Inferring latent interaction structures from observational time series is a fundamental yet challenging problem in dynamical systems. Existing deep learning methods employ unidirectional information aggregation via incoming edges, failing to identify the mutual dependencies prevalent in real dynamics as well as the feedback effects induced by sampling intervals, which leads to inferential bias. To address this, we propose the D ual- A ttention R elational I nference (DARI), a framework designed to learn latent interaction structures from dynamical observations. DARI employs a coupled bidirectional attention mechanism to model forward and feedback dynamics, effectively decoupling information flow from the underlying interaction structure. Extensive synthetic experiments demonstrate competitive structural recovery performance across diverse graph topologies, including undirected, directed, and weighted graphs. Experiments on COVID-19 data further show that the inferred transmission structures are consistent with real-world population mobility patterns. In addition, the elimination of costly edge-wise computations in DARI leads to substantial gains in both runtime and memory efficiency. Code is available at https://anonymous.4open.science/r/DARI-778C .

Text-attributed graphs have recently garnered significant attention due to their wide range of applications in web domains. Existing methodologies employ word embedding models for acquiring text representations as node features, which are subsequently fed into Graph Neural Networks (GNNs) for training. Recently, the advent of Large Language Models (LLMs) has introduced their powerful capabilities in information retrieval and text generation, which can greatly enhance the text attributes of graph data. Furthermore, the acquisition and labeling of extensive datasets are both costly and time-consuming endeavors. Consequently, few-shot learning has emerged as a crucial problem in the context of graph learning tasks. In order to tackle this challenge, we propose a lightweight paradigm called LLM4NG, which adopts a plug-and-play approach to establish supervision signals by leveraging LLMs for node generation. Specifically, we utilize LLMs to extract semantic information from the labels and generate samples that belong to these categories as exemplars. Subsequently, we employ an edge predictor to capture the structural information inherent in the raw dataset and integrate the newly generated samples into the original graph. This approach harnesses LLMs for enhancing class-level information and seamlessly introduces labeled nodes and edges without modifying the raw dataset, thereby facilitating the node classification task in few-shot scenarios. Extensive experiments demonstrate the outstanding performance of our proposed paradigm, particularly in low-shot scenarios. For instance, in the 1-shot setting of the ogbn-arxiv dataset, LLM4NG achieves a 76% improvement over the baseline model.