Predicting motions of surrounding vehicles is critically important to help autonomous driving systems plan a safe path and avoid collisions. Although recent social pooling based LSTM models have achieved significant performance gains by considering the motion interactions between vehicles close to each other, vehicle trajectory prediction still remains as a challenging research issue due to the dynamic and high-order interactions in the real complex driving scenarios. To this end, we propose a wave superposition inspired social pooling (Wave-pooling for short) method for dynamically aggregating the high-order interactions from both local and global neighbor vehicles. Through modeling each vehicle as a wave with the amplitude and phase, Wave-pooling can more effectively represent the dynamic motion states of vehicles and capture their high-order dynamic interactions by wave superposition. By integrating Wave-pooling, an encoder-decoder based learning framework named WSiP is also proposed. Extensive experiments conducted on two public highway datasets NGSIM and highD verify the effectiveness of WSiP by comparison with current state-of-the-art baselines. More importantly, the result of WSiP is more interpretable as the interaction strength between vehicles can be intuitively reflected by their phase difference. The code of the work is publicly available at https://github.com/Chopin0123/WSiP.

Text-Attributed Graphs (TAGs) are prevalent in various real-world scenarios, where each node is associated with a text attribute. Representation learning on TAGs relies on a comprehensive understanding of both the textual attributes associated with nodes and the topological connections between nodes. Recent works have tried to enhance graph neural networks (GNNs) with pre-trained language models (PLMs) for textual attribute learning, which has achieved promising result compared with shallow text representation learning models such as BoW. With the advent of more powerful LLMs, it is worth to further exploring how to effectively and efficiently incorporate the generic knowledge in LLMs to enhance GNNs. To this aim, we for the first time investigate whether the mainstream generative LLMs can provide high-quality generic text embedding for GNNs. To simultaneously use the generic knowledge in the LLM and the task-specific knowledge in the tuned PLM (TLM), we propose a dual channel knowledge fusion framework, named MoKGNN, to enhance the representation learning on TAGs. Specifically, we first propose a dual channel feature extraction stage to derive text embeddings that contain generic knowledge and task-specific knowledge separately. Next, we introduce a Knowledge Alignment Network to align the higher-dimensional General Embedding with the lower-dimensional Task-Specific Embedding. Finally, we adopt a mix-up operation to integrate the generic and task-specific textual attribute representation to enhance downstream GNNs. Extensive experiments on four TAG datasets demonstrate the effectiveness of our approach in improving the performance of GNNs in the tasks of node classification and link prediction.