Applied Soft Computing最新文献

{"title":"A sarcasm detection method based on modality inconsistencies and textual knowledge enhancement","authors":"Yuxin Han,&nbsp;Runtao Yang,&nbsp;Mingyu Zhu,&nbsp;Lina Zhang","doi":"10.1016/j.asoc.2025.113225","DOIUrl":"10.1016/j.asoc.2025.113225","url":null,"abstract":"<div><div>Sarcasm detection aims to identify emotional tendencies in tweets, which helps governments and enterprises monitor online public opinions. The Twitter platform can create messages, including images and texts. Existing sarcasm detection methods mainly focus on extracting high-level semantic information from images while ignoring textual information. However, previous research has demonstrated that text is more important than images in sentiment analysis tasks. Inspired by this, we reduce the involvement of image information and investigate the sarcasm detection from a textual perspective. First, we divide the text in the primary dataset into pure text and hashtags. The hashtags are fused with high-frequency words in the pure text. Then, considering the differences on the data distribution between the training corpus of Bidirectional Encoder Representation from Transformers (BERT) and the sarcasm detection corpus, we use the Twitter sentiment analysis corpus to further pre-train the BERT model, obtaining the Basic_BERT and Hash_BERT models as feature extractors for the pure text and hashtags. Furthermore, to better play the role of the text in this task, a cross-gate mechanism method is proposed by a cross-attention transformer module and a similarity constraint. The cross-attention transformer module is used to generate a representation of intra-modal and inter-modal fusion while the similarity constraint is used to achieve a balance between the original modal representation and the fused modal representation. On the sarcasm detection dataset, the proposed model achieves an F1-score of 87.22%, an improvement of 3.30% over the most advanced model.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113225"},"PeriodicalIF":7.2,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143947774","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Few-shot outliers classification in high-speed railway track geometry based on self-supervised transformer","authors":"Yu Liu,&nbsp;Jinzhao Liu,&nbsp;Wenxuan Zhang,&nbsp;Sen Yang,&nbsp;Kai Tao,&nbsp;Fei Yang","doi":"10.1016/j.asoc.2025.113281","DOIUrl":"10.1016/j.asoc.2025.113281","url":null,"abstract":"<div><div>External disturbances, data transmission, sensor signal offsets and weather conditions are common sources of outliers in track geometry inspection data. The infrequent occurrence of these outliers leads to a scarcity of labeled samples, making high-accuracy few-shot classification challenging with traditional supervised learning methods. To address this, we propose a method for the few-shot classification of outliers in high-speed railway track geometry inspection data based on a self-supervised transformer. First, the self-supervised transformer pre-trains on a substantial amount of unlabeled data, enabling the model to learn and extract fundamental features and patterns from the inspection data. Next, the limited labeled outliers are used to fine-tune the model, enhancing its adaptability to the task of classifying outliers. This approach allows for the automatic identification and classification of outliers even with limited labeled data and without prior knowledge. Experimental results demonstrate that the proposed method accurately identifies and classifies outliers such as local burr, turnout gauge widening, constant section of unilateral gauge data, and abnormal distribution in track geometry inspection data, achieving a classification accuracy of up to 97.8 % and an F1-score as high as 97.9 %. This performance surpasses five supervised baselines by 4–26 % in accuracy and by 5–35 % in F1-score. Moreover, the method maintains an accuracy rate exceeding 92 % across different inspection trains and lines, demonstrating excellent generalization performance.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"178 ","pages":"Article 113281"},"PeriodicalIF":7.2,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144090720","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Knowledge-informed multiplication convolution generalization network for interpretable equipment diagnosis under unknown speed domains","authors":"Rui Liu ,&nbsp;Xiaoxi Ding ,&nbsp;Benyuan Ye ,&nbsp;Yuanyuan Xu ,&nbsp;Jiahai Huang ,&nbsp;Hongyu Lv","doi":"10.1016/j.asoc.2025.113263","DOIUrl":"10.1016/j.asoc.2025.113263","url":null,"abstract":"<div><div>The generalization performance and interpretability of intelligent fault diagnosis methods under unknown speed domains are crucial concerns in real industry practice. However, existing solutions seldom address both issues simultaneously, restricting their development prospects. Motivated by these challenges, this study puts forward a signal-processing-collaborated deep learning architecture—knowledge-informed multiplication convolution generalization network (KI-MCGN), which is composed of three layers, called adaptive mode capturer (AMCer), prior knowledge pooler (PKPer) and classifier. Informed by the mode response characteristics of fault vibration signals, AMCer first tailors several speed-fused multiplication filtering kernels (SF-MFKs) for adaptive mining of fault-related modes. To improve the generalization capability, the center frequency and bandwidth coefficient of SF-MFKs are no longer defined directly, but are innovatively fitted by multiple trainable coefficients with regard to the speed information. This novel speed fusion strategy allows SF-MFK to not only learn the mapping relationship between the speed information and the distribution of fault-included modes, but also to autonomously adjust its modal filtering scale in unknown speed domains. In light of the excellent comprehensibility of prior indicators in characterizing the health status of equipment, a novel pooler named PKPer is presented subsequently. It pools each extracted mode into 12 frequency-domain modal prior indicators (MPIs). Eventually, two dense layers are adopted as the classifier to output the ultimate decision. In particular, considering the distribution difference of mode features across different speed domains, local-domain generalization is further integrated to assist the model extract generalized features. The comparison results from two experimental cases demonstrate that the proposed KI-MCGN architecture outperforms the other eight state-of-the-art approaches and three ablation models. Meanwhile, comprehensive visualization analysis not only validates the modal filtering potency of SF-MFKs under unknown speed domains, but also explores the guiding meaning of MPIs for the final diagnosis. It can be also foreseen that the proposed KI-MCGN framework is expected to provide reliable and explainable intelligent decision-making for equipment maintenance under unknown speed domains.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"179 ","pages":"Article 113263"},"PeriodicalIF":7.2,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144178412","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Integrating large language models with cross-modal data fusion for advanced intelligent transportation systems in sustainable cities development","authors":"Jiachen Jiang ,&nbsp;Yang Li ,&nbsp;Jing Nie ,&nbsp;Jingbin Li ,&nbsp;Baoqin Wen ,&nbsp;Thippa Reddy Gadekallu","doi":"10.1016/j.asoc.2025.113278","DOIUrl":"10.1016/j.asoc.2025.113278","url":null,"abstract":"<div><div>With the advent of the 6 G network era, autonomous navigation systems, as a key driving force for sustainable cities innovation and development, have raised higher demands for navigation and perception technologies. Against this backdrop, this paper proposes a multi-level cross-modal fusion method based on large language models (LLMs) for environmental perception in autonomous navigation systems. Specifically, the system first utilizes Transformer and PointNet models to extract and encode features from multimodal data obtained from cameras and LiDAR sensors. Subsequently, the cross-modal multi-head self-attention mechanism enhances the information exchange between different data types, enabling richer feature representations. Finally, a Mixture of Experts (MoE) model is deployed as a secondary fusion module, and a pre-fetching strategy is introduced to reduce the communication latency of the model. Experimental results indicate that this method outperforms other state-of-the-art 3D object detection algorithms on the KITTI dataset. Specifically, when detecting \"car\" targets, the mAP value of this method exceeds that of the next best method, Fast-CLOCs, by 2.08 %. When detecting \"pedestrian\" targets, the mAP exceeds that of the next best method, Epnet+ +, by 1.88 %. The proposed method also demonstrates a higher convergence speed and greater throughput than other strategies in the comparison of expert pre-fetching strategies. For example, in the case of the Transformer model, the EPP strategy reduces the time required to achieve target accuracy by 40 %, 17 %, and 25 % compared to Tutel, Lina, and Janus, respectively. Additionally, the MCF model shows robustness in handling data from sensors with varying data collection frequencies and timestamps, and it performs well across different weather conditions and lighting scenarios.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113278"},"PeriodicalIF":7.2,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068239","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Question answering over temporal knowledge graphs based on time sensitive graph neural network","authors":"Luyi Bai,&nbsp;Linshuo Xu,&nbsp;Lin Zhu","doi":"10.1016/j.asoc.2025.113257","DOIUrl":"10.1016/j.asoc.2025.113257","url":null,"abstract":"<div><div>Answering query questions accurately on large-scale knowledge graph has always been the key of question answering system. Recently, answering temporal questions on temporal knowledge graphs has attracted wide attention. However, all rely on predefined temporal knowledge or structured data, which often limits their ability to generalize to more dynamic or unstructured temporal information. These models can also struggle to deal with ambiguous or complex temporal reasoning, especially when dealing with long or overlapping time frames. Therefore, we propose a new <strong>T</strong>ime <strong>S</strong>ensitive <strong>G</strong>raph <strong>N</strong>eural <strong>N</strong>etwork <strong>Q</strong>uestion <strong>A</strong>nswering model (TSGNN-QA), which not only addresses the limitations of existing knowledge graph QA models in learning and representing temporal information, but also enhances the ability to infer and predict implicit temporal information and improve multi-hop node connectivity and node ability, providing a new solution for temporal knowledge graph QA tasks. On the one hand, we use temporal graph neural network to learn temporal knowledge graph, and improve its time encoding part by hyperplane technology to improve the time sensitivity of the model. On the other hand, we use gated recurrent unit to reason and predict the hidden temporal information, so as to alleviate the problem of query accuracy degradation caused by lack of temporal information. In addition, in view of the fact that some question entities and answer entities are not directly connected in the temporal question answering process, it may take multiple entities to find the answer entity. This paper establishes and increases the multi-hop connection between nodes and carries out reasoning scoring, which improves the multi-hop query ability of the model. In the query module of TSGNN-QA model, we optimize and decodes the multi-head attention mechanism and multi-layer perceptron. Finally, experimental results show that the TSGNN-QA model has significant advantages in temporal question-answering queries. Its experimental results for Hits@1 on the CronQuestions dataset and Complex-CronQuestions dataset outperform the best baseline model by 12.03 % and 1.52 %, respectively.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113257"},"PeriodicalIF":7.2,"publicationDate":"2025-05-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143941730","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Improving the performance of fault diagnosis network via interpretable projection information","authors":"Biao He ,&nbsp;Pengfei Dong ,&nbsp;Yi Qin","doi":"10.1016/j.asoc.2025.113284","DOIUrl":"10.1016/j.asoc.2025.113284","url":null,"abstract":"<div><div>The performance of fault diagnosis networks for rotating machine mainly depends on the classifiers and feature extractors. To improve the performance of the two components, a novel regularization and a modified normalization module are proposed based on the interpretable projection information. Specifically, the classifier is firstly studied from the perspective of projection mechanism rather than the traditional base representation. It is found that a negative correlation is beneficial to classifiers, and a corresponding regularization is proposed for improving its ability. Meanwhile, by analyzing the convolution operations in convolution networks from the projection perspective, we find that the projection information will be affected by the traditional batch normalization block followed with a rectified linear unit, if the batch size is huge and the number of classes is small. To solve this problem, a novel normalization module, which is designed based on absolute value operation, is proposed to fully retain the projection information while effectively suppressing the noises in features. Finally, the proposed method is used to improve several typical fault diagnostic networks, and the fault diagnosis experiments on rolling bearings and planetary gearboxes demonstrate that the proposed method can effectively improve the performance of diagnosis networks.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113284"},"PeriodicalIF":7.2,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068240","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Intrusion detection in IoT and wireless networks using image-based neural network classification","authors":"Yanxia Sun ,&nbsp;Zenghui Wang","doi":"10.1016/j.asoc.2025.113236","DOIUrl":"10.1016/j.asoc.2025.113236","url":null,"abstract":"<div><div>Telecommunication networks play more and more important role in our modern times, and there are significant security risks associated with both wireless and wired networks. These risks stem from various malicious actions and security threats that have emerged with the development of Fourth Generation (4G), Fifth Generation (5G), and Internet of Things (IoT) networks. Machine learning (ML) algorithms have been applied to Intrusion Detection Systems (IDSs) due to their capacity to their ability to detect complex network traffic patterns. Deep learning (DL) networks are highly effective in processing images and videos and they have potential to solve other types of data. Given the characteristics of network traffic records used for intrusion detection in wireless and wired networks, we propose a simple data preprocessing method to convert the data into a grid-structured format, making it compatible with image processing networks. To validate the proposed structure, modified LeNet networks have been used for intrusion detection based on the NSL-KDD and CICIoV2024 (Canadian Institute for Cybersecurity Internet of Vehicles 2024 dataset) benchmark datasets. The simulation results indicate that methods based on extracted features may not always guarantee improved performance. The proposed Image Classification Neural Network-based Intrusion Detection (ICNN-ID) outperforms the compared existing methods. The multiclass classification experimental results show that the proposed LeNet-based IDS achieved a test accuracy (TAC) of 89.97% for NSL-KDD and nearly 100% (99.996%) for CICIoV2024. Additionally, it offers higher accuracy and improved robustness compared to a one-dimensional CNN and a recent deep learning model that integrates deep convolutional neural networks (DCNN) and bidirectional long short-term memory (BiLSTM).</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113236"},"PeriodicalIF":7.2,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143941727","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AdvancedScoreCAM: Enhancing visual explainability through hierarchical upsampling","authors":"HaoJun Zhao,&nbsp;Mohd Halim Mohd Noor","doi":"10.1016/j.asoc.2025.113265","DOIUrl":"10.1016/j.asoc.2025.113265","url":null,"abstract":"<div><div>Deep learning models have achieved remarkable success across various domains. However, the intricate nature of these models often hinders our understanding of their decision-making processes. Explainable AI methods such as Class Activation Mapping (CAM) become indispensable in providing intuitive explanations for these model decisions. Previous CAM-based methods often employed simple upsampling operations, resulting in the loss of contextual information. In this work, we propose a simple yet highly effective approach, AdvancedScoreCAM (ASC), which introduces a concurrent upsampling and fusion pipeline method to enhance visual explainability. Our proposed method introduces a direct and progressive upsampling pipeline, which can fully extracts contextual information during the upsampling process. This improvement is achieved by selectively integrating contextual details within the upsampled activation layers. Through extensive experiments and qualitative comparisons on two datasets, we demonstrate that ASC consistently produces clearer and more interpretable heatmaps that better reflect the model’s decision-making process compared to previous methods. Our code is available at <span><span>https://github.com/jiiaozi/AdvancedScoreCAM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113265"},"PeriodicalIF":7.2,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144084376","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Mutable hierarchy feature selection based on generalized fuzzy rough sets","authors":"Zilong Lin ,&nbsp;Yaojin Lin ,&nbsp;Chenxi Wang ,&nbsp;Jinkun Chen","doi":"10.1016/j.asoc.2025.113233","DOIUrl":"10.1016/j.asoc.2025.113233","url":null,"abstract":"<div><div>Hierarchical classification divides data into correlated sub-tasks from coarse to fine. Compared to flat classification, it is more complex and suffers from the curse of dimensionality. Existing hierarchical Fuzzy Rough Sets (FRS) methods only stay at the fine-grained to select the features, which is a fine-grained search strategy. Thus, we propose a coarse-grained search strategy and use Generalized Fuzzy Rough Sets (GFRS) to enhance its robustness. Furthermore, we introduce the concept of tree fragmentation. Though assessing the degree of fragmentation in the tree structure, it selects an appropriate granularity search strategy for hierarchical tree-structured datasets. Finally, we combine both granularity search strategies and collectively named - Mutable Hierarchy Search Strategy (MHSS); the entire proposed algorithm is named - Mutable Hierarchy Feature Selection Based on Generalized Fuzzy Rough Sets (MHFS). We compare two FRS-based feature selection algorithms, three hierarchical optimization methods, and six flat feature selection methods. Extensive experiments demonstrate the performance of our method.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113233"},"PeriodicalIF":7.2,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143947779","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An adaptive multi-factor integrated forecasting model based on periodic reconstruction and random forest for carbon price","authors":"Shunyu Zhao ,&nbsp;Yelin Wang ,&nbsp;Jianwei Deng ,&nbsp;Zheng Li ,&nbsp;Gen Deng ,&nbsp;Zhi Chen ,&nbsp;Youjie Li","doi":"10.1016/j.asoc.2025.113274","DOIUrl":"10.1016/j.asoc.2025.113274","url":null,"abstract":"<div><div>Accurate and appropriate carbon price prediction can provide a quantitative benchmark for the decision-making of government and investors, promoting the rapid development of carbon market. However, the inherently complexity of carbon price affected by multiple external factors poses a challenge for accurate forecasting. Thus, an adaptive multi-factor integrated hybrid model based on periodic reconstruction and random forest is developed for carbon price prediction. In the model, the improved decomposition method and periodic reconstruction are introduced to fully extract and integrate the hidden laws, which realizes the efficient and accurate prediction under multiple time scales. Considering the disparities of carbon markets, a three-stage influencing factors screening framework is proposed based on random forest, achieving the adaptive prediction by using the selected external factors to modify the forecasting of carbon prices. Four representative carbon markets in China (i.e., Shanghai, Guangdong, Shenzhen, and Hubei) are employed for empirical analysis. The results reveal that carbon price can be affected by energy and financial markets in short-term fluctuations, while its long-term trends are mainly influenced by climate and policy effects. Compared with other benchmark models, the proposed adaptive model considering multiple factors is reasonable and effective to predict carbon price with different characteristics that the average MAPE and RMSE are 0.3977 and 0.5036, respectively. Therefore, the proposed model not only provides a reliable tool for carbon price prediction, but also provides a unique perspective for governments and investors to explore the multi-time scale influencing factors of carbon price variations, which helps stakeholders understand the market rules and make appropriate decisions.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"177 ","pages":"Article 113274"},"PeriodicalIF":7.2,"publicationDate":"2025-05-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144088692","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}