Ontology-Aware Prediction from Rules: A Reconciliation-Based Approach
Abstract
Our work is related to the general problem of constructing predictions for decision support issues. It relies on knowledge expressed by numerous rules with homogeneous structure, extracted from various scientific publications in a specific domain. We propose a predictive approach that takes two stages: a reconciliation stage which identifies groups of rules expressing a common experimental tendency and a prediction stage which generates new rules, using both descriptions coming from experimental conditions and groups of reconciled rules obtained in stage one. The method has been tested with a case study related to food science and it has been compared to a classical approach based on decision trees. The results are promising in terms of accuracy, completeness and error rate.