In EDS ansehen

Feature-centric mobile applications clustering using GPT-3 enabled context enrichment.

Gespeichert in:
Bibliographische Detailangaben
Titel: Feature-centric mobile applications clustering using GPT-3 enabled context enrichment.
Autoren: Saeed, Sana1,2 (AUTHOR) FA21-PCS-001@cuiatd.edu.pk, Bilal, Kashif1,3 (AUTHOR) kashif.bilal@aasu.edu.kw, Shuja, Junaid4 (AUTHOR) jshuja@semo.edu, Namoun, Abdallah5 (AUTHOR) a.namoun@iu.edu.sa, Ibrahim, Isa Ali6 (AUTHOR) ibrahim.ali@futo.edu.ng
Quelle: Cluster Computing. Dec2025, Vol. 28 Issue 15, p1-20. 20p.
Schlagwörter: *MOBILE apps, *HIERARCHICAL clustering (Cluster analysis), *CLUSTERING algorithms, *DATA mining, *DATA analysis, *GENERATIVE pre-trained transformers, *SEMANTICS, *USER experience
Abstract: Mobile application marketplaces contain millions of applications to help mobile users in their routine task. There are multiple applications for performing single task and grouping similar application together can enhance over all user experience. However, searching for similar application from such a large search space is challenging task. Existing application store ranking systems are not capacitive enough to deal with a multifaceted nature of problem as they group similar applications by using meta-data rather than applications functions and features. Context-based clustering is a highly recommended approach for analysis and categorization of unlabeled textual data. Textual description with the mobile applications contains the latent semantic meanings. However, existing studies for application clustering relies on conventional embedding techniques that struggle to extract context from application descriptions. Hence, existing methods lags behind in encapsulating the true diversity of application functionalities. To overcome these twin limitations, our study performs context enrichment, by using Named Entity Recognition (NER) and (Generative Transformer model) GPT-3. NER identifies key entities in application descriptions while GPT-3 enriches context, aiding in precise grouping of mobile applications by capturing semantic meaning. For similar application grouping, clustering is employed on the state-of-the-art transformer-based contextual embedding generated form the contextually enriched text. Moreover, to construct multiple perspectives of the application store at varying granularity levels, hierarchical clustering is used. Evaluation on large dataset shows the efficiency of proposed method based on silhouette coefficient and Davies–Bouldin score. [ABSTRACT FROM AUTHOR]
Datenbank: Academic Search Index
Beschreibung
Abstract:Mobile application marketplaces contain millions of applications to help mobile users in their routine task. There are multiple applications for performing single task and grouping similar application together can enhance over all user experience. However, searching for similar application from such a large search space is challenging task. Existing application store ranking systems are not capacitive enough to deal with a multifaceted nature of problem as they group similar applications by using meta-data rather than applications functions and features. Context-based clustering is a highly recommended approach for analysis and categorization of unlabeled textual data. Textual description with the mobile applications contains the latent semantic meanings. However, existing studies for application clustering relies on conventional embedding techniques that struggle to extract context from application descriptions. Hence, existing methods lags behind in encapsulating the true diversity of application functionalities. To overcome these twin limitations, our study performs context enrichment, by using Named Entity Recognition (NER) and (Generative Transformer model) GPT-3. NER identifies key entities in application descriptions while GPT-3 enriches context, aiding in precise grouping of mobile applications by capturing semantic meaning. For similar application grouping, clustering is employed on the state-of-the-art transformer-based contextual embedding generated form the contextually enriched text. Moreover, to construct multiple perspectives of the application store at varying granularity levels, hierarchical clustering is used. Evaluation on large dataset shows the efficiency of proposed method based on silhouette coefficient and Davies–Bouldin score. [ABSTRACT FROM AUTHOR]
ISSN:13867857
DOI:10.1007/s10586-024-05082-2