LLM-TFP: Integrating large language models with spatio-temporal features for urban traffic flow prediction

Haitao Cheng, Zibin Gong, Chang Wang

Abstract: Highlights•Propose LLM-TFP that combines LLMs with spatio-temporal features for traffic flow.•Develop a tokenizer to convert traffic data into LLM-compatible tokens.•Design an adapter to link tokens with a pre-trained BERT model.•Demonstrate robust generalization in few-shot and zero-shot prediction.