Table Transformation Rule Learner

As we know, table data is a popular data form in industry and scientific research fields. However, sometimes the original table data could not meet updating requirements in real applications, so we need to convert them into required form. In this paper we propose an approach to learn the transformation rules that convert original table data to target form. Based on Inductive Logic Programming(ILP), we design a learning system called Table Transformation Rule Learner (TTRL). It uses specific predicates and background knowledge for this task to generate table transformation rules. We implement a unique heuristic function (HF) in TTRL to accelerate searching process for rule generation, and we use semi-supervised learning (SSL) in order to obtain more information especially from small set of sample data. We also address the problem like over-generalization which may occur when having only positive training examples in ILP learning process. We test our program in several kinds of table data, and the result shows that the transformation rules can be learned correctly. Moreover, our designed searching strategy can greatly reduce the time cost of searching rules.