Despite the potential of Multi-Agent Systems (MAS), this technology has not been widely adopted by industry yet. Due to its complexity, errors in modelling activities can be costly. Early validation of MAS models can prevent rework or building a system non-compliant with client's specification. We propose a general ontology-based process to validate any kind of software models that can be adapted in a broad range of software development projects. We illustrate this for MAS development as its complexity justifies additional costs associated with applying our add-on validation process. This work provides early evidence of the soundness of our approach. We successfully validate and improve the quality of MAS models for a real-life development project showing that our MAS models validation process can contribute to harnessing the commercial potential of MAS technology.
Ontologies provide a mechanism for representing domain
knowledge to a varying degree of formalism [
Conference’10, Month 1–2, 2010, City, State, Country.
Copyright 2010 ACM 1-58113-000-0/00/0010…$10.00. potentially reducing the development and maintenance costs of software systems. We provide a methodology-independent and ontology-based add-on to facilitate the creation of models especially in certain scenarios: In developments of inexperienced modelers, to guide their work and avoid errors; in initial MAS developments of experienced modelers in any other technology, as agents have many important particularities which cannot be found in other paradigms; in projects where the domain is complex or unknown, for experienced and inexperienced modelers alike; in projects dealing with the same domain, to enable reuse of the generated domain knowledge (i.e. ontologies). Whilst the focus of our illustrations is on applying ontologies to improve the development of MAS models, we expect that our approach is easily adaptable to other development paradigms such as agile methods.
Many existing works focus on the use of ontologies to MAS. Of
these many focus on the process itself. For example, in [
Our proposed ontology-based MAS software models validation (Figure 1) consists of five activities. Our proposal is an add-on to the development process and it is completely independent of the underlying software models or their development methodology.
In the Ontology Acquisition activity a suitable ontology is retrieved from an existing repository, otherwise one is built using the most suitable ontology engineering techniques. In the Ontology Augmentation activity, the ontology is augmented to represent features related to the chosen development paradigm. Domain concepts are annotated to link them to paradigm concepts and relations between them are created according to existing relations defined for the paradigm. In our case study we identify the MAS terms Goal, Role, Activity, Environment and Agent and relations between them such as Role responsible for Goal, Agent plays Role or Activity follows Activity. In the Ontology Validation activity, members of the development team validate the ontology with the client to reach a common understanding and compliance to client’s conceptualisation. In the Software Models Validation activity, the models are validated against the augmented ontology. This activity provides the control element for new iterations. A new iteration will be necessary as long as any non-trivial recommendation is made to improve the quality of the models. In the Software Models Improvement activity, the recommendations are analysed by the developers to choose which to apply and which to ignore. After improving the quality of the models according to chosen recommendations, the new set of models will be validated in the next iteration. In our case study we validate and improve goal, agent, interaction, scenario, organisation, role and environment models for MAS development.
Development proceeds with each iteration further along the sequence of workproducts required by the chosen methodology. The development and validation of the software models are intertwined and done concurrently. Problems of reviewed models are fixed before their full development. Any models yet to be commenced in that iteration will take advantage of the recommendations avoiding compounded errors. This has been proven in our case study, as recommendations made in the first iteration to the interaction model were used to prevent errors in the development of the scenario during the second iteration.
We apply ontologies to validate and improve the quality of software models. We take into account the domain as specified by the client’s requirements, bridging any communication gap between clients and developers. Models are validated as soon as they are available, fixing errors as they arise and avoiding compounding and propagating errors to later phases of the development. To integrate our validation add-on seamlessly into the development process, we use an iterative, incremental and concurrent development process. The process iterates over intermediate versions of the model to achieve high quality. It is incremental in nature, not all the models are considered during each iteration. It is concurrent as development and validation activities overlap. Applying our process can incur additional development cost and requires a cost justification. It is particularly appealing in critical software application where errors can be very costly and disastrous. This cost overhead may also be justified in the scenarios described in Section 1. That said, the cost of the validation can be greatly reduced by more effective reuse of existing ontologies. With advent of the Semantic Web, more ontologies are made available. More importantly, there is a great scope for generating the amendment proposals automatically, harnessing automatic reasoning capabilities of ontologies. Indeed, we are now studying this possibility with the expectation to develop a tool that can significantly alleviate the burden of the details of the ontologymediated validation process. In the future, we also intend to apply the ontology-mediated software model validation process to further cases studies to fine-tune it and to test our forthcoming tool.