Embedded AI Framework for Low-Latency Sentiment Analysis in Smart Content Curation Systems

Abstract

The fast spread of digital content on multiple networks has caused the growth of the need of real time sentiment-informed analytics to improve user involvement and content delivery techniques. Convincing cloud-based AI Natural language processing (NLP) solutions, as powerful as they are, typically have a disadvantage in latency, bandwidth consumption and data privacy, in bandwidth-sensitive and low-latency settings. To overcome such issues, the framework of a new embedded artificial intelligence (AI) is provided in this paper to handle low-latency sentiment analysis in intelligent content curation systems. The suggested framework uses quantized deep learning networks that are particularly trained to run on resource-constrained edge computing platforms, i.e., the NVIDIA Jetson Nano and the Raspberry Pi 5. The use of a blend of hardware-aware pruning, 8-bit quantization, and mixed-precision training methods by the framework leads to a large scale of reductions in model complexity, power consumption without leading to predictive accuracy decrease. As experimentally shown, the optimised model has a high classification of 92 percent on a benchmark sentiment dataset and a reduction inference latency of less than 150 milliseconds, which is much lower than the limit of the real-time processing. In addition to that, power consumption is also minimised by 34 percent against baseline full-precision models, which underscores appropriateness of the strategy to embedded application with energy-saving. The system architecture integrates an effective data acquisition system and light weight text preprocessing platform and a real time inference system, which allows the smooth integration into intelligent media applications. This paper creates a base of the forthcoming embedded NLP systems, those able to work autonomously and effectively in semi-autonomic settings. It also highlights the possibility of using affordable hardware to do more intricate AI work and, in general, expand accessibility and scalability to real-life applications. These findings make this framework a practical option towards real-time sentiment analysis in intelligent communication infrastructure, personalised recommendation systems, as well as adaptive media content moderation in mobile and embedded systems.