User modeling by confabulation theory

In this paper, we propose a method to generate fictitious user data from a collection of real user data by using confabulation theory. In confabulation theory, the maximum cogent argument for a number of facts is approximated by the maximization of the product of all single-fact cogencies. It has been argued that this approximation is reasonable in cognition. Here, we use this method to generate new user data-sets by successively filling-in positions of a data-set in dependence from already filled-in positions. The proposed method has been applied to the evaluation of questionnaire data about acceptance of several kinds of quality losses in multimedia content distribution. Besides of being able to generate additional user data following the distribution of the collected user data, the number of cogency factors that were used to achieve the best match (four in the studied case) also provides some additional qualitative insight into the acquired data that cannot be achieved by another method.