Modularization in Object Oriented Software: A Comparative Study

The process of software modularization is essential to comprehending complex software systems. There are a variety of implementation approaches that provide different approaches to software architectural delineation. The advancement of programming languages and techniques presents novel challenges in determining suitable approaches for creating cohesive and separate modules. In this study, modularization increases the understanding of software quality by breaking apart the source code incorporated into packages into coherent and independent groupings. This study assesses the effectiveness of various architectures that integrate clustering algorithms, i.e. Agglomerative, K-means, and K-medoids, with three word-embedding techniques, i.e. BERT, FastText, and GloVe. The evaluation was conducted on five java-based applications with varying number of packages developed by different prominent developers. The primary objective of this study is to identify the optimal combination of clustering and word embedding algorithms that could yield high cohesion and low coupling of java-class modules. The results demonstrate that the combination of K-means and FastText outperforms other architectures in terms of cohesion, with a CH Index value of 10.66 and coupling with a DB Index value of 1.76 of clustered classes. In addition, regarding accuracy and similarity with ground truth, the combination of distance-based clustering algorithms and FastText outperformed other architectures.