Zusammenfassung

Intelligent assistants are ubiquitous and will grow in importance. Apple's well-known assistant Siri uses Active Ontologies to process user input and to model the provided functionalities. Supporting new features requires extending the ontologies or even building new ones. The question is no longer "How to build an intelligent assistant?" but "How to do it efficiently?"

 

We propose EASIER, an approach to automate building and extending Active Ontologies. EASIER identifies new services automatically and classifies unseen service providers with a clustering-based approach. It proposes ontology elements for new service categories and service providers respectively to ease ontology building.

We evaluate EASIER with 292 form-based web services and two different clustering algorithms from Weka, DBScan and spectral clustering. DBScan achieves a F1 score of 51% in a ten-fold cross validation but is outperformed by spectral clustering, which achieves a F1 score of even 70%.

@inproceedings{,
author={Blersch, Landh{\"a}u{\ss}er},
title={Easier: An Approach to Automatically Generate Active Ontologies for Intelligent Assistants},
year=2016,
month=7,
booktitle={International Symposium on Information Systems and Software Engineering: ISSE 2016},
abstract={Intelligent assistants are ubiquitous and will grow in importance. Apple's well-known assistant Siri uses Active Ontologies to process user input and to model the provided functionalities. Supporting new features requires extending the ontologies or even building new ones. The question is no longer "How to build an intelligent assistant?" but "How to do it efficiently?"

 

We propose EASIER, an approach to automate building and extending Active Ontologies. EASIER identifies new services automatically and classifies unseen service providers with a clustering-based approach. It proposes ontology elements for new service categories and service providers respectively to ease ontology building.

We evaluate EASIER with 292 form-based web services and two different clustering algorithms from Weka, DBScan and spectral clustering. DBScan achieves a F1 score of 51% in a ten-fold cross validation but is outperformed by spectral clustering, which achieves a F1 score of even 70%.},
}