Abstract
One of the most important purposes in the analysis of questionnaire data is to get profiles for the target group(s). As the result of profiling gives a great influence on the planning marketing strategy, the reliability of profiling is very important. Correspondence Analysis and Association Rule mining are typical methods for profiling
and useful to grasp the relationships between categorical variables in data. However, the results of these methods may "overfit" to the data and be different from the behavior in the general population, especially when the sample size is small. This paper proposes a new profiling method considering behavior in the general population by probabilities. It derives the probabilistic distributions of each choice on a question by Bayesian approach for each attribute, and visualizes the similarities between these distributions. This paper applies the proposed method to an actual questionnaire data on the impression of cloths for men's suits. It shows that a user can profile the data for the relationship between each attribute and the answer to the essential question considering the uncertainty of extracted rules. It also shows that the visualization supports a user to grasp the similarities between the probabilistic distributions for each attribute and to extract the characteristics of attributes.
