Applied Soft Computing最新文献_第9页

Evolving granular fuzzy control: Overview, case study on the chaotic Hénon map, and research outlook 演化颗粒模糊控制：综述、混沌hsamnon图的个案研究及研究展望

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114639

Daniel Leite , Reinaldo Palhares , Igor Škrjanc , Fernando Gomide

This paper highlights the relevance of evolving granular fuzzy systems in adaptive control and fuzzy modeling, particularly for learning in dynamic, nonstationary environments. These systems incrementally construct rule-based models—such as predictors and controllers operating in open- or closed-loop configurations—by adapting both structure and parameters from data streams. This provides a flexible and autonomous alternative to traditional parametric-adaptive approaches. We consolidate foundational concepts in fuzzy and adaptive control, positioning evolving systems as data-driven extensions of classical schemes. Key challenges are discussed, including safety-aware adaptation to drift, memory mechanisms, interpretability, and principled structural evolution. Building on these foundations, we develop a more mature formulation of the state-space evolving granular modeling and control framework (SS-EGM/SS-EGC), introducing a decay-rate–oriented treatment that advances the methodology beyond mere LMI feasibility toward online optimality. A compact case study on the chaotic Hénon map illustrates the approach: an online SS-EGM learned from data streams supports SS-EGC synthesis that stabilizes the map under bounded inputs. One-step prediction accuracy and decay-rate estimates confirm real-time viability. The framework provides a flexible basis that can be further extended in multiple directions to address the identified challenges.

本文强调了进化颗粒模糊系统在自适应控制和模糊建模中的相关性，特别是在动态，非平稳环境中的学习。这些系统通过调整数据流的结构和参数，逐步构建基于规则的模型——比如在开环或闭环配置中运行的预测器和控制器。这为传统的参数自适应方法提供了灵活和自主的替代方案。我们巩固了模糊和自适应控制的基本概念，将进化系统定位为经典方案的数据驱动扩展。讨论了主要挑战，包括对漂移的安全意识适应，记忆机制，可解释性和原则性结构进化。在这些基础上，我们开发了一种更成熟的状态空间演化颗粒建模和控制框架（SS-EGM/SS-EGC），引入了一种面向衰减率的处理方法，将该方法从单纯的LMI可行性推进到在线最优性。一个关于混沌hsamnon映射的紧凑案例研究说明了这种方法：从数据流中学习的在线SS-EGM支持SS-EGC合成，在有界输入下稳定映射。一步预测精度和衰减率估计确认了实时可行性。该框架提供了一个灵活的基础，可以在多个方向进一步扩展，以应对已确定的挑战。

{"title":"Evolving granular fuzzy control: Overview, case study on the chaotic Hénon map, and research outlook","authors":"Daniel Leite , Reinaldo Palhares , Igor Škrjanc , Fernando Gomide","doi":"10.1016/j.asoc.2026.114639","DOIUrl":"10.1016/j.asoc.2026.114639","url":null,"abstract":"<div><div>This paper highlights the relevance of evolving granular fuzzy systems in adaptive control and fuzzy modeling, particularly for learning in dynamic, nonstationary environments. These systems incrementally construct rule-based models—such as predictors and controllers operating in open- or closed-loop configurations—by adapting both structure and parameters from data streams. This provides a flexible and autonomous alternative to traditional parametric-adaptive approaches. We consolidate foundational concepts in fuzzy and adaptive control, positioning evolving systems as data-driven extensions of classical schemes. Key challenges are discussed, including safety-aware adaptation to drift, memory mechanisms, interpretability, and principled structural evolution. Building on these foundations, we develop a more mature formulation of the state-space evolving granular modeling and control framework (SS-EGM/SS-EGC), introducing a decay-rate–oriented treatment that advances the methodology beyond mere LMI feasibility toward online optimality. A compact case study on the chaotic Hénon map illustrates the approach: an online SS-EGM learned from data streams supports SS-EGC synthesis that stabilizes the map under bounded inputs. One-step prediction accuracy and decay-rate estimates confirm real-time viability. The framework provides a flexible basis that can be further extended in multiple directions to address the identified challenges.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114639"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979797","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}

引用次数: 0

PMNet: A progressive MLP-based framework for time series forecasting with long-short term dependency synergy PMNet：一个渐进的基于mlp的框架，用于具有长短期依赖协同作用的时间序列预测

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114635

Shijun Chen , Hua Wang , Fan Zhang

Time series forecasting has wide-ranging applications across various domains. In recent years, MLP-based forecasting models have attracted increasing attention. However, their ability to capture complex temporal patterns remains limited. To address this, we propose PMNet—a progressive sequence modeling framework based on MLPs—from the perspective of collaborative modeling of short-term and long-term dependencies. In time series, short-term dependencies carry crucial fluctuation signals and instantaneous variations, while long-term dependencies reflect broader trend evolutions. The organic integration of both plays a vital role in improving predictive accuracy. To this end, we introduce the Multi-source Progressive Coupling Block (MPC), which progressively enhances the model’s ability to capture short-term dependencies and local dynamics through block-wise multi-source feature fusion and information propagation. Meanwhile, we employ the Adaptive Structural Perception Block (ASP) to introduce a dual-domain guidance mechanism equipped with stochastic selectivity and differentiation, enabling structural enhancement of global temporal patterns. Working in synergy, these two modules unify fine-grained short-term modeling with long-range global structure representation, significantly improving the model’s capacity for deep temporal pattern understanding. Extensive experiments validate the effectiveness of the proposed approach.

时间序列预测在各个领域有着广泛的应用。近年来，基于mlp的预测模型越来越受到人们的关注。然而，它们捕捉复杂时间模式的能力仍然有限。为了解决这个问题，我们从短期和长期依赖关系协同建模的角度，提出了基于mlp的渐进式序列建模框架pmnet。在时间序列中，短期依赖关系携带关键的波动信号和瞬时变化，而长期依赖关系反映更广泛的趋势演变。两者的有机结合对提高预测精度起着至关重要的作用。为此，我们引入了多源渐进耦合块（MPC），通过分块多源特征融合和信息传播，逐步增强模型捕捉短期依赖关系和局部动态的能力。同时，我们利用自适应结构感知块（ASP）引入了一种具有随机选择性和差异性的双域引导机制，实现了全局时间格局的结构增强。这两个模块协同工作，将细粒度短期建模与长期全局结构表示统一起来，显著提高了模型对深度时间模式理解的能力。大量的实验验证了该方法的有效性。

{"title":"PMNet: A progressive MLP-based framework for time series forecasting with long-short term dependency synergy","authors":"Shijun Chen , Hua Wang , Fan Zhang","doi":"10.1016/j.asoc.2026.114635","DOIUrl":"10.1016/j.asoc.2026.114635","url":null,"abstract":"<div><div>Time series forecasting has wide-ranging applications across various domains. In recent years, MLP-based forecasting models have attracted increasing attention. However, their ability to capture complex temporal patterns remains limited. To address this, we propose PMNet—a progressive sequence modeling framework based on MLPs—from the perspective of collaborative modeling of short-term and long-term dependencies. In time series, short-term dependencies carry crucial fluctuation signals and instantaneous variations, while long-term dependencies reflect broader trend evolutions. The organic integration of both plays a vital role in improving predictive accuracy. To this end, we introduce the Multi-source Progressive Coupling Block (MPC), which progressively enhances the model’s ability to capture short-term dependencies and local dynamics through block-wise multi-source feature fusion and information propagation. Meanwhile, we employ the Adaptive Structural Perception Block (ASP) to introduce a dual-domain guidance mechanism equipped with stochastic selectivity and differentiation, enabling structural enhancement of global temporal patterns. Working in synergy, these two modules unify fine-grained short-term modeling with long-range global structure representation, significantly improving the model’s capacity for deep temporal pattern understanding. Extensive experiments validate the effectiveness of the proposed approach.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114635"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980369","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}

引用次数: 0

Advanced generative artificial intelligence and deep learning for long-term container throughput forecasting under stochastic disruptions 基于高级生成人工智能和深度学习的随机干扰下集装箱长期吞吐量预测

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114610

Ngoc Cuong Truong , Duy Anh Nguyen

In the face of interruptions that can majorly impact global supply chains, strategic planning and operational efficiency depend on accurate long-term forecasting of container throughput. This paper presents an advanced forecasting system that incorporates nonlinear wavelet decomposition to break down complicated time series data into easier-to-understand parts. These components are then predicted through the use of a hybrid model that blends deep learning methods, particularly the Stacked Long Short-Term Memory (Stacked LSTM) network, with the Wasserstein Generative Adversarial Network with Gradient Penalty (Wasserstein GAN-GP). Using this hybrid technique, the model can better forecast disruption-induced changes in container throughput and capture nonlinear patterns. A case study is conducted at Busan Port, one of the world’s busiest container ports, comprising five major container terminals. The proposed hybrid model achieves forecasting accuracy of 91.09 % for the most complex 40FT container dataset, compared to less than 85 % for traditional methods. For 20FT and 10FT datasets, the model maintains accuracy above 90 %, consistently outperforming baseline approaches. The proposed model is evaluated against conventional forecasting methods using the Long Short-Term Memory (LSTM), Stacked Long Short-Term Memory (Stacked LSTM), and other hybrid models as benchmarks. Across all datasets, the hybrid approach consistently records the lowest MAE, MSE, and RMSE values, validating its robustness under growing data complexity and stochastic disruptions. The study's practical ramifications demonstrate how the suggested approach could assist port officials, terminal operators, and logistics stakeholders make data-driven choices amidst volatile global trade conditions.

面对可能严重影响全球供应链的中断，战略规划和运营效率取决于对集装箱吞吐量的准确长期预测。本文提出了一种先进的预测系统，该系统采用非线性小波分解将复杂的时间序列数据分解成更容易理解的部分。然后，通过使用混合模型来预测这些组件，该模型混合了深度学习方法，特别是堆叠长短期记忆（堆叠LSTM）网络，以及带有梯度惩罚的Wasserstein生成对抗网络（Wasserstein GAN-GP）。使用这种混合技术，该模型可以更好地预测中断引起的集装箱吞吐量变化，并捕获非线性模式。以釜山港为例进行了研究。釜山港是世界上最繁忙的集装箱港口之一，由五个主要集装箱码头组成。对于最复杂的40FT容器数据集，所提出的混合模型的预测精度为91.09 %，而传统方法的预测精度低于85 %。对于20FT和10FT数据集，该模型的准确率保持在90% %以上，始终优于基线方法。采用长短期记忆（LSTM）、堆叠长短期记忆（堆叠LSTM）和其他混合模型作为基准，对该模型进行了对比。在所有数据集中，混合方法始终记录最低的MAE、MSE和RMSE值，验证了其在不断增长的数据复杂性和随机中断下的稳健性。该研究的实际影响表明，建议的方法如何帮助港口官员、码头运营商和物流利益相关者在动荡的全球贸易条件下做出数据驱动的选择。

{"title":"Advanced generative artificial intelligence and deep learning for long-term container throughput forecasting under stochastic disruptions","authors":"Ngoc Cuong Truong , Duy Anh Nguyen","doi":"10.1016/j.asoc.2026.114610","DOIUrl":"10.1016/j.asoc.2026.114610","url":null,"abstract":"<div><div>In the face of interruptions that can majorly impact global supply chains, strategic planning and operational efficiency depend on accurate long-term forecasting of container throughput. This paper presents an advanced forecasting system that incorporates nonlinear wavelet decomposition to break down complicated time series data into easier-to-understand parts. These components are then predicted through the use of a hybrid model that blends deep learning methods, particularly the Stacked Long Short-Term Memory (Stacked LSTM) network, with the Wasserstein Generative Adversarial Network with Gradient Penalty (Wasserstein GAN-GP). Using this hybrid technique, the model can better forecast disruption-induced changes in container throughput and capture nonlinear patterns. A case study is conducted at Busan Port, one of the world’s busiest container ports, comprising five major container terminals. The proposed hybrid model achieves forecasting accuracy of 91.09 % for the most complex 40FT container dataset, compared to less than 85 % for traditional methods. For 20FT and 10FT datasets, the model maintains accuracy above 90 %, consistently outperforming baseline approaches. The proposed model is evaluated against conventional forecasting methods using the Long Short-Term Memory (LSTM), Stacked Long Short-Term Memory (Stacked LSTM), and other hybrid models as benchmarks. Across all datasets, the hybrid approach consistently records the lowest MAE, MSE, and RMSE values, validating its robustness under growing data complexity and stochastic disruptions. The study's practical ramifications demonstrate how the suggested approach could assist port officials, terminal operators, and logistics stakeholders make data-driven choices amidst volatile global trade conditions.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114610"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980376","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}

引用次数: 0

Differential-game algorithm for resource optimization of wireless sensor network 无线传感器网络资源优化的微分博弈算法

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114641

Ning Yao , Dong Liu , Bai Chen , Xiaochen Hao , Rongrong Yin

In response to the resource competition optimization problem in wireless sensor network, this paper first uses differential evolution algorithm to transform single-object multi-objective optimization into multi-object single-objective optimization, and combines game theory to design a new differential game algorithm. The algorithm conforms to the differential evolution framework on the outer layer and meets the requirements of game theory on the inner layer. And theoretically proved the low complexity and algorithm convergence. Then, the designed differential game algorithm was applied to the wireless sensor network to design the channel allocation algorithm. The convergence and effectiveness of the algorithm were verified through simulation. Finally, the differential game algorithm was discussed, and it was theoretically proved that the algorithm is also applicable to multi-objective optimization of multiple objects, providing a design idea for low complexity convergence multi-objective optimization algorithms.

针对无线传感器网络中的资源竞争优化问题，本文首先利用差分进化算法将单目标多目标优化转化为多目标单目标优化，并结合博弈论设计了一种新的差分博弈算法。该算法在外层符合微分进化框架，在内层满足博弈论的要求。并从理论上证明了该算法的低复杂度和收敛性。然后，将所设计的差分博弈算法应用到无线传感器网络中，设计信道分配算法。通过仿真验证了该算法的收敛性和有效性。最后对微分博弈算法进行了讨论，从理论上证明了该算法同样适用于多目标的多目标优化，为低复杂度收敛的多目标优化算法提供了设计思路。

引用次数: 0

Temporal knowledge graph reasoning based on time layering and fuzzy logic rules 基于时间分层和模糊逻辑规则的时态知识图推理

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114598

Luyi Bai , Qianwen Xiao , Tongji Liu , Kejie Guo

Temporal knowledge graph (TKG) reasoning aims to address the incompleteness of temporal knowledge graphs by inferring unknown facts from known ones. However, existing methods either overlook fuzzy semantics or apply rigid temporal constraints—creating a critical gap in handling real-world ambiguity and event causality. This paper proposes a time hierarchy and fuzzy logic rule-based multi-hop path reasoning model TiFuLa to tackle these challenges, a multi-hop path reasoning model that computes fuzzy membership of relations via frequency, duration, recency, and stability, categorizes relations into strong/weak temporal correlation using Relation-Time Correlation (RTC) and learns rule axioms aligned with each category. Finally, by applying these rules, the quadruples are inferred from the selected case rules and ranked by integrating scores from fuzzy membership and temporal differences. Experimental results on four benchmark datasets show this method enhances the logic rule-based multi-hop path reasoning in temporal knowledge graphs through the calculation of fuzzy membership degrees for relational predicates and time hierarchical processing. On the ICEWS18 dataset, TiFuLa’s performance improved by an average of nearly 17 percentage points. On the ICEWS0515 dataset, the improvement of TiFuLa is more significant, with an average increase of 24 percentage points in MRR and Hits @ (1/3/10) indicators.

时间知识图（TKG）推理旨在通过从已知事实推断未知事实来解决时间知识图的不完全性问题。然而，现有的方法要么忽略模糊语义，要么应用严格的时间约束，这在处理现实世界的模糊性和事件因果关系方面造成了严重的差距。为了解决这些问题，本文提出了一种基于时间层次和模糊逻辑规则的多跳路径推理模型TiFuLa，该多跳路径推理模型通过频率、持续时间、近时性和稳定性计算关系的模糊隶属度，使用关系-时间相关性（RTC）将关系分类为强/弱时间相关性，并学习与每个类别一致的规则公理。最后，通过应用这些规则，从所选的案例规则中推断出四组，并通过模糊隶属度和时间差异的积分进行排序。在4个基准数据集上的实验结果表明，该方法通过计算关系谓词的模糊隶属度和时间层次处理，增强了时间知识图中基于逻辑规则的多跳路径推理。在ICEWS18数据集上，TiFuLa的性能平均提高了近17个百分点。在ICEWS0515数据集上，TiFuLa的改进更为显著，MRR和Hits @（1/3/10）指标平均提高了24个百分点。

{"title":"Temporal knowledge graph reasoning based on time layering and fuzzy logic rules","authors":"Luyi Bai , Qianwen Xiao , Tongji Liu , Kejie Guo","doi":"10.1016/j.asoc.2026.114598","DOIUrl":"10.1016/j.asoc.2026.114598","url":null,"abstract":"<div><div>Temporal knowledge graph (TKG) reasoning aims to address the incompleteness of temporal knowledge graphs by inferring unknown facts from known ones. However, existing methods either overlook fuzzy semantics or apply rigid temporal constraints—creating a critical gap in handling real-world ambiguity and event causality. This paper proposes a time hierarchy and fuzzy logic rule-based multi-hop path reasoning model TiFuLa to tackle these challenges, a multi-hop path reasoning model that computes fuzzy membership of relations via frequency, duration, recency, and stability, categorizes relations into strong/weak temporal correlation using Relation-Time Correlation (RTC) and learns rule axioms aligned with each category. Finally, by applying these rules, the quadruples are inferred from the selected case rules and ranked by integrating scores from fuzzy membership and temporal differences. Experimental results on four benchmark datasets show this method enhances the logic rule-based multi-hop path reasoning in temporal knowledge graphs through the calculation of fuzzy membership degrees for relational predicates and time hierarchical processing. On the ICEWS18 dataset, TiFuLa’s performance improved by an average of nearly 17 percentage points. On the ICEWS0515 dataset, the improvement of TiFuLa is more significant, with an average increase of 24 percentage points in MRR and Hits @ (1/3/10) indicators.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114598"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146024245","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}

引用次数: 0

Using sentiment analysis and CEEMDAN to learn the preferences of consumer groups: A case study of online hotel reviews 利用情感分析和CEEMDAN了解消费者群体的偏好：以在线酒店评论为例

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114625

Dan Wang , Zaiwu Gong , Guo Wei , María Ángeles Martínez , Enrique Herrera-Viedma

With the rapid development of big data technologies, online reviews have emerged as a crucial platform where consumers can express their opinions. However, the unstructured nature, temporal characteristics, and implicit and ambiguous nature of these preferences pose formidable challenges for researchers. This paper introduces a data-driven model designed to learn and interpret consumer preferences. The approach begins with fine-grained sentiment analysis of online reviews via the deep learning model known as Bidirectional Encoder Representations from Transformers (BERT). The Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) method is then applied to extract the residual trend component from the data. The mean value of this component serves as an approximate representation of group-consistent sentiment features, providing quantitative support for the subsequent preference modeling. Following this, the UTilités Additives (UTA) method is used to construct a preference disaggregation model that aids in inferring the classification value system of consumer groups and identifying the core factors prioritized by different consumer segments. To validate the model, experiments were conducted using real-world hotel industry review data. The results suggest that the proposed model not only enhances the interpretability of consumer group preferences but also refines and provides insights into their classification value system. Moreover, it enables hotel managers to accurately identify consumer needs, optimize resource allocation, enhance market competitiveness, and offer a scientific basis for evidence-based management decisions tailored to consumer preferences.

随着大数据技术的快速发展，在线评论已经成为消费者表达意见的重要平台。然而，这些偏好的非结构化、时间特征、隐含性和模糊性给研究人员带来了巨大的挑战。本文介绍了一个旨在学习和解释消费者偏好的数据驱动模型。该方法首先通过称为“变形金刚双向编码器表示”（BERT）的深度学习模型对在线评论进行细粒度的情感分析。采用自适应噪声完全集成经验模态分解（CEEMDAN）方法提取残差趋势分量。该分量的均值作为群体一致情绪特征的近似表示，为后续的偏好建模提供定量支持。在此基础上，利用utilitims Additives （UTA）方法构建偏好分解模型，推断消费群体的分类价值体系，识别不同消费群体优先考虑的核心因素。为了验证该模型，我们使用真实的酒店行业评论数据进行了实验。结果表明，该模型不仅增强了消费者群体偏好的可解释性，而且对消费者群体偏好的分类价值体系进行了细化和洞察。此外，它还能使酒店管理者准确识别消费者需求，优化资源配置，增强市场竞争力，为根据消费者偏好进行循证管理决策提供科学依据。

{"title":"Using sentiment analysis and CEEMDAN to learn the preferences of consumer groups: A case study of online hotel reviews","authors":"Dan Wang , Zaiwu Gong , Guo Wei , María Ángeles Martínez , Enrique Herrera-Viedma","doi":"10.1016/j.asoc.2026.114625","DOIUrl":"10.1016/j.asoc.2026.114625","url":null,"abstract":"<div><div>With the rapid development of big data technologies, online reviews have emerged as a crucial platform where consumers can express their opinions. However, the unstructured nature, temporal characteristics, and implicit and ambiguous nature of these preferences pose formidable challenges for researchers. This paper introduces a data-driven model designed to learn and interpret consumer preferences. The approach begins with fine-grained sentiment analysis of online reviews via the deep learning model known as Bidirectional Encoder Representations from Transformers (BERT). The Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) method is then applied to extract the residual trend component from the data. The mean value of this component serves as an approximate representation of group-consistent sentiment features, providing quantitative support for the subsequent preference modeling. Following this, the UTilités Additives (UTA) method is used to construct a preference disaggregation model that aids in inferring the classification value system of consumer groups and identifying the core factors prioritized by different consumer segments. To validate the model, experiments were conducted using real-world hotel industry review data. The results suggest that the proposed model not only enhances the interpretability of consumer group preferences but also refines and provides insights into their classification value system. Moreover, it enables hotel managers to accurately identify consumer needs, optimize resource allocation, enhance market competitiveness, and offer a scientific basis for evidence-based management decisions tailored to consumer preferences.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114625"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039863","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}

引用次数: 0

Transformer optimization algorithm for selecting tokens based on genetic algorithm 基于遗传算法的变压器令牌选择优化算法

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114632

Tao Zhou , Yuxia Niu , Huiling Lu , Yujie Guo , Long Liu , Huiyu Zhou

In recent years, Transformer are widely recognized in the field of artificial intelligence for its excellent performance. However, the tokens number is increased exponentially in Transformer, "How to select token?" and "How to evaluate the importance of each token ?" are key focus areas in current research. To solve the above problems, this paper proposes a global token selection model GAFormer based on Genetic Algorithm. Firstly, Aiming to "How to select token?", binary encoding strategy, whole-part coupled fitness function, elitism strategy-roulette selection, two-point crossover, and one-point mutation are used to realize genetic evolution; Secondly, Aiming to "How to evaluate the token importance?", a whole-part coupled fitness function is constructed, using whole gene selection to obtain "probability vector", using part gene selection to obtain "selection result vector", and using the scoring strategy of whole-part coupled to obtain the fitness value of each chromosome; Finally, the model is verified on 2 public lung X-ray datasets. On dataset 1, its accuracy, precision, recall, F1, and specificity are reached 96.17 %, 92.33 %, 92.35 %, 92.50 %, and 97.45 % respectively, on dataset 2, the corresponding results are 95.84 %, 93.23 %, 93.41 %, 93.26 %, and 96.30 % respectively. This model provides a new idea for token selection in Transformer, which has positive significance for the further development of the large medical model based on Transformer.

近年来，Transformer因其优异的性能在人工智能领域得到广泛认可。然而，在Transformer中令牌数量呈指数级增长，“如何选择令牌”和“如何评估每个令牌的重要性”是当前研究的重点领域。为了解决上述问题，本文提出了一种基于遗传算法的全局令牌选择模型GAFormer。首先，针对“如何选择token”问题，采用二值编码策略、整体耦合适应度函数、精英策略-轮盘选择、两点交叉、一点突变等方法实现遗传进化；其次，针对“如何评估令牌重要性”问题，构建了整体部分耦合适应度函数，利用整体基因选择获得“概率向量”，利用部分基因选择获得“选择结果向量”，利用整体部分耦合评分策略获得各染色体的适应度值；最后，在2个公开的肺x射线数据集上对模型进行了验证。数据集1,其准确性、精密,记得,F1和特异性 %达到96.17,92.33 % 92.35 %, % 92.50和97.45分别 %,在数据集2,相应的结果是95.84 % 93.23 % 93.41 %, % 93.26和96.30分别 %。该模型为Transformer中的令牌选择提供了一种新的思路，对基于Transformer的大型医疗模型的进一步开发具有积极意义。

{"title":"Transformer optimization algorithm for selecting tokens based on genetic algorithm","authors":"Tao Zhou , Yuxia Niu , Huiling Lu , Yujie Guo , Long Liu , Huiyu Zhou","doi":"10.1016/j.asoc.2026.114632","DOIUrl":"10.1016/j.asoc.2026.114632","url":null,"abstract":"<div><div>In recent years, Transformer are widely recognized in the field of artificial intelligence for its excellent performance. However, the tokens number is increased exponentially in Transformer, \"How to select token?\" and \"How to evaluate the importance of each token ?\" are key focus areas in current research. To solve the above problems, this paper proposes a global token selection model GAFormer based on Genetic Algorithm. Firstly, Aiming to \"How to select token?\", binary encoding strategy, whole-part coupled fitness function, elitism strategy-roulette selection, two-point crossover, and one-point mutation are used to realize genetic evolution; Secondly, Aiming to \"How to evaluate the token importance?\", a whole-part coupled fitness function is constructed, using whole gene selection to obtain \"probability vector\", using part gene selection to obtain \"selection result vector\", and using the scoring strategy of whole-part coupled to obtain the fitness value of each chromosome; Finally, the model is verified on 2 public lung X-ray datasets. On dataset 1, its accuracy, precision, recall, F1, and specificity are reached 96.17 %, 92.33 %, 92.35 %, 92.50 %, and 97.45 % respectively, on dataset 2, the corresponding results are 95.84 %, 93.23 %, 93.41 %, 93.26 %, and 96.30 % respectively. This model provides a new idea for token selection in Transformer, which has positive significance for the further development of the large medical model based on Transformer.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114632"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145979788","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}

引用次数: 0

Wildfire spots analysis and forecasting: Evaluation of univariate and multivariate based on variational mode decomposition models 野火点分析与预测：基于变分模态分解模型的单变量与多变量评价

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114606

Vinicius Lovatel Rocha , Ramon Gomes da Silva , Gilson Adamczuk Oliveira , Matheus Henrique Dal Molin Ribeiro

Every year, forest fires affect large areas of Santa Catarina State in Brazil, causing significant environmental, economic, and property damage. This study aims to predict fire outbreaks in Lages and Xanxerê, municipalities with high incidence rates, using data from the Santa Catarina Military Fire Department (CBMSC). The goal is to support more efficient resource management and personnel allocation during high-risk periods. The time series comprises 70 monthly observations, covering February 2019 to December 2024. The methodology combines univariate and multivariate models with Variational Mode Decomposition (VMD) to generate short-term forecasts for one-, two-, and three-month horizons, applying a rolling-window cross-validation protocol. The data were tested in their original form and with logarithmic and square root transformations to enhance predictive performance. Evaluation measures included RMSE, MAE, sMAPE, and error standard deviation. In Lages, the best result was achieved using square-root transformed data and the VMD–LASSO model (RMSE = 1.2074; sMAPE = 58.90). In Xanxerê, the best performance was achieved by applying the same model to the original data (RMSE = 1.5248; sMAPE = 67.79). These findings provide a decision–support tool for CBMSC, enabling more accurate planning and resource mobilization during critical periods and strengthening wildfire prevention and response strategies.

每年，森林火灾都会影响巴西圣卡塔琳娜州的大片地区，造成重大的环境、经济和财产损失。本研究旨在利用圣卡塔琳娜军事消防局（CBMSC）的数据，预测Lages和Xanxerê这两个高发病率城市的火灾爆发。目标是在高风险时期支持更有效的资源管理和人员分配。该时间序列包括70个月的观测，涵盖2019年2月至2024年12月。该方法将单变量和多变量模型与变分模式分解（VMD）相结合，应用滚动窗口交叉验证协议，生成1个月、2个月和3个月的短期预测。数据以原始形式进行测试，并使用对数和平方根转换来提高预测性能。评价指标包括RMSE、MAE、sMAPE和误差标准差。在Lages中，使用平方根变换数据和VMD-LASSO模型（RMSE = 1.2074; sMAPE = 58.90）获得了最好的结果。在Xanxerê中，将相同的模型应用于原始数据获得了最好的性能（RMSE = 1.5248; sMAPE = 67.79）。这些发现为CBMSC提供了决策支持工具，可以在关键时期实现更准确的规划和资源动员，并加强野火预防和响应策略。

{"title":"Wildfire spots analysis and forecasting: Evaluation of univariate and multivariate based on variational mode decomposition models","authors":"Vinicius Lovatel Rocha , Ramon Gomes da Silva , Gilson Adamczuk Oliveira , Matheus Henrique Dal Molin Ribeiro","doi":"10.1016/j.asoc.2026.114606","DOIUrl":"10.1016/j.asoc.2026.114606","url":null,"abstract":"<div><div>Every year, forest fires affect large areas of Santa Catarina State in Brazil, causing significant environmental, economic, and property damage. This study aims to predict fire outbreaks in Lages and Xanxerê, municipalities with high incidence rates, using data from the Santa Catarina Military Fire Department (CBMSC). The goal is to support more efficient resource management and personnel allocation during high-risk periods. The time series comprises 70 monthly observations, covering February 2019 to December 2024. The methodology combines univariate and multivariate models with Variational Mode Decomposition (VMD) to generate short-term forecasts for one-, two-, and three-month horizons, applying a rolling-window cross-validation protocol. The data were tested in their original form and with logarithmic and square root transformations to enhance predictive performance. Evaluation measures included RMSE, MAE, sMAPE, and error standard deviation. In Lages, the best result was achieved using square-root transformed data and the VMD–LASSO model (RMSE = 1.2074; sMAPE = 58.90). In Xanxerê, the best performance was achieved by applying the same model to the original data (RMSE = 1.5248; sMAPE = 67.79). These findings provide a decision–support tool for CBMSC, enabling more accurate planning and resource mobilization during critical periods and strengthening wildfire prevention and response strategies.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"190 ","pages":"Article 114606"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980367","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}

引用次数: 0

A multimodal small sample pipeline leak detection method based on spatio-temporal fusion 基于时空融合的多模态小样本管道泄漏检测方法

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114628

Jingyi Lu , Yuhang Wang , Jing Chen , Yao Chen , Dongmei Wang , Peng Wang

In the industry, leak detection is a critical task to ensure the safe operation of oil and gas pipelines. Due to the low frequency and short duration of pipeline leakage in actual operation, and the small leakage is easily drowned by noise, these factors affect the accuracy and generalization of the leakage diagnosis model. This study proposes a new multimodal few-shot leak detection method based on spatio-temporal fusion. First, we combine VQ-VAE and short-time Fourier transform (STFT) to enhance one-dimensional leak signals. We then build a spatio-temporal bidirectional (STB) dataset by combining the enhanced signals with images converted from time-series data, improving data reliability. Next, we design a spatio-temporal fusion (STF) network, which uses the PKO algorithm to optimize key model parameters and integrates an intra-inter module (IIM) attention mechanism to fuse multi-source feature information, thereby boosting model performance. Finally, the proposed method is validated using pipeline data under different working conditions. The results show that this method achieves an accuracy of 98.67 % in identifying various types of leak signals, comparative experiments demonstrate that this method outperforms other models, demonstrating strong robustness.

在工业中，泄漏检测是确保油气管道安全运行的一项关键任务。由于实际运行中管道泄漏频率低、持续时间短，小泄漏容易被噪声淹没，这些因素影响了泄漏诊断模型的准确性和泛化性。提出了一种基于时空融合的多模态少弹泄漏检测方法。首先，结合VQ-VAE和短时傅里叶变换（STFT）增强一维泄漏信号；然后，我们将增强信号与从时间序列数据转换的图像相结合，构建时空双向（STB）数据集，提高数据可靠性。接下来，我们设计了一个时空融合（STF）网络，该网络使用PKO算法来优化模型关键参数，并集成了一个模块间（IIM）关注机制来融合多源特征信息，从而提高模型的性能。最后，利用不同工况下的管道数据对所提方法进行了验证。结果表明，该方法对各类泄漏信号的识别准确率达到98.67 %，对比实验表明，该方法优于其他模型，具有较强的鲁棒性。

{"title":"A multimodal small sample pipeline leak detection method based on spatio-temporal fusion","authors":"Jingyi Lu , Yuhang Wang , Jing Chen , Yao Chen , Dongmei Wang , Peng Wang","doi":"10.1016/j.asoc.2026.114628","DOIUrl":"10.1016/j.asoc.2026.114628","url":null,"abstract":"<div><div>In the industry, leak detection is a critical task to ensure the safe operation of oil and gas pipelines. Due to the low frequency and short duration of pipeline leakage in actual operation, and the small leakage is easily drowned by noise, these factors affect the accuracy and generalization of the leakage diagnosis model. This study proposes a new multimodal few-shot leak detection method based on spatio-temporal fusion. First, we combine VQ-VAE and short-time Fourier transform (STFT) to enhance one-dimensional leak signals. We then build a spatio-temporal bidirectional (STB) dataset by combining the enhanced signals with images converted from time-series data, improving data reliability. Next, we design a spatio-temporal fusion (STF) network, which uses the PKO algorithm to optimize key model parameters and integrates an intra-inter module (IIM) attention mechanism to fuse multi-source feature information, thereby boosting model performance. Finally, the proposed method is validated using pipeline data under different working conditions. The results show that this method achieves an accuracy of 98.67 % in identifying various types of leak signals, comparative experiments demonstrate that this method outperforms other models, demonstrating strong robustness.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114628"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039855","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}

引用次数: 0

Multimodal transformers for image and audio polyphonic music transcription 用于图像和音频复调音乐转录的多模态变压器

IF 6.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Applied Soft Computing

Pub Date : 2026-01-12 DOI: 10.1016/j.asoc.2026.114643

María Alfaro-Contreras, Noelia Luna-Barahona, Carlos Pérez-Sancho, Jorge Calvo-Zaragoza, Jose J. Valero-Mas

The attainment of structured representations of music sources—commonly referred to as music transcription—has long been a central challenge in the field of Music Information Retrieval. Traditionally, research has addressed this task through Optical Music Recognition for sheet music and through Automatic Music Transcription for audio recordings. More recently, multimodal approaches that integrate both image and audio modalities have emerged, aiming to exploit their complementary strengths. However, these existing methods have so far been evaluated only in controlled settings with monophonic music. In this work, we present the first multimodal image-and-audio framework for polyphonic music transcription, built upon the Transformer architecture. Specifically, we design and evaluate six distinct modality-fusion strategies, differing in the stage at which the modalities are integrated (early, intermediate, or late fusion). Our results demonstrate that multimodality can be beneficial for polyphonic transcription—producing comparable or superior performance across all datasets tested and, in the best-performing experiments, improving the performance of the best unimodal transcription scenario by 9% on average—but its impact depends on the strategy: some fusion schemes yield consistent gains, whereas others fail to improve upon the unimodal baseline.

音乐来源的结构化表示的实现-通常被称为音乐转录-长期以来一直是音乐信息检索领域的核心挑战。传统上，研究通过对乐谱的光学音乐识别和对录音的自动音乐转录来解决这一任务。最近，整合图像和音频模式的多模式方法已经出现，旨在利用它们的互补优势。然而，到目前为止，这些现有的方法仅在单音音乐的受控环境中进行了评估。在这项工作中，我们提出了第一个多模态图像和音频框架，用于复调音乐转录，建立在Transformer架构之上。具体来说，我们设计并评估了六种不同的模式融合策略，这些策略在模式整合的阶段不同（早期、中期或晚期融合）。我们的研究结果表明，多模态对复调转录是有益的——在所有测试的数据集中产生相当或更好的性能，并且在表现最好的实验中，将最佳单模态转录场景的性能平均提高9%——但其影响取决于策略：一些融合方案产生一致的收益，而另一些融合方案在单模态基线上没有改善。

{"title":"Multimodal transformers for image and audio polyphonic music transcription","authors":"María Alfaro-Contreras, Noelia Luna-Barahona, Carlos Pérez-Sancho, Jorge Calvo-Zaragoza, Jose J. Valero-Mas","doi":"10.1016/j.asoc.2026.114643","DOIUrl":"10.1016/j.asoc.2026.114643","url":null,"abstract":"<div><div>The attainment of structured representations of music sources—commonly referred to as music transcription—has long been a central challenge in the field of Music Information Retrieval. Traditionally, research has addressed this task through Optical Music Recognition for sheet music and through Automatic Music Transcription for audio recordings. More recently, multimodal approaches that integrate both image and audio modalities have emerged, aiming to exploit their complementary strengths. However, these existing methods have so far been evaluated only in controlled settings with monophonic music. In this work, we present the first multimodal image-and-audio framework for polyphonic music transcription, built upon the Transformer architecture. Specifically, we design and evaluate six distinct modality-fusion strategies, differing in the stage at which the modalities are integrated (early, intermediate, or late fusion). Our results demonstrate that multimodality can be beneficial for polyphonic transcription—producing comparable or superior performance across all datasets tested and, in the best-performing experiments, improving the performance of the best unimodal transcription scenario by 9% on average—but its impact depends on the strategy: some fusion schemes yield consistent gains, whereas others fail to improve upon the unimodal baseline.</div></div>","PeriodicalId":50737,"journal":{"name":"Applied Soft Computing","volume":"191 ","pages":"Article 114643"},"PeriodicalIF":6.6,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146039722","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}

引用次数: 0