Traditional methods compress the two objectives of deviation and support into a single number via a single scalar quality measure (e.g., a weighted combination with an exponent), which essentially selects a fixed cut point on the Pareto frontier and therefore cannot accommodate different user support preferences.