Regarding the evaluation of AR-QAS, 11 users participated in the study and were asked 22 questions, as shown in Table 1, yielding 242 collected query sentences in this experiment. The study measured the performance of AR-QAS through a method known as k-fold cross-validation Training and testing were performed 242 times (i.e., k = 242). In iteration i, query sentence i was selected as the test corpus, and the other query sentences were collectively used to train the system for the values provided by each question. In this experiment, the feasibility of applying ontology and classification algorithms to QAS was evaluated. Table 2 compares the accuracy of the traditional rule-based reasoning (RBR) and NB classifier with the proposed methods outlined in this study. The results reveal that the accuracy can be higher when the ontology technique is combined with the method. However, because not all of the attributes in the experiments were conditionally independent, the accuracy rate of the NB classifier was not higher. Furthermore,
similar data may not exist in historical records; consequently, the correct class cannot be retrieved using the kNN algorithm. Therefore, a higher accuracy rate of 98.76% results from a combination of the most reliable approaches, which are ANN and Ontology. Therefore, this study applies ANN to Ontology in the AR-QAS.