Reading Material Classification (RMC) determines Determining the particular readability graded reading material from an unclassified text based on its text readability. RMC have used Natural Language Processing (NLP) methods, i.e., machine-learning-based RMC, to overcome disadvantages of using syntactic features, i.e., insufficiency for modelling the levels of text reading difficulty. Concepts can be varied somewhat between different contexts, therefore “contextual concept-variants” wanted in our ontologies which will result in Contextual Ontology (CO) by using basic NLP techniques such as stemming and word sense disambiguation. Rule-Based Systems (RBSs) are Knowledge-Based Systems whose knowledge is structured by rules. Given RMC as the test-bed, we propose to combine CO with RBSs for synergising their advantages, to improve RBS performance by adding contextual ontological processing into RBSs. This addition will also provide an extended inferencing to RBSs, so that (a) in pre-processing of inferencing, a powerful CO can be extracted, (b) in in-processing of inferencing CO can be utilised for making inferences along with some features (rule strength incremental modification, the state of memory affection and system’s weight), and (c) in post-processing in inferencing, CO can be exploited. Based on the evaluation experiments, we do not claim that our proposed method is better than machine-learning-based RMC. Instead, our system performance is just on a par with them. Rather than beating those methods, we aimed to use RMC to show that adding contextual ontological processing into RBSs (RMC-RBS+CO) presents a considerable benefit for RBSs than not adding it (RMC-RBS+O). 31.25% and 25.89% improvements can be obtained for validation and testing data, respectively.
contextual ontology, machine learning, reading material classification, rule-based systems.