A Clustering-Based Approach to the Mining of Analogical Proportions

This paper presents an approach aimed at mining a new type of pattern in data, namely analogical proportions. An analogical proportion expresses the equality of the relationships between the attributes of two pairs of structured objects. This notion is investigated in the database context for the discovery of different forms of "parallels" between pairs of tuples. First, we give a formal definition of the analogical proportion in the setting of relational databases. Then we focus on the problem of mining analogical proportions. We propose to use a clustering approach for enumerating parallels occurring in a relation, thus discovering analogical proportions.