“…First, we consider samples with the highest expected pruning cardinality, e.g., 'convertible' and 'sedan' for type, 'red' and 'blue' for color, and 'Ferrari' and 'Honda' for brand. Among these, we decide to obtain a preference elicitation on 'red' and 'blue' first, since its expected pruning cardinality (Theorem 1) is the highest, e.g., 1 2 (60 + 40). Once the elicitation result is obtained, for instance 'red' brand 'blue', for each object with 'red', we can prune out objects with 'blue' sharing the same remaining attribute values.…”