Yujiang Liu

The accurate prediction of Remaining Useful Life (RUL) is fundamental to Prognostics and Health Management (PHM), enabling predictive maintenance and ensuring the operational safety of complex industrial systems. While deep learning models have demonstrated significant potential in RUL estimation, existing approaches often struggle with noisy sensor signals, inconsistent predictions across overlapping time windows, and a lack of explicit modeling for the underlying temporal structure of the degradation process. This paper proposes a novel multi-view temporal structure-aware learning framework to address these challenges. The framework introduces a multi-view temporal perturbation mechanism that generates multiple perspectives of the degradation state, coupled with a Transformer-based backbone to capture long-range dependencies. To enhance stability and physical rationality, we design a multi-view consistency regularization term and a temporal ordering constraint learning mechanism. These components ensure that the model produces stable predictions across temporal shifts and adheres to the inherent monotonic degradation patterns. Experimental evaluations conducted on the NASA C-MAPSS dataset demonstrate that the proposed method significantly improves prediction accuracy, stability, and structural consistency compared to state-of-the-art baselines.

When extracting entities, relations, and their associated words from scientific literature, it is imperative to consider the supporting role of feature words on the extraction results. These feature words can provide local semantic information and be combined with the global feature representation of the sentence, improving the accuracy of information extraction. However, existing methods, when fusing local semantic feature words with global features, due to ineffective distinction between the influence of feature words and non-feature words, result in limited enhancement on model performance. To solve this problem, we propose a feature words adversarial scheme (FWAS) with dual pointer method. This method implements a dynamic filtering mechanism for feature words through feature pointers, in order to semantically enhance the encoding of the original text. Simultaneously, an inverse feature pointer is designed to establish a negative weight decay mechanism, weakening interference of non-key vocabulary. During joint training, annotation information for entity relations is introduced to supervise the dual feature selection mechanism. Experimental results on three public scientific information extraction datasets demonstrate that our method consistently outperforms strong baselines, achieving up to 4.9% improvement in F1-score. This method offers a new perspective for information extraction tasks in scientific and technical literature and provides scalable optimization directions for subsequent research.

Extracting relation quintuple and feature words from unstructured text is a prelude to the construction of the scientific knowledge base. At present, the prior works use explicit clues between entities to study this task but ignore the use and the association of the feature words. In this work, we propose a new method to generate self-adaptive feature words from the original text for every single sample. These words can add additional correlation information to the knowledge graph. We allow the model to generate a new word representation and apply it to the original sentence to judge the relation type and locate the head and tail of the relation quintuple. Compared with the previous works, the feature words increase the flexibility of relying on information and improve the explanatory ability. Extensive experiments on scientific field datasets illustrate that the self-adaptive feature words method (SAFW) is good at ferreting out the unique feature words and obtaining the core part for the quintuple. It achieves good performance on four public datasets and obtains a markable performance improvement compared with other baselines.

AI chatbots talk to patients in multi-turn conversations in order to imitate real doctors and meet the patients’ needs during prior medical inquiries. However, existing works ignore the fact that doctors can convey two different tones answers to patients through feedback information, that is, question and statement. This inflexible method urges us to seek a new response strategy that outputs a tone indicator and uses it to limit the range of words output. In this paper, we propose a novel chat model based on question-and-answer juxtaposition (QAJCM), which simultaneously optimize three parts: the basic response, the generated counterpart, and the type of the basic response. These are all produced from the basic response, the real result. In specific, the model outputs a question, a statement, and a type to ensure that the type gives correct choice in the first two parts. Experiments on MedDG and HaoDF datasets show that the BLEU-average scores are up to 20.28 and 3.03, which are 45% and 49% higher than the baselines, respectively. According to the experimental results, we have made obvious improvements and verify that our work can respond to patients in a correct tone in the medical inquiry scenario.