CrossTrafficLLM: Traffic Forecasting and Natural Language

A new study introduces CrossTrafficLLM, a GenAI-driven framework designed to enhance Intelligent Transportation Systems (ITS). The core innovation lies in its ability to simultaneously predict future traffic conditions and generate natural language descriptions of those conditions.

Data and Semantics Alignment

The main challenge addressed by CrossTrafficLLM is aligning quantitative traffic data with the qualitative semantics of natural language. The framework uses Large Language Models (LLMs) within a unified architecture to overcome this obstacle. This approach allows the generated textual context to refine the accuracy of predictions, ensuring that reports are based on the forecasts themselves.

Architecture and Performance

Technically, the system employs a text-guided adaptive graph convolutional network to integrate high-level semantic information with the traffic network structure. Evaluations on the BJTT dataset demonstrate that CrossTrafficLLM surpasses state-of-the-art methods in both traffic forecasting performance and text generation quality.

Implications for Intelligent Transportation Systems

By unifying prediction and description generation, CrossTrafficLLM offers a more interpretable and actionable approach to generative traffic intelligence, with significant advantages for modern ITS applications.