T h refers to a threshold for the ratio between the number of vertices of two non exact occurrences of P PU . If T h is equal to one, the occurrences must be isomorphic between them. The output of uEP-FPDA are two matrices, score and F reqM . Rows in the score matrix represent each vertex u in P M . Columns represent the results of the algorithm for different pattern sizes6 . If u belongs to a highly frequent subgraph in P M of size j then score(u, j) will be also high. F reqM is a matrix with |VP M | rows and k columns, where each cell indicates the frequency of a discovered pattern.

Exact, Inexact and Partial Pattern Matching In realistic scenarios where an exact match of a pattern is unlikely, partial and inexact matching become relevant. Inexact pattern matching provides good, but not exact solutions to the matching problem. In this case, pattern instances can incorporate additional elements not described in the pattern, nevertheless they must not affect the structural properties of the pattern. Partial pattern matches identify exact but incomplete matches of patterns. Partial instances of patterns might exist due to modifications or evolution of previously instantiated patterns.

Labels in LVP M can be mapped to specialisations of or equivalent labels in LVP P . In this manner, generalised patterns are considered as families of patterns where the parent pattern contains the roots of tree-structured taxonomies for business concepts in specific domains. Child patterns contains one or more child concepts connected to root concepts in the hierarchy defined by the taxonomy. Note that using the uG-PMA algorithm requires the existence of an implemented taxonomy in which the algorithm can search for semantically corresponding terms.