Abstract

This research investigates the potential of machine learning algorithms as an alternative approach to rule-based systems for generating Recursive Object Model (ROM) diagrams. The existing rule-based approach suffers from limitations and challenges, and this study aims to explore the possibility of overcoming these limitations by leveraging machine learning techniques.
To achieve the research objectives, software was developed to gather labelled data for our supervised learning problem. A model comprised of Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) models was created and trained using the labelled data. The proposed model takes a pair of words and a sentence as inputs and classifies the appropriate relations among the pairs. Subsequently, a comprehensive evaluation was conducted to assess the effectiveness of the proposed model.
The evaluation process involved a comparative analysis between the proposed model and a baseline model, an evaluation of the proposed model on unseen data, and an investigation into the capability of the design model in addressing the limitations of the rule-based system. The evaluation results demonstrate the superiority of the proposed model. Firstly, the proposed model achieved an exceptional accuracy of 97 percent in the training process, surpassing the baseline model's accuracy of approximately 61 percent. Secondly, the proposed model exhibited an accuracy of 96 percent on unseen data, thus showcasing its ability to generalize effectively to new instances. Lastly, when comparing the proposed intelligent system with the rule-based system, although the proposed methodology exhibited minor errors in generating ROM diagrams for certain scenarios, the findings underscore the potential of the proposed model in mitigating the limitations of the rule-based system.