Combining Ontologies and Machine Learning to Capture Tacit Knowledge in Complex Decision Making
- Speaker: Dr Vania Dimitrova, School of Computing at the University of Leeds
- Date: Tuesday, 19 April 2016 from 17:00 to 18:00
- Location: Malet 151
Complex decision making in domains with high impact, such as infrastructure management, is a challenging task that requires the consideration of a large number of parameters and their dependencies. For example, in the case of tunnel management, accurate pathology diagnosis and early risk assessment are critical for making cost effective maintenance plans. The common practice is that such decisions are made by a small number of domain experts, who follow their intuitions and apply tacit knowledge gained over many years of experience. This results in unsustainable subjective decision models, where knowledge can be lost when experts leave. We tackle these issues in the context of tunnel management within the EU project NETTUN (nettun.org). Following ontology engineering methodologies, we have engaged domain experts with extensive experience in tunnel management in a knowledge elicitation process to identify the concepts they investigate and the rules they apply when diagnosing a tunnel.
Ontological models can have some limitations. Firstly, they may not be able to capture the true complexity of the decision process. Then, the process of validating these models is an important one, but laborious and slow. It is hard to identify missing or inaccurate rules, and some rules are ‘more reliable’ than others, but experts typically cannot articulate this information. Furthermore, there are aspects of the decision process, such as risk assessment and potential for further degradation of tunnel portions, which take into account a number of parameters so that experts are unable to elucidate. To address these challenges, we adopt supervised machine learning models, taking advantage of the existence of provenance data with past observations and expert decisions. We use the ontology to enrich the feature space for machine learning methods that derive risk assessment rules and extend the knowledge model. We are also adopting machine learning to learn rules that may be applied to assess the pathologies, so that to facilitate ontology validation.
Bio: Dr Dimitrova (http://www.comp.leeds.ac.uk/vania/) is an Associate Professor in the School of Computing at the University of Leeds, and Co-Director of the Leeds Research Centre in Digital Learning. Her research is in user-aware intelligent systems, focusing on knowledge-enriched user modelling and adaptation, crowd-wisdom, interactive data exploration, knowledge capture and ontological modelling. She has led research activities in several projects developing user-aware intelligent systems to augment human intellect, involving interdisciplinary partnerships with researchers from Education, Business, Psychology, Engineering and end users from industries (e.g. Ordnance Survey, Rolls Royce, SNCF, various SMEs) and charities (via LeedsACTS!). She has chaired premier international conferences on user modelling and intelligent learning environments, and has served as a member of the editorial boards of major journals in these areas (IEEE TLT, UMIAI, IJAIED).