Fundamental Research on Automated Approach in Detecting Contradictions in Requirements

Abstract

Dialectics, in philosophy, and logic, is the art of examining or discussing the truth of assumptions through the opposition of opposite ideas, resulting in the reconciliation of them. The three-stage dialectic process normally employed by philosophers such as Hegel and Marx [1] involves the thesis, then the antithesis, and finally the synthesis. This paper gives an attention to the automation of the dialectical process in software engineering. Hence, the generation of synthesis from contradictory statements using the more sophisticated Natural Language Processing techniques is put to fore through this paper. The present study[2] envisages a new form of requirement analysis, now in software engineering, through the assistance of Natural Language Processing principles. This will focus on the dialectical process. The study employs the RoBERTa model from Hugging Face for the contradictions found in software requirements datasets from Kaggle. The application of the model gives datasets for analysis to obtain the desired accuracy. The result is achieved by evaluating the model performances on three different systems, each having different datasets. The first system has 174 requirements, 8 of them being contradictions, and an accuracy of 94%. The second system gives an accuracy of 95% with a slightly larger dataset. This, however, drops the accuracy to 84% for the third system, which contains a considerably larger dataset. The paper systematically applies NLP techniques to preprocess and tokenize data; further, it resorts to a fine-tuned, pre-trained RoBERTa model for detecting contradictions and requirement analysis. Results are evaluated against accuracy, precision, recall, F1 scores, and through confusion matrices. The findings[3] show the real effectiveness of the dialectic process in finding contradictions and improving requirement quality, but also underline challenges related to bigger datasets. We provide evidence that NLP-driven dialectical methods have potential within software requirement analysis, since we have indicated what can be achieved and where limitations may be. It provides a foundation for future work on this important issue. It concludes by emphasizing the role of dialectic reasoning in the resolution of conflicting requirements, and it opens a variety of further research issues on the development of advanced dialectic models, automated argumentation frameworks, and interactive tools for stakeholder collaboration.