Adaptive Interactive Systems Practical 2: User Modelling
You can work individually in this practical, or with somebody else (as you prefer).
This week we will look at incorporating User Models into the Arts Tour.
we will represent a painting as a vector with 4 very simple features:
"Abstract", "Landscape", "Portrait", "Animal". Each painting has one or more of these features present.
We want to learn what kind of paintings the user likes, as a function of these features. Our algorithm is very simple:
- Initially set the weight of each feature to 1.
- Each time a user likes a painting, double the weights of the features of that painting.
- Each time a user dislikes a painting, half the weights of the features of that painting.
Play with the demo at
User Modelling Demo, and see how the user model changes with your explicit preferences for paintings. Feel free to look at and modify the code, which is again in php. You can find a zipped version here.
In the User Modelling Demo, the weights for features were initialised to 1 (Coldstart problem) and then adapted as the user provides explicit ratings. Can we get around the Coldstart issue? Can we avoid requiring the user to provide explicit ratings?
- Identify and specify potential stereotypes
- Determine relevant user characteristics for adaptation
- What can the system ask the user, what is difficult to ask?
- What can the system observe?
- Are the user's personality traits a useful thing to model? Look at the Big Five Personality test below (Task 3).
1. Individual Traits
We discussed a number of individual traits in the lecture. You can have a look at the following to get more insight in what some of these do, and the kind of questionnaires used: