SWMLAlgorithms

Various implementation of machine learning algorithms for Semantic Web knowledge bases

The software provides the implementation of the algorithms for inducing:

• Terminological Decision Trees [2,7,8]
• Terminological Random Forests [2,5]
• Evidential Terminological Decision Trees and Forests [1,3,4,6]
• kernelized perceptron
• k-nearest neighbor [9]

in order to assess the class-membership of OWL individuals

Import the project.

Clone the repository via GIT and import the project using Maven.

Running the software.

For running the program you must to specify the parameters for the experiments in the file experiments.properties. The parameters to be specifiied are reported below:

• NUMGENCONCEPTS, the number of artificial learning problems to be generated in class-membership prediction tasks
• PURITY_THRESHOLD, the purity threshold fdecision tree learning
• originalImbalance, the original imbalance ratio (employed by TRF and ETRFs)
• SEED, the seed for random generator employed by the refinement operator
• NTREES, the forest size
• nonspecificityControl, a flag to impose a prepruning criterion over ETDT learning algoritgm
• design (CROSSVALIDATION/BOOTSTRAP), the design of the experiments
• task, the task to be solved (e.g. CLASSMEMBERSHIPREDICTION )
• BINARYCLASSIFICATION, a boolean to decide if the learning problem must be solved under CWA or OWA
• NFOLDS, the number of folds or run for the experiments
• algorithm, the algorithmn to be used in th experiments (e.g. classifiers.knn.QHDNN)
• urlOwlFile, the file containing the knowledge base (the software supports only OWL file)
• M3, for the M-estimate probability
• samplingrate, the stratified sampling rate adopted by (E)TRFs
• missingValueTreatmentForTDT, a boolean value to decide if the branches of a TDT must be navigated in parallel when an intermediate test return an unknown value
• pruning, the pruning strategy for TDT (NOPRUNING/,REP/PEP)
• beam, the number of candidates generated via refinement operator
• refinementoperator, the refinement opertator to be employed (classifiers.refinementOperator.RefinementOperator)
• CCP, a flag to decide if the information gain must be calculated using the CCP
• mutagenicAsDataPropertyPredicti, a flag to decide the learning problme over CARCINOGENESIS ontology

Publications.

[1] Giuseppe Rizzo, Nicola Fanizzi, Claudia d'Amato, Floriana Esposito: Approximate classification with web ontologies through evidential terminological trees and forests. Int. J. Approx. Reasoning 92: 340-362 (2018)

[2]Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Tree-based models for inductive classification on the Web Of Data. J. Web Sem. 45: 1-22 (2017)

[3] Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Inductive Classification Through Evidence-Based Models and Their Ensembles. ESWC 2015: 418-433

[4]Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi: On the Effectiveness of Evidence-Based Terminological Decision Trees. ISMIS 2015: 139-149

[5]Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Tackling the Class-Imbalance Learning Problem in Semantic Web Knowledge Bases. EKAW 2014: 453-468

[6] Giuseppe Rizzo, Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Towards Evidence-Based Terminological Decision Trees. IPMU (1) 2014: 36-45

[7]Nicola Fanizzi, Claudia d'Amato, Floriana Esposito: Towards the induction of terminological decision trees. SAC 2010: 1423-1427

[8] Nicola Fanizzi, Claudia d'Amato, Floriana Esposito: Induction of Concepts in Web Ontologies through Terminological Decision Trees. ECML/PKDD (1) 2010: 442-457

[9]Claudia d'Amato, Nicola Fanizzi, Floriana Esposito: Query Answering and Ontology Population: An Inductive Approach. ESWC 2008: 288-302