User Modelling via Facebook and the Social Web

in Cyberpsychology, My Research, Social Bits @ October 31, 2010

As I am on the second year for my MSc in Cyberpsychology, I have started work on my research project. My interest in User Experience, Online Education and Machine Learning coupled with my professional background in HCI, Interaction Design & Development for the web and for E-Learning systems, lead me to start exploring new ways to improve the Learning Experience.

The educational aspiration of the 21st century is to evolve how people learn; to unleash the potential and genius of each individual. As pupils and students from all over the world move to online learning environments they create a rich but inhomogeneous ecosystem. The challenge for advanced E-Learning systems is to respond by tailoring the learning experience towards the personality and learning capacity of every learner. Indeed, technological and pedagogical breakthroughs have lead to the development of dynamic E-Learning platforms that identify the user’s preferences and personal characteristics and adapt the learning process towards the individual. Such systems have shown to be successful in learning, as they improve the User Experience, engagement levels and, most importantly, the learning outcome.

The usual methodology towards customised learning environments is the creation of a User Model, which holds a set of user characteristics, attributes and preferences that the system believes to be an accurate representation of the learner. An important piece of information for an E-Learning system is the Learning Style of a student. Learning Styles are a sub-set of the User Model and describe the way one learns and can be extremely helpful in the adaptation of learning material. The most popular Learning Style model is Felder-Solomon’s Inventory of Learning Styles (ILS). It involves a questionnaire of 44 questions that measure the learner’s inclinations towards the following characteristics:

  • active/reflective
  • sensing/intuitive
  • visual/verbal
  • sequential/global

The information contained within a User Model is used as the basis for intelligent adaptation of deliverables such as the learning material and the UI. Two techniques have been used so far for the creation of User Models: questionnaires and behaviour/performance monitoring. Both techniques have significant drawbacks, as explained below.

Learning Style questionnaires must be completed before a student starts using an E-Learning system and although some useful data can be gathered, the reality is that questionnaires are disruptive to the learning process, they only provide data from a single point in time (learning styles and user characteristics can evolve/change) and finally, the validity of the students’ answers is debatable since users can provide false information for reasons related to self-image representation, misunderstanding of the questionnaire or plain boredom.

Behaviour and performance monitoring occurs after a student has started using an E-Learning system and is continuous during the learning process. This is a highly transparent and non-disruptive way to profile the user but extremely slow to produce useful results and highly reliant on the system’s capabilities to identify characteristics of the learner that might be useful.

The question then becomes, how can an accurate (or accurate-enough) User Model can be created prior to the learning process, without the use of a questionnaire, that will continue to evolve along with the learner? The problem is really a matter of finding relevant data about the user, possibly from third-party sources, that can be used as a replacement to questionnaires and provide accurate information. The answer might lie within the Social Web.

As people embrace the tremendous growth of the Social Web they create online profiles, connect with each other and leave vast amounts of online traces behind them.  My hypothesis is that useful quantifiable information about a person (preferences, psychological traits, etc) can be inferred from large abstract data sets, like Facebook’s Open Graph, and consequently used to build accurate-enough User Models that can be used to provide an adaptive learning experience.

I will document the progress of my research in this blog, under the My Research category.