Feature-Centric Mobile Applications Clustering with Generative AI
Sana Saeed, Kashif Bilal, Abdallah Namoun, Isa Ali Ibrahim, Junaid Shuja · 2024
Utilizing context-based clustering offers a robust method for analyzing and categorizing unlabeled textual data. Textual descriptions of mobile applications often contain latent semantic meanings. Many of current studies still rely on, traditional embedding techniques, however, traditional embedding methods struggle to extract the contextual meanings from textual descriptions of mobile applications. As a result, these methods frequently fail to capture the full range of application functionalities. Moreover, mobile app marketplaces contain millions of applications and comparing a single application with all the available apps (pair wise similarity matching) on such a large scale is practically infeasible. Our study addresses these limitations by leveraging advanced transformer-based models to enhance context. These models produce contextual embeddings that accurately represent semantic meaning, enabling precise clustering of mobile applications offering similar features and services. We apply hierarchical clustering to these sophisticated contextual embeddings. Hierarchical clustering presents multiple perspectives of the app store at different levels of detail. Our evaluation on dataset shows the effectiveness of our approach, demonstrated by silhouette coefficient and Davies-Bouldin score results.