Abstract:Traffic prediction is a vital component of intelligent transportation systems. By reasoning about traffic patterns in both the spatial and temporal dimensions, accurate and interpretable predictions can be provided. A considerable challenge in traffic prediction lies in handling the diverse data distributions caused by vastly different traffic conditions occurring at different locations. LLMs have been a dominant solution due to their remarkable capacity to adapt to new datasets with very few labeled data samples, i.e., few-shot adaptability. However, existing forecasting techniques mainly focus on extracting local graph information and forming a text-like prompt, leaving LLM- based traffic prediction an open problem. This work presents a probabilistic LLM for traffic forecasting with three highlights. We propose a graph-aware LLM for traffic prediction that considers proximal traffic information. Specifically, by considering the traffic of neighboring nodes as covariates, our model outperforms the corresponding time-series LLM. Furthermore, we adopt a lightweight approach for efficient domain adaptation when facing new data distributions in few-shot fashion. The comparative experiment demonstrates the proposed method outperforms the state-of-the-art LLM-based methods and the traditional GNN- based supervised approaches. Furthermore, Strada-LLM can be easily adapted to different LLM backbones without a noticeable performance drop.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are the complex spatial and temporal dependencies in traffic prediction, as well as the challenges to the model generalization ability posed by the differences in data distribution in different regions and at different times. Specifically: 1. **Spatial Dependence**: Changes in traffic volume are affected by the topological structure of the urban road network. In particular, the traffic conditions on upstream roads affect those on downstream roads through the transfer effect, and vice versa. This interaction makes traffic prediction complex. 2. **Temporal Dependence**: Traffic volume changes over time, showing periodicity and trends, which are usually affected by holidays, working hours, and other social events. 3. **Data Distribution Diversity**: There are significant differences in the distribution of traffic data in different regions and time periods, which poses a huge challenge to finding a general - purpose model. Existing traffic prediction methods mainly focus on extracting local graph information and forming text prompts, but these methods have limitations when dealing with large - scale graph data, especially in terms of insufficient adaptability when facing new data distributions. Therefore, this paper proposes a new graph - based large - language model (Graph LLM) - Strada - LLM, aiming to solve these problems and improve the accuracy and generalization ability of traffic prediction. ### Main Contributions 1. **Probabilistic LLM**: Strada - LLM is the first probabilistic LLM specifically designed for spatio - temporal traffic prediction, and it can adapt to new data distributions with a small number of samples. 2. **Graph - Awareness**: Strada - LLM takes spatial dependence into account and enhances the graph - awareness ability of the model by implicitly encoding the network graph. 3. **Light - weight Domain Adaptation**: A low - rank method is adopted for efficient domain adaptation, enabling it to quickly adapt to new data sets. 4. **Comprehensive Evaluation**: Evaluation is carried out using multiple real - world traffic data sets, and the results show that the prediction error of Strada - LLM is significantly reduced and it outperforms the baseline methods. 5. **Flexibility**: Strada - LLM can be combined with different known LLM backbone models without significantly degrading performance. Through these innovations, Strada - LLM aims to provide a more effective and flexible traffic prediction solution, especially suitable for scenarios with diverse and complex data distributions.

Strada-LLM: Graph LLM for traffic prediction

Spatial-Temporal Large Language Model for Traffic Prediction

TPLLM: A Traffic Prediction Framework Based on Pretrained Large Language Models

Modeling Dynamic Traffic Flow As Visibility Graphs: A Network-Scale Prediction Framework for Lane-Level Traffic Flow Based on Lpr Data

Network-Scale Traffic Modeling and Forecasting with Graphical Lasso and Neural Networks

Lane-Level Traffic Flow Prediction with Heterogeneous Data and Dynamic Graphs

LLM Multimodal Traffic Accident Forecasting

LSTM Variants Meet Graph Neural Networks for Road Speed Prediction

Lanczos Method for Spatio‐temporal Graph Convolutional Networks to Forecast Expressway Flow

Network-Scale Traffic Modeling And Forecasting With Graphical Lasso

Traj-LLM: A New Exploration for Empowering Trajectory Prediction with Pre-trained Large Language Models

The Prediction of Multistep Traffic Flow Based on AST-GCN-LSTM

Leveraging Graph Neural Network with LSTM for Traffic Speed Prediction

Adaptive Graph Attention and Long Short-Term Memory-Based Networks for Traffic Prediction

A Novel Attention-Based Dynamic Multi-Graph Spatial-Temporal Graph Neural Network Model for Traffic Prediction

A Dynamic Spatio-Temporal Deep Learning Model for Lane-Level Traffic Prediction

Embracing Large Language Models in Traffic Flow Forecasting

LSTM Network: a Deep Learning Approach for Short-Term Traffic Forecast

Towards Explainable Traffic Flow Prediction with Large Language Models

Large Language Models Powered Context-aware Motion Prediction in Autonomous Driving

MTL-LSTM: Multi-Task Learning-based LSTM for Urban Traffic Flow Forecasting