academic knowledge only from her tweets? We develop a methodology that allows us to perform extensive experiments and to validate our approaches. The initial results indicate that indeed, there is enough information in users' tweet to create academic knowledge profiles, though creating a good profile is a very challenging task.

The long-term objective is to alleviate the cold-start problems of systems that require knowledge about the user's expertise in order to perform effectively. Currently, the two most commonly used strategies are either to present the user with a pre-usage questionnaire (which is often long and tedious to fill in) or to derive a user profile from the user's initial interactions with the system. Both approaches are problematic and a better solution is to infer a user's expertise from information that is already publicly available about the user online.

Two research questions will be addressed in the long term:

• User profile representation: a user profile that encodes the expertise of a user is likely to require a very structured representation that can cover a wide range of knowledge items as well as a wide range of expertise levels.
• Credibility of inferred expertise: if we assume that we can extract a user's knowledge about a topic from the user's social Web signals, we still have to deal with the problem of estimating the credibility of the user's claims. A dishonest user may use his social Web profiles to position himself as having expertise in topic X (to improve his job chances, etc.), and thus we need to derive a methodology that can attach credibility estimates to each statement.

In the ImREAL project, in particular for the medical simulator use case, we can think of the set of topics (the learner's expertise) as the different interview strategies commonly employed in doctor-patient interviews. Each learner, who uses the simulator for the first time, is asked to provide his social Web IDs. The envisioned knowledge profiling module is then used to infer the learner's knowledge for each of the topics. The simulation is adapted accordingly, focusing on interview strategies the learner is unfamiliar with. Overall, this leads to a more effective use of the learner's time as the simulation is directed towards those concepts the learner is (relatively) unfamiliar with.

Claudia Hauff and Geert-Jan Houben. Deriving Knowledge Profiles from Twitter. In Proceedings of 6th European conference on Technology enhanced learning: towards ubiquitous learning (EC-TEL), Palermo, Italy, September 2011. [PDF] [Slides]