FTMnodes: Fuzzy tree mining based on partial inclusion

Mining frequent patterns from huge databases have been addressed for many years and results have been applied to many fields, including banking, marketing, biology, health, etc. Fuzzy approaches have been proposed in order to soften the constraints on the patterns found by the algorithms. However, when dealing with complex databases such as tree databases (as it is for instance the case for XML databases), only a few methods have been proposed in order to handle soft constraints in discovering the frequent subtrees from a forest of trees. Such algorithms can hardly deal with real data in a soft manner. Indeed, they consider a subtree as fully included in the super-tree, meaning that all the nodes must appear. In this paper, we extend this definition to fuzzy inclusion based on the idea that a tree is included to a certain degree within another one. This fuzzy degree being correlated to the number of matching nodes. We propose the FTMnodes method together with the associated definitions, and we report the experiments lead on synthetical and real databases, showing the interest of our approach.