<scp>MAD</scp> ‐ <scp>KG</scp> : A Hybrid Knowledge Graph Construction Framework for Mobile App Development Domain

ABSTRACT Objective Mobile App Development (MAD) is a rapidly evolving domain with an extensive and continuously expanding body of knowledge, making it difficult for developers to efficiently identify relevant tools, frameworks, best practices, and emerging trends. This study aims to develop a domain‐specific Knowledge Graph (KG) that enables intelligent knowledge management and semantic understanding to support developers in navigating the dynamic MAD ecosystem. Method We propose MAD‐KG, a domain‐specific KG that integrates deep learning, ontology engineering, and Large Language Models (LLMs). The framework employs XLNet to generate contextual embeddings from unstructured text, followed by a BiLSTM network to model sequential dependencies and a self‐attention mechanism to emphasize salient contextual information. A Conditional Random Field (CRF) layer produces coherent entity labels. The extracted knowledge is enriched through the integration of GPT‐4 and MAD‐Onto, a specialized ontology, enabling dynamic knowledge enrichment and semantic reasoning. Knowledge is constructed from heterogeneous data sources, including X (formerly Twitter), GitHub, and Stack Overflow, to capture technical concepts, developer interactions, and emerging trends. Results Experimental evaluation demonstrates that MAD‐KG achieves state‐of‐the‐art performance, obtaining an F1‐score of 89% for named entity recognition and 78.9% for relation extraction. Ablation studies confirm the contribution of each architectural component, highlighting the effectiveness of combining transformer‐based language models, sequential learning, attention mechanisms, ontology‐driven reasoning, and LLM‐based knowledge enrichment. Conclusions MAD‐KG provides a scalable and adaptive framework for intelligent knowledge management in MAD. By combining structured knowledge representation with ontology‐guided reasoning and the adaptability of LLMs, the proposed framework maintains an up‐to‐date and semantically rich knowledge base that enhances developer productivity, reduces redundant effort, and accelerates innovation in the rapidly evolving MAD domain.