Paper Title
Optimizing Communication Data Processing With Knowledge Graph and On-Device Large Language Models for Scalable Personalized Insights and Recommendations

Abstract
The exponential growth of communication data across multiple platforms has led to significant challenges in processing and extracting meaningful insights in real-time. Current solutions often rely on cloud-based processing, which introduces privacy concerns and latency issues. These solutions also face token limitations when using Large Language Models (LLMs), making it difficult to process and analyse vast datasets efficiently. In contrast, this paper presents a novel approach that integrates adaptable knowledge graphs, focusing on on-device processing. By condensing large datasets into compact, high-dimensional knowledge graph representations, the proposed system enables LLMs to operate efficiently within token constraints while preserving the depth of relationships and insights. The use of on-device knowledge graphs ensures that communication data is never sent to the cloud, thereby maintaining user privacy. The system dynamically constructs knowledge graphs to convert communication data into actionable embeddings, providing real-time insights without compromising data security.