Recent research in time series forecasting has demonstrated significant advancements using models based on the Transformer architecture. However, most existing studies treat each variable as an independent channel, which fails to effectively capture the dynamic dependencies among multiple variables. This limitation impedes the model's ability to understand complex dynamic relationships in real-world data, thereby reducing predictive accuracy. To address this issue, we propose a novel framework called MTS-CAM (Multi-Time-Scale Convolutional Attention Mechanism). MTS-CAM captures various cycles and trends across different time scales and employs a convolutional attention module to identify key features, while dynamically adjusting the importance of each channel and spatial position within the feature map. This comprehensive approach enables the model to account for dynamic relationships and feature variations among variables. Experimental results on six benchmark time series forecasting datasets validate the effectiveness of the proposed model, showing that it significantly outperforms existing methods and achieves state-of-the-art performance.

Compared to large language models, traditional multimodal sentiment analysis frameworks are constrained by their classification heads, resulting in poor performance on zero-shot tasks. Moreover, due to limitations in visual encoders and multimodal fusion modules, most existing frameworks can only process a small number of images, leading to a loss of visual information. In light of these issues, this paper proposes a new framework, MSACC. This framework enhances the model’s zero-shot performance by adopting a contrastive classification method and reduces the loss of visual information through visual relation extraction and three-dimensional sentiment analysis. We conducted extensive experiments on the Yelp dataset. The experimental results show that MSACC outperforms models of the same category in zero-shot MSA tasks, achieving a 48% performance improvement. Furthermore, compared to the large language model ChatGLM2-6B, MSACC still achieved a 7% performance increase while saving 90% of the model size. In addition, in supervised tasks, MSACC also achieved a 3.27% performance improvement compared to the baseline model.

The combined use of multiple drugs has been widely used in the treatment of many diseases, so accurate prediction of drug-drug interaction (DDI) plays a critical role in ensuring the health of patients' medication and improving the efficiency of drug development. However, for many existing DDI prediction methods (KGNN, TIGER, DANN-DDI, etc.), They fail to fully utilize the potential information in drug background data, limiting the predictive power of unknown drug interactions. Therefore, this paper proposes a drug interaction model based on knowledge graph and drug background data (KGDB-DDI), which can effectively integrate knowledge graph and drug background data and use the integrated data to predict DDI. We evaluated the model on DrugBank and KEGG datasets, and the AUC and AUPR of the KGDB-DDI model on the DrugBank dataset reached 0.9952. The experimental results show that the KGDB-DDI model outperforms the classic and other state-of-the-art models and has excellent prediction performance. In addition, multi-angle ablation studies further demonstrate its effectiveness and its potential in predicting drug interactions.

Transfer learning plays a crucial role in low-resource machine translation by addressing the challenge of poor model performance due to limited data in low-resource languages, thereby improving translation accuracy. Current research methods not only utilize pre-trained parent models for parameter initialization and fine-tuning but also use the soft labels output by these parent models to enhance the consistency between parent and child models. However, even if the parent model performs well, there are still instances where certain token predictions are unstable. During training, if the child model incorporates these unstable token predictions, it can hinder its learning effectiveness; the child model might not fully comprehend the parent model’s prediction strategy, potentially affecting overall translation performance. To address this, we propose a training strategy called Token-Level Filter Training, designed to effectively filter out unstable token predictions from the parent model, thereby transferring the parent model’s positive knowledge to the child model. Additionally, we introduce a hierarchical ranking loss method to help the child model better learn the parent model’s prediction strategies and sequence order, thus enhancing translation accuracy and fluency. Experimental results show that our method outperforms baseline methods on the public datasets Global Voices (Id, Ca, Hu, Pl) and WMT17 (Turkish–English), with BLEU score improvements of 1.47, 0.91, 0.50, 0.54, and 0.55, respectively. These results demonstrate the effectiveness and superiority of the proposed method.