Enhancing Last-Mile Delivery: Social Media Insights and Deep Learning Applications

Accurate traffic forecasting has become essential due to increased urban deliveries driven by growing e-commerce and urban expansion. This surge has increased traffic in large cities, resulting in delays and numerous accidents. The convenience of online shopping and home delivery continues to fuel the e-commerce sector. As this sector grows, the complexity of its challenges increases, necessitating quicker solutions. These challenges often extend beyond the control of delivery companies when influenced by external elements such as traffic congestion or adverse weather conditions, particularly in the last-mile delivery. This issue is intensified in regions lacking extensive traffic sensor networks, such as less developed countries. This research contributes by (1) establishing a contextual groundwork for existing traffic prediction frameworks, (2) employing social media and diverse traffic-related data (including weather conditions, significant locations, and event schedules) through social network analysis to refine traffic prediction accuracy, and (3) presenting a method applicable to partially observable traffic situations. The methodology integrates advanced deep learning techniques like Long Short-Term Memory Network. In addition, a future extension aims to develop a robust real-time traffic prediction system, Traffic GPT, utilizing deep learning and transformer methodologies, optimized for diverse urban environments with real-time updates through transfer learning. Key to this approach is the use of sentiment analysis supported by training on datasets like the IMDB sentiment dataset with techniques like AWD-LSTM’s attention mechanisms, and Graph Convolutional Networks, along with tools for analyzing social media sentiment.