Jiaxing Shang

Multivariate time-series (MTS) forecasting is of critical importance in a wide range of real-world applications, including demand planning and energy load forecasting. However, effective forecasting models typically rely heavily on domain-specific expertise and carefully designed, task-oriented architectures. Although large language models (LLMs) have recently demonstrated remarkable generalization and reasoning capabilities across diverse domains, their direct application to time-series data remains challenging due to the lack of explicit semantic structure in raw numerical sequences. Existing studies have largely failed to systematically activate the intrinsic reasoning capabilities of LLMs, instead treating them primarily as sequence mapping tools. To address these limitations, we propose CMGA: Cross-modal graph alignment for unlocking LLM reasoning in multivariate time series forecasting, this framework that transforms semantically sparse multivariate time series into structured entity–relation graphs, enabling cross-modal translation from numerical signals to LLM-interpretable semantics. Specifically, CMGA constructs a patch graph from multivariate time series to explicitly capture latent spatio-temporal dependencies, and employs a graph neural network to learn structured representations. Furthermore, the model introduces temporal semantic prompting and a cross-modal attention mechanism to align numerical temporal patterns with language-level semantic representations. Extensive experiments on multiple multivariate time series forecasting benchmark datasets validate the effectiveness of CMGA. The proposed framework exhibits strong generalization capability and robustness, particularly in data-scarce and zero-shot settings, validating the effectiveness of integrating graph-structured modeling and semantic alignment to activate LLM reasoning for time-series forecasting.

Fake news detection is a hot topic in the social media mining research community. Recent studies have shown that sentiment signals could significantly benefit the detection performance. However, most existing methods treat sentiment merely as auxiliary features, while the more sophisticated social sentiment interactions were rarely explored. In this paper, we propose a novel framework named ReFEND, which leverages the sentiment resonances among the social users (i.e., social sentiment resonances) and the sentiment relationship between news content and user comments to improve the detection performance. Specifically, we first utilize sentiment scorers to assess the sentiment of comments and identify users’ emotional tendencies. Then we creatively construct a sentiment-aware multi-relational graph to capture social sentiment resonances evoked by the content and the interactions between comments and news. Next, we leverage the relational graph convolutional network (RGCN), which specializes in handling multi-relational graph data, to learn the interactions on sentiment-aware graph. To our best knowledge, this is the first effort to leverage social sentiment resonances for fake news detection. Experimental results on three datasets indicate that ReFEND significantly outperforms the state-of-the-art sentiment-based methods in terms of F1 and accuracy. Besides, ablation studies demonstrate the effectiveness of components designed in ReFEND.

In recent years, the prevalence of modern flight data sensing systems, represented by Quick Access Recorder (QAR), has made it viable to improve flight safety by analyzing massive flight parameters. However, existing studies predominantly concentrate on predefined safety events where the expert-annotated data are highly scarce and imbalanced. In this paper, we address this issue by uncovering abnormal flight patterns from tremendous unlabeled data to discover potential and unexpected safety risks. To this end, we propose DUVET, a reconstruction-based DUal View Enhanced Transformer model for flight data anomaly detection, which plays a vital role in aviation safety monitoring. Specifically, DUVET adopts patching and channel independence techniques to reduce model parameters and improve Transformer’s computational efficiency while preserving its semantic representation capacity. Based on the observation that flight safety risks usually relate to both long-term trend and short-term patterns, we creatively propose Flight Trend-Aware Attention and Bell attention mechanisms to capture both global and local dependencies, enhancing the model reconstruction ability. We evaluate DUVET on six public real-world datasets and one QAR dataset, demonstrating that DUVET achieves state-of-the-art performance in general time series anomaly detection task. Further experiments on the QAR dataset show that DUVET meets key criteria of anomaly detection for flight safety in terms of stability, training efficiency and detection accuracy. Case study shows that DUVET can effectively uncover hidden flight safety risks.

Graph neural network (GNN)-based approaches have achieved remarkable success in temporal knowledge graph (TKG) reasoning. Despite these advances, two critical challenges remain: 1) inadequate modeling of local contextual dynamics, which limits the adaptability of entity and relation representations to specific queries and 2) inadequate mechanisms for handling emerging patterns, that is, novel interactions absent from historical data, which reduces predictive performance in dynamic environments. To address these limitations, we propose TCDR–PD, a temporal and contextual dynamic representation network with pattern decomposition. TCDR–PD introduces a temporal and contextual dynamic representation learning (TCDR) module to capture both global temporal trends and query-specific contextual dynamics, enabling more precise embeddings. Additionally, the pattern decomposition (PD) prediction module explicitly disentangles the prediction of recurring and emerging patterns, enabling tailored strategies to improve reasoning performance. Experiments on four benchmark datasets demonstrate that TCDR–PD outperforms state-of-the-art methods, effectively supporting stable reasoning over evolving TKGs.

Flight safety remains a central concern in civil aviation. Recently, increasing attention has been given to leveraging high-dimensional temporal flight data, typically collected by Quick Access Recorders (QAR), to predict safety events. However, most existing studies focus on individual events, overlooking the latent correlations among multiple events. For instance, during landing, a maneuver that reduces the risk of one event may inadvertently increase the risk of another. Predicting multiple events introduces three main challenges: 1)handling parameters recorded at inconsistent sampling rates, 2) learning task-specific parameter importance for interpretability, and 3) modeling complex temporal dependencies for accuracy. To address these, we propose MultiSafe, a deep multi-task learning model for predicting and interpreting multiple flight safety events. First, to process parameters with heterogeneous frequencies, we introduce a Multi-Scale Shared Encoder with adaptive convolutional kernels to unify representations across tasks. Next, a gating-based Parameter Selector learns task-specific parameter importance, enabling interpretable predictions. Finally, a Temporal Decoder with fine-grained attention captures intricate temporal dependencies among parameters. Experiments on a dataset of 37,518 A320 flight records demonstrate that MultiSafe outperforms state-of-the-art baselines in prediction accuracy, while its interpretability offers actionable insights to assist pilots in enhancing flight safety.