Study and application of new methods for ontology matching

Resumo

In the last years we have experienced the growing interest of the communities in further discovering and interacting with their environments. To this end we have been witnesses of initiatives to develop smart environments to control our houses, hospitals, factories, etc., in an attempt to increase our quality of life and improve our experiences as users. Such initiatives have been possible due to the Internet of Things (IoT), considered as the next step in the evolution of the information and communication technologies (ICT). The IoT is taking connectivity to new devices, such as RFID tags, sensors, actuators, etc., and with these devices, new ways to interact with our environments. However, these initiatives do not end with smart homes or factories but had also spread to Smart Cities. Smart Cities is a field on the rise that attracts interest not only of the research community but also of companies and public governments. From a user point of view, a smart city can be seen as an integration of the services, such as traffic management, public transport scheduling, etc., devoted to making cities more citizen friendly, efficient and sustainable. However, to provide the users with such services, the research community should consider the smart city from a lower point of view and regard it as an environment where different agents producing and/or consuming information coexist. Hence, in a smart city are equally important the agents that obtain measurements from the environment and those that act on behalf of an user to answer a certain query. The knowledge of any of these agents, as well as the smart city’s knowledge, can be semantically represented using different ontologies. To have an open smart city that is fully accessible to any agent and hence to provide the enhanced services to the final users, there is the need to ensure a seamless information exchange between the different agents and the city. If the different agents and the smart city use a common ontology, the information exchange would be straight. However, in most cases, this is not the case. A smart city is usually developed in several stages and possibly by different parties since it is common that public and private deployments coexist within the same smart city. In each one of such stages, a new sub-system is added to offer a new range of services, which rely on a new set of devices and which define new types of agents to interact both with the devices and with the final users. In order to ensure, that this new sub-system is compatible with the previous ones, some approaches propose forcing the ontologies of the different types of agents to comply with that of the smart city. Another approach would be to provide the smart city with a set of predefined correspondences between the concepts represented in the different ontologies and exploit such equivalences as bridges to translate the information. However, these solutions would either contradict the open nature of a smart city, as in the first case, or limit its evolution, in the second. For instance, using the first approach, a user-agent from another city would not be able to use any service and it would not be possible to reuse sub-systems between smart cities. Following the second approach, it would be necessary to manually update all the equivalences in case that an ontology changed and propagate such changes all over the smart city. In this thesis we describe a novel proposal to tackle this problem: relying on ontology matching techniques to guarantee the automatic information exchange between the agents and the smart city. In order to develop our proposal we have studied both the background on ontology matching and smart cities aiming at gaining knowledge on the different matching solutions, on how to apply them to a practical scenario as a smart city and what the requirements of ontology matching in smart cities are. To gain such knowledge we have performed a review of the literature published on ontology matching within the last decade and we have also performed a practitioner-oriented survey to obtain first-hand knowledge on the main challenges in ontology matching. This review and subsequent survey supported our intuition that the amount of practical applications where ontology matching is used is far below the theoretical developments and that this fact still remains as an area of concern for practitioners. With this background review, we have also confirmed that the integration of ontology matching solutions in smart cities has not been addressed before. By themselves, both the literature review and survey are also useful tools for new practitioners approaching the field get a general idea at a stroke. We have applied this knowledge to the definition of a new ontology matching algorithm, which we have named OntoPhil, to be integrated in a smart city and to assume the task of matching the different agent’s ontologies to the ontology of the smart city avoiding human interaction. In this algorithm we have used different matching techniques that exploit both the linguistic features of the ontologies and their internal structures. In the scope of this algorithm we have also defined a new measure to determine the similarity between the concepts in the ontologies to match. This algorithm works in three main steps. First, by relying on the similarity measure that we have defined, OntoPhil computes some initial correspondences between the entities of the ontologies. We named these initial correspondences binding points. Then, taking these binding points as pivots, we rely on structural matching techniques to discover other binding points to exploit them as well, this way iteratively building a set of possible correspondences between the ontologies. As final step, this set of correspondences is filtered to produce the final output of the algorithm, where only the most promising correspondences are kept. The next step in the evolution of our thesis was validating the defined algorithm. To do this, we have followed a twofold approach. First, as matching algorithm, we have used the benchmarks and datasets provided by the Ontology Alignment Evaluation Initiative (OAEI’13), to assess the overall behavior of OntoPhil in different matching tasks. These tests also allowed the comparison of OntoPhil with some state-of-the-art matching algorithms. Such comparison was done by relying on the results of the OAEI matching tasks and also supported by a subsequent statistical evaluation procedure. For this statistical evaluation we have applied the Friedman’s test and the Holm’s procedure on the results of one of the OAEI matching tasks. The results obtained account for the validity of OntoPhil as matching algorithm and place it among the top performing algorithms. Besides, the application of the statistical procedure can be used as an example for other researchers as procedure to compare different versions of their algorithms or an algorithm to some reference others. The second part of the validation was to test the performance of OntoPhil as part of a smart city. To this end, we have defined some specific experiments, involving the SOFIA ontology for smart cities and several agents’ ontologies. These agents’ ontologies represent the main type of agents that are deployed in a smart city and were retrieved online out of websites devoted to projects and initiatives of IoT and integrated in our testbed. The results obtained corroborate the usefulness of ontology matching in smart cities and open the door to further developments in this field.