Recommender Systems (RS) are a commonly used filtering mechanism to cope with the information overload problem – i.e., automatically providing relevant information out of very large information spaces by considering the context (e.g. user preferences, task, history etc.). Primarily, RS are used to provide results in form of text, or list items, but recently, they were also used for recommending other kinds of information – the visualizations.
We provide a Recommendation Dashboard (RD) that includes several interactive visualizations for exploring and analyzing results of a Recommender System (RS). The RD generates the visualizations automatically. To avoid providing user with junk visuals the RS tracks user behavior, and provides visualizations that best match to that behavior. Thus, the central component of the RS is a user model, or profile – a collection of parameters that must be tracked in order to feed RS with enough information about the interactions and in the end to build a user profile.
The student has to implement an algorithm which tracks and collects users’ interaction with the visualizations and which defines behavioral patterns based on collected data. The patterns will be used to infer users’ next action using a content-based RS and in turn to recommend visualizations which address users’ task in the best possible way.
This mission can be summarized as follows:
· Research on tracking, user profiling and behavioral patterns (ideally for CB-RS)
· Analysis and evaluation of tracking method (finding best approach)
· Definition of user profile for the project
As an additional step the student has to integrate tools or implement own methods to modulate visual saliency within the visualizations (i.e. steer user’s attention), for example to support:
· Peak and trough detection to detect the global maximum and minimum point
· Changes over the time
That information will be used by the CB-RS to improve the recommendations.