TCTNew

The project aims at discovering disjointness axiom through the induction of a Terminological Cluster Tree (TCT), an unsupervised model that extends first order logic tree in order to be compliant with SW representations.

How to run

The project can be imported as a Maven project after the repository is cloned. For the sake of comparisons, the project allows to discover disjointness axioms via 3 possible approaches:

• terminological cluster tree
• pearson coefficient correlation
• negative association rule mining Each method can be invoked by passing as argument on of the following value: "tct", "corr, "apriori"

The progam can be configured modifying the file experiments.properties that contains the value of the following parameters:

• the seed for controlling the random aspects of the algorithms
• FOLDS: the number of run required for an experiment (in order to mitigate the effectiveness of random choices)
• beam: the number of candidate refinements generated via refinement operator and used as features of the tree
• distance: the distance measure adopted by the terminologicla cluster tree induction algorithm (admissible values: simpleDistance1, simpleDistance2, entropicSimpleDistance1, entropicSimpleDistance2, sqrtDistance1, sqrtDistance2)
• timeout, a timeout for stopping the induction of a TCT (0 means that the algorithm does not use a timeout)
• nResults, to trunk the results output by the extraction of a TCT ((0 means that the algorithm extracts all the disjointness axioms)
• refinementoperator, the refinement operator adopted by a TCT (admissible values: single , i.e. a single thread refinement operator, -spark, i.e. the spark implementation of the previuous one)
• prototype, the kind of prototypical individuals used to split the set of individuals (admissbile values: single, a single linkage approach to cluster the individuals, medoids, i.e. a partitioning around the medoids of two clusters of individuals)
• split, the type of split used to partition the set of individuals rooted to the current node (admissible values: instance, i.e. according to the instance check, or prototype, i.e. according to the prototypes-- the medoids by default)

Publications

• Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Terminological Cluster Trees for Disjointness Axiom Discovery. ESWC (1) 2017: 184-201

• Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Induction of Terminological Cluster Trees. URSW@ISWC 2016: 49-60