The rapid development of information technology and accelerated digitalization have led to an explosive growth of data across various fields.As a key technology for knowledge representation and sharing,knowledge graph...The rapid development of information technology and accelerated digitalization have led to an explosive growth of data across various fields.As a key technology for knowledge representation and sharing,knowledge graphs play a crucial role by constructing structured networks of relationships among entities.However,data sparsity and numerous unexplored implicit relations result in the widespread incompleteness of knowledge graphs.In static knowledge graph completion,most existing methods rely on linear operations or simple interaction mechanisms for triple encoding,making it difficult to fully capture the deep semantic associations between entities and relations.Moreover,many methods focus only on the local information of individual triples,ignoring the rich semantic dependencies embedded in the neighboring nodes of entities within the graph structure,which leads to incomplete embedding representations.To address these challenges,we propose Two-Stage Mixer Embedding(TSMixerE),a static knowledge graph completion method based on entity context.In the unit semantic extraction stage,TSMixerE leveragesmulti-scale circular convolution to capture local features atmultiple granularities,enhancing the flexibility and robustness of feature interactions.A channel attention mechanism amplifies key channel responses to suppress noise and irrelevant information,thereby improving the discriminative power and semantic depth of feature representations.For contextual information fusion,a multi-layer self-attentionmechanism enables deep interactions among contextual cues,effectively integrating local details with global context.Simultaneously,type embeddings clarify the semantic identities and roles of each component,enhancing the model’s sensitivity and fusion capabilities for diverse information sources.Furthermore,TSMixerE constructs contextual unit sequences for entities,fully exploring neighborhood information within the graph structure to model complex semantic dependencies,thus improving the completeness and generalization of embedding representations.展开更多
Interactivity in online newspapers is the focus of this chapter in eliciting readers’evaluation of Zambian online newspapers.This aspect of the study investigates and characterises the motivations(gratification sough...Interactivity in online newspapers is the focus of this chapter in eliciting readers’evaluation of Zambian online newspapers.This aspect of the study investigates and characterises the motivations(gratification sought)for use of interactivity features(“process motivation”)and how widely they are used.It also attempts to ascertain the gratification obtained from their use among readers.The probable relationships between use of the interactivity features(“audience interactivity”)and gratification obtained from them(“process gratification”)and the impact of the perceived credibility of the online newspapers on gratification are also examined.Past studies present mixed results on use of interactivity and gratification obtained from it.This study finds that use of interactivity in Zambian online newspapers is at a low level,although among the three broad categorisations of features of online newspapers,interactivity attracts greater use than hyper-textuality and multi-mediality.Human interactivity features-“knowing what others think about an issue”,“chat on the Facebook page of the newspaper”,“ability to navigate on the Facebook page of the newspaper”,and“posting own comments on stories”-are the main motivations for use of online newspapers,the most frequently used,and the most gratifying to the readers.While readers express an interest in interacting with other readers via online newspapers,they seem less interested in posting their own stories as“citizen journalists”and linking up with the publishers and editors.This finding challenges the notion that all new media are catalysts of participatory and cyclic communication.展开更多
Heart disease includes a multiplicity of medical conditions that affect the structure,blood vessels,and general operation of the heart.Numerous researchers have made progress in correcting and predicting early heart d...Heart disease includes a multiplicity of medical conditions that affect the structure,blood vessels,and general operation of the heart.Numerous researchers have made progress in correcting and predicting early heart disease,but more remains to be accomplished.The diagnostic accuracy of many current studies is inadequate due to the attempt to predict patients with heart disease using traditional approaches.By using data fusion from several regions of the country,we intend to increase the accuracy of heart disease prediction.A statistical approach that promotes insights triggered by feature interactions to reveal the intricate pattern in the data,which cannot be adequately captured by a single feature.We processed the data using techniques including feature scaling,outlier detection and replacement,null and missing value imputation,and more to improve the data quality.Furthermore,the proposed feature engineering method uses the correlation test for numerical features and the chi-square test for categorical features to interact with the feature.To reduce the dimensionality,we subsequently used PCA with 95%variation.To identify patients with heart disease,hyperparameter-based machine learning algorithms like RF,XGBoost,Gradient Boosting,LightGBM,CatBoost,SVM,and MLP are utilized,along with ensemble models.The model’s overall prediction performance ranges from 88%to 92%.In order to attain cutting-edge results,we then used a 1D CNN model,which significantly enhanced the prediction with an accuracy score of 96.36%,precision of 96.45%,recall of 96.36%,specificity score of 99.51%and F1 score of 96.34%.The RF model produces the best results among all the classifiers in the evaluation matrix without feature interaction,with accuracy of 90.21%,precision of 90.40%,recall of 90.86%,specificity of 90.91%,and F1 score of 90.63%.Our proposed 1D CNN model is 7%superior to the one without feature engineering when compared to the suggested approach.This illustrates how interaction-focused feature analysis can produce precise and useful insights for heart disease diagnosis.展开更多
Semantic segmentation of remote sensing images is a critical research area in the field of remote sensing.Despite the success of Convolutional Neural Networks(CNNs),they often fail to capture inter-layer feature relat...Semantic segmentation of remote sensing images is a critical research area in the field of remote sensing.Despite the success of Convolutional Neural Networks(CNNs),they often fail to capture inter-layer feature relationships and fully leverage contextual information,leading to the loss of important details.Additionally,due to significant intraclass variation and small inter-class differences in remote sensing images,CNNs may experience class confusion.To address these issues,we propose a novel Category-Guided Feature Collaborative Learning Network(CG-FCLNet),which enables fine-grained feature extraction and adaptive fusion.Specifically,we design a Feature Collaborative Learning Module(FCLM)to facilitate the tight interaction of multi-scale features.We also introduce a Scale-Aware Fusion Module(SAFM),which iteratively fuses features from different layers using a spatial attention mechanism,enabling deeper feature fusion.Furthermore,we design a Category-Guided Module(CGM)to extract category-aware information that guides feature fusion,ensuring that the fused featuresmore accurately reflect the semantic information of each category,thereby improving detailed segmentation.The experimental results show that CG-FCLNet achieves a Mean Intersection over Union(mIoU)of 83.46%,an mF1 of 90.87%,and an Overall Accuracy(OA)of 91.34% on the Vaihingen dataset.On the Potsdam dataset,it achieves a mIoU of 86.54%,an mF1 of 92.65%,and an OA of 91.29%.These results highlight the superior performance of CG-FCLNet compared to existing state-of-the-art methods.展开更多
To address the challenges of multi-scale differences,complex background interference,and unstable small target positioning in visual inspection of power towers,the existing methods still face issues such as insufficie...To address the challenges of multi-scale differences,complex background interference,and unstable small target positioning in visual inspection of power towers,the existing methods still face issues such as insufficient feature interaction and unstable confidence estimation,which lead to performance degradation in complex backgrounds and occlusion conditions.This paper proposes a precise inspection method for key power tower components using autonomous drone positioning.To this end,this paper makes three key improvements to the you only look once version 11(YOLOv11)framework.First,it constructs C3k2-adaptive multi-receptive field block(C3k2-AMRB),combining multiple dilated convolutions with a reparameterization mechanism to significantly expand the receptive field and enhance multi-scale feature extraction.Second,it designs a hierarchical wavelet interaction unit(HWIU),which leverages high-and low-frequency decomposition and reconstruction of wavelet transform(WT)to achieve cross-scale semantic alignment,enhancing feature discriminability in complex backgrounds.Third,it proposes a distribution-aware confidence refinement head(DACR-Head),which adaptively calibrates classification confidence based on the statistical characteristics of the predicted bounding-box corner distribution,improving the localization stability and accuracy of small targets.Experiments on the inspection of power line assets dataset(InsPLAD)dataset show that the integrated approach achieves a component detection accuracy at intersection over union(IoU)=0.5(CDA_(50))of 88.3%and a component detection robustness(CDR_(50:95))of 69.8%,respectively,improvements of 4.4%and 7.0%over the baseline.展开更多
This paper proposed a novel multi-view interactive behavior recognition method based on local self-similarity descriptors and graph shared multi-task learning. First, we proposed the composite interactive feature repr...This paper proposed a novel multi-view interactive behavior recognition method based on local self-similarity descriptors and graph shared multi-task learning. First, we proposed the composite interactive feature representation which encodes both the spatial distribution of local motion of interest points and their contexts. Furthermore, local self-similarity descriptor represented by temporal-pyramid bag of words(BOW) was applied to decreasing the influence of observation angle change on recognition and retaining the temporal information. For the purpose of exploring latent correlation between different interactive behaviors from different views and retaining specific information of each behaviors, graph shared multi-task learning was used to learn the corresponding interactive behavior recognition model. Experiment results showed the effectiveness of the proposed method in comparison with other state-of-the-art methods on the public databases CASIA, i3Dpose dataset and self-built database for interactive behavior recognition.展开更多
Acoustic quality detection is vital in the manufactured products quality control field since it represents the conditions of machines or products.Recent work employed machine learning models in manufactured audio data...Acoustic quality detection is vital in the manufactured products quality control field since it represents the conditions of machines or products.Recent work employed machine learning models in manufactured audio data to detect anomalous patterns.A major challenge is how to select applicable audio features to meliorate model’s accuracy and precision.To relax this challenge,we extract and analyze three audio feature types including Time Domain Feature,Frequency Domain Feature,and Cepstrum Feature to help identify the potential linear and non-linear relationships.In addition,we design a visual analysis system,namely AFExplorer,to assist data scientists in extracting audio features and selecting potential feature combinations.AFExplorer integrates four main views to present detailed distribution and relevance of the audio features,which helps users observe the impact of features visually in the feature selection.We perform the case study with AFExplore according to the ToyADMOS and MIMII Dataset to demonstrate the usability and effectiveness of the proposed system.展开更多
In recent years,deep learning has been widely applied in the fields of recommendation systems and click-through rate(CTR)prediction,and thus recommendation models incorporating deep learning have emerged.In addition,t...In recent years,deep learning has been widely applied in the fields of recommendation systems and click-through rate(CTR)prediction,and thus recommendation models incorporating deep learning have emerged.In addition,the design and implementation of recommendation models using information related to user behavior sequences is an important direction of current research in recommendation systems,and models calculate the likelihood of users clicking on target items based on their behavior sequence information.In order to explore the relationship between features,this paper improves and optimizes on the basis of deep interest network(DIN)proposed by Ali’s team.Based on the user behavioral sequences information,the attentional factorization machine(AFM)is integrated to obtain richer and more accurate behavioral sequence information.In addition,this paper designs a new way of calculating attention weights,which uses the relationship between the cosine similarity of any two vectors and the absolute value of their modal length difference to measure their relevance degree.Thus,a novel deep learning CTR prediction mode is proposed,that is,the CTR prediction network based on user behavior sequence and feature interactions deep interest and machines network(DIMN).We conduct extensive comparison experiments on three public datasets and one private music dataset,which are more recognized in the industry,and the results show that the DIMN obtains a better performance compared with the classical CTR prediction model.展开更多
This paper presents a feature-based method for machining process planning in integrated product designing and manufacturing system for CE(Concurrent Engineering) application. The feature setup generation and machining...This paper presents a feature-based method for machining process planning in integrated product designing and manufacturing system for CE(Concurrent Engineering) application. The feature setup generation and machining sequence can be determined automatically in this system. The set of knowledge-based rules for process planning and manufacturability evaluation is provided and can be shared by all stages of full product life-cycle. An approach for MTAD (Multiple Tool Axis Direction) feature setup generation is presented and the appropriate Tool Axis Direction(TAD) is chosen to minimize the total setup numbers of a part. The classification and process planning of interacting feature are discussed and the knowledge-based rules are used to solve the feature interaction problem.展开更多
This paper presents methodologies and technologies of feature_based integrated product designing and manufacturing system for CE(Concurrent Engineering) application. The product information is represented on the basis...This paper presents methodologies and technologies of feature_based integrated product designing and manufacturing system for CE(Concurrent Engineering) application. The product information is represented on the basis of hierarchical and dynamic structure of feature representation. The Object_Oriented feature modeling method is adopted to represent the feature classification, feature relationship and feature interaction. The set of knowledge_based rule for process planing and manufacturiability evaluation is provided and can be shared by all stages of full product life_cycle. The feature_based machining operation and machining sequence can be determined automatically. The machining process of the machining feature can be determined according to the set of knowledge_based rule.展开更多
Online advertising click-through rate(CTR) prediction is aimed at predicting the probability of a user clicking an ad,and it has undergone considerable development in recent years.One of the hot topics in this area is...Online advertising click-through rate(CTR) prediction is aimed at predicting the probability of a user clicking an ad,and it has undergone considerable development in recent years.One of the hot topics in this area is the construction of feature interactions to facilitate accurate prediction.Factorization machine provides second-order feature interactions by linearly multiplying hidden feature factors.However,real-world data present a complex and nonlinear structure.Hence,second-order feature interactions are unable to represent cross information adequately.This drawback has been addressed using deep neural networks(DNNs),which enable high-order nonlinear feature interactions.However,DNN-based feature interactions cannot easily optimize deep structures because of the absence of cross information in the original features.In this study,we propose an effective CTR prediction algorithm called CAN,which explicitly exploits the benefits of attention mechanisms and DNN models.The attention mechanism is used to provide rich and expressive low-order feature interactions and facilitate the optimization of DNN-based predictors that implicitly incorporate high-order nonlinear feature interactions.The experiments using two real datasets demonstrate that our proposed CAN model performs better than other cross feature-and DNN-based predictors.展开更多
Gaze following aims to interpret human-scene interactions by predicting the person’s focal point of gaze.Prevailing approaches often adopt a two-stage framework,whereby multi-modality information is extracted in the ...Gaze following aims to interpret human-scene interactions by predicting the person’s focal point of gaze.Prevailing approaches often adopt a two-stage framework,whereby multi-modality information is extracted in the initial stage for gaze target prediction.Consequently,the efficacy of these methods highly depends on the precision of the previous modality extraction.Others use a single-modality approach with complex decoders,increasing network computational load.Inspired by the remarkable success of pre-trained plain vision transformers(ViTs),we introduce a novel single-modality gaze following framework called ViTGaze.In contrast to previous methods,it creates a novel gaze following framework based mainly on powerful encoders(relative decoder parameters less than 1%).Our principal insight is that the inter-token interactions within self-attention can be transferred to interactions between humans and scenes.Leveraging this presumption,we formulate a framework consisting of a 4D interaction encoder and a 2D spatial guidance module to extract human-scene interaction information from self-attention maps.Furthermore,our investigation reveals that ViT with self-supervised pre-training has an enhanced ability to extract correlation information.Many experiments have been conducted to demonstrate the performance of the proposed method.Our method achieves state-of-the-art performance among all single-modality methods(3.4%improvement in the area under curve score,5.1% improvement in the average precision)and very comparable performance against multi-modality methods with 59% fewer parameters.展开更多
Multimodal Sentiment analysis refers to analyzing emotions in infor-mation carriers containing multiple modalities.To better analyze the features within and between modalities and solve the problem of incomplete multi...Multimodal Sentiment analysis refers to analyzing emotions in infor-mation carriers containing multiple modalities.To better analyze the features within and between modalities and solve the problem of incomplete multimodal feature fusion,this paper proposes a multimodal sentiment analysis model MIF(Modal Interactive Feature Encoder For Multimodal Sentiment Analysis).First,the global features of three modalities are obtained through unimodal feature extraction networks.Second,the inter-modal interactive feature encoder and the intra-modal interactive feature encoder extract similarity features between modal-ities and intra-modal special features separately.Finally,unimodal special features and the interaction information between modalities are decoded to get the fusion features and predict sentimental polarity results.We conduct extensive experi-ments on three public multimodal datasets,including one in Chinese and two in English.The results show that the performance of our approach is significantly improved compared with benchmark models.展开更多
Correction to:GraphFM:Graph Factorization Machines for Feature Interaction Modelling DOI:10.1007/s11633-024-1505-5 Authors:Shu Wu,Zekun Li,Yunyue Su,Zeyu Cui,Xiaoyu Zhang,Liang Wang The article GraphFM:Graph Factoriza...Correction to:GraphFM:Graph Factorization Machines for Feature Interaction Modelling DOI:10.1007/s11633-024-1505-5 Authors:Shu Wu,Zekun Li,Yunyue Su,Zeyu Cui,Xiaoyu Zhang,Liang Wang The article GraphFM:Graph Factorization Machines for Feature Interaction Modelling,written by Shu Wu,Zekun Li,Yunyue Su,Zeyu Cui,Xiaoyu Zhang,Liang Wang,was originally published without Open Access.After publication,the authors decided to opt for Open Choice and to make the article an Open Access publication.展开更多
Factorization machine(FM)is a prevalent approach to modelling pairwise(second-order)feature interactions when dealing with high-dimensional sparse data.However,on the one hand,FMs fail to capture higher-order feature ...Factorization machine(FM)is a prevalent approach to modelling pairwise(second-order)feature interactions when dealing with high-dimensional sparse data.However,on the one hand,FMs fail to capture higher-order feature interactions suffering from combinatorial expansion.On the other hand,taking into account interactions between every pair of features may introduce noise and degrade the prediction accuracy.To solve these problems,we propose a novel approach,the graph factorization machine(GraphFM),which naturally represents features in the graph structure.In particular,we design a mechanism to select beneficial feature interactions and formulate them as edges between features.Then the proposed model,which integrates the interaction function of the FM into the feature aggregation strategy of the graph neural network(GNN),can model arbitrary-order feature interactions on graph-structured features by stacking layers.Experimental results on several real-world datasets demonstrate the rationality and effectiveness of our proposed approach.The code and data are available at https://github.com/CRIPAC-DIG/GraphCTR.展开更多
Object detection serves as a challenging yet crucial task in computer vision.Despite significant advancements,modern detectors remain struggling with task alignment between localization and classification.In this pape...Object detection serves as a challenging yet crucial task in computer vision.Despite significant advancements,modern detectors remain struggling with task alignment between localization and classification.In this paper,Global Collaborative Learning(GCL)is introduced to address these challenges from often-overlooked perspectives.First,the essence of GCL is reflected in the label assignment of the detector.Adjusting the loss function to transform samples with strong localization yet weak classification into high-quality samples in both tasks,provides more effective training signals,enabling the model to capture key consistent features.Second,the spirit of GCL is embodied in the head design.By enabling global feature interaction within the decoupled head,the approach ensures that final predictions are made more comprehensively and robustly,thereby preventing the two independent branches from converging into suboptimal solutions for their respective tasks.Extensive experiments on the challenging MS COCO and CrowdHuman datasets demonstrate that the proposed GCL method substantially enhances performance and generalization capabilities.展开更多
With the rapid advancement of mechanical automation and intelligent processing technology,ac-curately measuring the surfaces of complex parts has emerged as a significant research challenge.Robotic measurement technol...With the rapid advancement of mechanical automation and intelligent processing technology,ac-curately measuring the surfaces of complex parts has emerged as a significant research challenge.Robotic measurement technology plays a crucial role in facilitating rapid quality inspections during the manufacturing process due to its inherent flexibility.However,the irregular shapes and viewpoint occlusions of complex parts complicate precise measurement.To address these challenges,this work proposes a point cloud registration network for robotic scanning systems and introduces a DBR-Net(Dual-line Registration Network)to overcome the issues of low overlap rates and perspective occlusion that currently impede the registration of certain workpieces.First,feature extraction is performed using a bilinear encoder and multi-level feature interactions of both point-wise and global features.Subsequently,the features are sampled through unanimous voting and fed into the RANSAC(Random Sample Consensus)algorithm for pose estimation,enabling multi-view point cloud registration.Experimental results demonstrate that this method significantly outperforms many existing techniques in terms of feature extraction and registration accuracy,thereby enhancing the overall performance of point cloud registration.展开更多
CO_(2) injection not only effectively enhances oil recovery(EOR)but also facilitates CO_(2) utilization and storage.Rapid screening and optimization of CO_(2)-EOR operations is urgently needed for unconventional reser...CO_(2) injection not only effectively enhances oil recovery(EOR)but also facilitates CO_(2) utilization and storage.Rapid screening and optimization of CO_(2)-EOR operations is urgently needed for unconventional reservoirs.However,it remains challenging due to a limited understanding of fluid flow in multiscale porous media and the problem complexity invoked by numerous factors.This work developed a new interpretable machine learning(ML)framework to specifically address this issue.Three different methods,namely random forest(RF),support vector regression(SVR),and artificial neural network(ANN),were used to establish proxy models using the data from a specific unconventional reservoir,and the RF model demonstrated a preferable performance.To enhance the interpretability of the established models,the multiway feature importance analysis and Shapley Additive Explanations(SHAP)were proposed to quantify the contribution of individual features to the model output.Based on the results of model interpretability,the genetic algorithm(GA)was coupled with RF(RF-GA model)to optimize the CO_(2)-EOR process.The proposed framework was validated by comparing the GA-RF predictions with simulation results under different reservoir conditions,which yielded a minimum relative error of 0.34%and an average relative error of 5.3%.The developed interpretable ML method was capable of rapidly screening suitable CO_(2)-EOR strategies based on reservoir conditions and provided a practical example for field applications.展开更多
Object detection in unmanned aerial vehicle(UAV)images plays an important role in environmental monitoring,post-disaster rescue and other fields.However,the complex context and small size of typical objects in UAV ima...Object detection in unmanned aerial vehicle(UAV)images plays an important role in environmental monitoring,post-disaster rescue and other fields.However,the complex context and small size of typical objects in UAV images pose great challenges to existing object detection algorithms.To address these issues,a new object detection method based on improved YOLOv7 with higher-order feature interaction(HOFI-YOLOv7)is proposed for UAV aerial images.Firstly,a new higher-order feature interaction(HOFI)module was constructed to promote the interaction of feature information.And then,a feature fusion network with effective channel attention(ECA)module was designed to improve the focus of the method on small objects.Finally,a focal extended intersection over union(Focal-EIOU)loss function is utilized to improve the detection accuracy of small objects.The experimental results on the VisDrone2019 dataset show that compared with the original model,the mAP50 and mAP50∶5∶95 of the improved model are increased by 3.37%and 1.77%,respectively.Furthermore,the proposed method exhibits superior generalization performance on the self-picked dataset.展开更多
Real-time conflict prediction at signalized intersections is crucial for urban road safety management.This study developed a real-time conflict prediction framework for signal-ized intersections using real-time video ...Real-time conflict prediction at signalized intersections is crucial for urban road safety management.This study developed a real-time conflict prediction framework for signal-ized intersections using real-time video data recognition technology and deep learning techniques,incorporating lane-level information and feature interactions.The modeling framework consists of three stages:real-time video data extraction and processing,the development of a deep and cross network(DCN)-based real-time traffic conflict prediction model,and conflict-driven factor interpretability analysis through Shapley additive expla-nations(SHAP).In the first stage,an efficient automated trajectory extraction system is designed to obtain vehicle trajectories in real time for dynamic traffic parameters and con-flict frequency identification.In the second stage,a DCN model is developed to construct the relationships between dynamic traffic parameters,including their interactions,and traffic conflicts.In the third stage,SHAP explores the impact mechanisms of different dynamic traffic parameters on traffic conflicts.The model’s predictive performance and interpretability are evaluated using intersection video data from Changsha City,China.The results are as follows.(1)In real-time traffic conflict prediction at signalized intersec-tions across different modified time-to-conflict thresholds(1.5 s and 3.0 s),the DCN model consistently outperformed statistical and machine learning models.(2)High traffic flows on main and secondary roads at signalized intersections significantly increase the com-plexity and frequency of conflicts,with varying sensitivity depending on the interaction of traffic flow,speed,and platoon length.(3)The proposed framework provides a safety measurement standard for data-driven road safety management methods.展开更多
基金supported by the National Natural Science Foundation of China(No.62267005)the Chinese Guangxi Natural Science Foundation(No.2023GXNSFAA026493)+1 种基金Guangxi Collaborative Innovation Center ofMulti-Source Information Integration and Intelligent ProcessingGuangxi Academy of Artificial Intelligence.
文摘The rapid development of information technology and accelerated digitalization have led to an explosive growth of data across various fields.As a key technology for knowledge representation and sharing,knowledge graphs play a crucial role by constructing structured networks of relationships among entities.However,data sparsity and numerous unexplored implicit relations result in the widespread incompleteness of knowledge graphs.In static knowledge graph completion,most existing methods rely on linear operations or simple interaction mechanisms for triple encoding,making it difficult to fully capture the deep semantic associations between entities and relations.Moreover,many methods focus only on the local information of individual triples,ignoring the rich semantic dependencies embedded in the neighboring nodes of entities within the graph structure,which leads to incomplete embedding representations.To address these challenges,we propose Two-Stage Mixer Embedding(TSMixerE),a static knowledge graph completion method based on entity context.In the unit semantic extraction stage,TSMixerE leveragesmulti-scale circular convolution to capture local features atmultiple granularities,enhancing the flexibility and robustness of feature interactions.A channel attention mechanism amplifies key channel responses to suppress noise and irrelevant information,thereby improving the discriminative power and semantic depth of feature representations.For contextual information fusion,a multi-layer self-attentionmechanism enables deep interactions among contextual cues,effectively integrating local details with global context.Simultaneously,type embeddings clarify the semantic identities and roles of each component,enhancing the model’s sensitivity and fusion capabilities for diverse information sources.Furthermore,TSMixerE constructs contextual unit sequences for entities,fully exploring neighborhood information within the graph structure to model complex semantic dependencies,thus improving the completeness and generalization of embedding representations.
文摘Interactivity in online newspapers is the focus of this chapter in eliciting readers’evaluation of Zambian online newspapers.This aspect of the study investigates and characterises the motivations(gratification sought)for use of interactivity features(“process motivation”)and how widely they are used.It also attempts to ascertain the gratification obtained from their use among readers.The probable relationships between use of the interactivity features(“audience interactivity”)and gratification obtained from them(“process gratification”)and the impact of the perceived credibility of the online newspapers on gratification are also examined.Past studies present mixed results on use of interactivity and gratification obtained from it.This study finds that use of interactivity in Zambian online newspapers is at a low level,although among the three broad categorisations of features of online newspapers,interactivity attracts greater use than hyper-textuality and multi-mediality.Human interactivity features-“knowing what others think about an issue”,“chat on the Facebook page of the newspaper”,“ability to navigate on the Facebook page of the newspaper”,and“posting own comments on stories”-are the main motivations for use of online newspapers,the most frequently used,and the most gratifying to the readers.While readers express an interest in interacting with other readers via online newspapers,they seem less interested in posting their own stories as“citizen journalists”and linking up with the publishers and editors.This finding challenges the notion that all new media are catalysts of participatory and cyclic communication.
基金supported by the Competitive Research Fund of the University of Aizu,Japan(Grant No.P-13).
文摘Heart disease includes a multiplicity of medical conditions that affect the structure,blood vessels,and general operation of the heart.Numerous researchers have made progress in correcting and predicting early heart disease,but more remains to be accomplished.The diagnostic accuracy of many current studies is inadequate due to the attempt to predict patients with heart disease using traditional approaches.By using data fusion from several regions of the country,we intend to increase the accuracy of heart disease prediction.A statistical approach that promotes insights triggered by feature interactions to reveal the intricate pattern in the data,which cannot be adequately captured by a single feature.We processed the data using techniques including feature scaling,outlier detection and replacement,null and missing value imputation,and more to improve the data quality.Furthermore,the proposed feature engineering method uses the correlation test for numerical features and the chi-square test for categorical features to interact with the feature.To reduce the dimensionality,we subsequently used PCA with 95%variation.To identify patients with heart disease,hyperparameter-based machine learning algorithms like RF,XGBoost,Gradient Boosting,LightGBM,CatBoost,SVM,and MLP are utilized,along with ensemble models.The model’s overall prediction performance ranges from 88%to 92%.In order to attain cutting-edge results,we then used a 1D CNN model,which significantly enhanced the prediction with an accuracy score of 96.36%,precision of 96.45%,recall of 96.36%,specificity score of 99.51%and F1 score of 96.34%.The RF model produces the best results among all the classifiers in the evaluation matrix without feature interaction,with accuracy of 90.21%,precision of 90.40%,recall of 90.86%,specificity of 90.91%,and F1 score of 90.63%.Our proposed 1D CNN model is 7%superior to the one without feature engineering when compared to the suggested approach.This illustrates how interaction-focused feature analysis can produce precise and useful insights for heart disease diagnosis.
基金funded by National Natural Science Foundation of China(61603245).
文摘Semantic segmentation of remote sensing images is a critical research area in the field of remote sensing.Despite the success of Convolutional Neural Networks(CNNs),they often fail to capture inter-layer feature relationships and fully leverage contextual information,leading to the loss of important details.Additionally,due to significant intraclass variation and small inter-class differences in remote sensing images,CNNs may experience class confusion.To address these issues,we propose a novel Category-Guided Feature Collaborative Learning Network(CG-FCLNet),which enables fine-grained feature extraction and adaptive fusion.Specifically,we design a Feature Collaborative Learning Module(FCLM)to facilitate the tight interaction of multi-scale features.We also introduce a Scale-Aware Fusion Module(SAFM),which iteratively fuses features from different layers using a spatial attention mechanism,enabling deeper feature fusion.Furthermore,we design a Category-Guided Module(CGM)to extract category-aware information that guides feature fusion,ensuring that the fused featuresmore accurately reflect the semantic information of each category,thereby improving detailed segmentation.The experimental results show that CG-FCLNet achieves a Mean Intersection over Union(mIoU)of 83.46%,an mF1 of 90.87%,and an Overall Accuracy(OA)of 91.34% on the Vaihingen dataset.On the Potsdam dataset,it achieves a mIoU of 86.54%,an mF1 of 92.65%,and an OA of 91.29%.These results highlight the superior performance of CG-FCLNet compared to existing state-of-the-art methods.
基金supported by the National Natural Science Foundation of China(No.61702347)Hebei Academy of Sciences Basic Research Operating Fund Project(No.2025PF21)。
文摘To address the challenges of multi-scale differences,complex background interference,and unstable small target positioning in visual inspection of power towers,the existing methods still face issues such as insufficient feature interaction and unstable confidence estimation,which lead to performance degradation in complex backgrounds and occlusion conditions.This paper proposes a precise inspection method for key power tower components using autonomous drone positioning.To this end,this paper makes three key improvements to the you only look once version 11(YOLOv11)framework.First,it constructs C3k2-adaptive multi-receptive field block(C3k2-AMRB),combining multiple dilated convolutions with a reparameterization mechanism to significantly expand the receptive field and enhance multi-scale feature extraction.Second,it designs a hierarchical wavelet interaction unit(HWIU),which leverages high-and low-frequency decomposition and reconstruction of wavelet transform(WT)to achieve cross-scale semantic alignment,enhancing feature discriminability in complex backgrounds.Third,it proposes a distribution-aware confidence refinement head(DACR-Head),which adaptively calibrates classification confidence based on the statistical characteristics of the predicted bounding-box corner distribution,improving the localization stability and accuracy of small targets.Experiments on the inspection of power line assets dataset(InsPLAD)dataset show that the integrated approach achieves a component detection accuracy at intersection over union(IoU)=0.5(CDA_(50))of 88.3%and a component detection robustness(CDR_(50:95))of 69.8%,respectively,improvements of 4.4%and 7.0%over the baseline.
基金Project(51678075)supported by the National Natural Science Foundation of ChinaProject(2017GK2271)supported by Hunan Provincial Science and Technology Department,China
文摘This paper proposed a novel multi-view interactive behavior recognition method based on local self-similarity descriptors and graph shared multi-task learning. First, we proposed the composite interactive feature representation which encodes both the spatial distribution of local motion of interest points and their contexts. Furthermore, local self-similarity descriptor represented by temporal-pyramid bag of words(BOW) was applied to decreasing the influence of observation angle change on recognition and retaining the temporal information. For the purpose of exploring latent correlation between different interactive behaviors from different views and retaining specific information of each behaviors, graph shared multi-task learning was used to learn the corresponding interactive behavior recognition model. Experiment results showed the effectiveness of the proposed method in comparison with other state-of-the-art methods on the public databases CASIA, i3Dpose dataset and self-built database for interactive behavior recognition.
基金National Key Research and Development Program of China(2020YFB1707700)National Natural Science Foundation of China(61972356,62036009)Fundamental Research Funds for the Provincial Universities of Zhejiang,China(RF-A2020001).
文摘Acoustic quality detection is vital in the manufactured products quality control field since it represents the conditions of machines or products.Recent work employed machine learning models in manufactured audio data to detect anomalous patterns.A major challenge is how to select applicable audio features to meliorate model’s accuracy and precision.To relax this challenge,we extract and analyze three audio feature types including Time Domain Feature,Frequency Domain Feature,and Cepstrum Feature to help identify the potential linear and non-linear relationships.In addition,we design a visual analysis system,namely AFExplorer,to assist data scientists in extracting audio features and selecting potential feature combinations.AFExplorer integrates four main views to present detailed distribution and relevance of the audio features,which helps users observe the impact of features visually in the feature selection.We perform the case study with AFExplore according to the ToyADMOS and MIMII Dataset to demonstrate the usability and effectiveness of the proposed system.
文摘In recent years,deep learning has been widely applied in the fields of recommendation systems and click-through rate(CTR)prediction,and thus recommendation models incorporating deep learning have emerged.In addition,the design and implementation of recommendation models using information related to user behavior sequences is an important direction of current research in recommendation systems,and models calculate the likelihood of users clicking on target items based on their behavior sequence information.In order to explore the relationship between features,this paper improves and optimizes on the basis of deep interest network(DIN)proposed by Ali’s team.Based on the user behavioral sequences information,the attentional factorization machine(AFM)is integrated to obtain richer and more accurate behavioral sequence information.In addition,this paper designs a new way of calculating attention weights,which uses the relationship between the cosine similarity of any two vectors and the absolute value of their modal length difference to measure their relevance degree.Thus,a novel deep learning CTR prediction mode is proposed,that is,the CTR prediction network based on user behavior sequence and feature interactions deep interest and machines network(DIMN).We conduct extensive comparison experiments on three public datasets and one private music dataset,which are more recognized in the industry,and the results show that the DIMN obtains a better performance compared with the classical CTR prediction model.
文摘This paper presents a feature-based method for machining process planning in integrated product designing and manufacturing system for CE(Concurrent Engineering) application. The feature setup generation and machining sequence can be determined automatically in this system. The set of knowledge-based rules for process planning and manufacturability evaluation is provided and can be shared by all stages of full product life-cycle. An approach for MTAD (Multiple Tool Axis Direction) feature setup generation is presented and the appropriate Tool Axis Direction(TAD) is chosen to minimize the total setup numbers of a part. The classification and process planning of interacting feature are discussed and the knowledge-based rules are used to solve the feature interaction problem.
文摘This paper presents methodologies and technologies of feature_based integrated product designing and manufacturing system for CE(Concurrent Engineering) application. The product information is represented on the basis of hierarchical and dynamic structure of feature representation. The Object_Oriented feature modeling method is adopted to represent the feature classification, feature relationship and feature interaction. The set of knowledge_based rule for process planing and manufacturiability evaluation is provided and can be shared by all stages of full product life_cycle. The feature_based machining operation and machining sequence can be determined automatically. The machining process of the machining feature can be determined according to the set of knowledge_based rule.
文摘Online advertising click-through rate(CTR) prediction is aimed at predicting the probability of a user clicking an ad,and it has undergone considerable development in recent years.One of the hot topics in this area is the construction of feature interactions to facilitate accurate prediction.Factorization machine provides second-order feature interactions by linearly multiplying hidden feature factors.However,real-world data present a complex and nonlinear structure.Hence,second-order feature interactions are unable to represent cross information adequately.This drawback has been addressed using deep neural networks(DNNs),which enable high-order nonlinear feature interactions.However,DNN-based feature interactions cannot easily optimize deep structures because of the absence of cross information in the original features.In this study,we propose an effective CTR prediction algorithm called CAN,which explicitly exploits the benefits of attention mechanisms and DNN models.The attention mechanism is used to provide rich and expressive low-order feature interactions and facilitate the optimization of DNN-based predictors that implicitly incorporate high-order nonlinear feature interactions.The experiments using two real datasets demonstrate that our proposed CAN model performs better than other cross feature-and DNN-based predictors.
基金supported by the National Science and Technology Major Project(No.2022YFB4500602).
文摘Gaze following aims to interpret human-scene interactions by predicting the person’s focal point of gaze.Prevailing approaches often adopt a two-stage framework,whereby multi-modality information is extracted in the initial stage for gaze target prediction.Consequently,the efficacy of these methods highly depends on the precision of the previous modality extraction.Others use a single-modality approach with complex decoders,increasing network computational load.Inspired by the remarkable success of pre-trained plain vision transformers(ViTs),we introduce a novel single-modality gaze following framework called ViTGaze.In contrast to previous methods,it creates a novel gaze following framework based mainly on powerful encoders(relative decoder parameters less than 1%).Our principal insight is that the inter-token interactions within self-attention can be transferred to interactions between humans and scenes.Leveraging this presumption,we formulate a framework consisting of a 4D interaction encoder and a 2D spatial guidance module to extract human-scene interaction information from self-attention maps.Furthermore,our investigation reveals that ViT with self-supervised pre-training has an enhanced ability to extract correlation information.Many experiments have been conducted to demonstrate the performance of the proposed method.Our method achieves state-of-the-art performance among all single-modality methods(3.4%improvement in the area under curve score,5.1% improvement in the average precision)and very comparable performance against multi-modality methods with 59% fewer parameters.
文摘Multimodal Sentiment analysis refers to analyzing emotions in infor-mation carriers containing multiple modalities.To better analyze the features within and between modalities and solve the problem of incomplete multimodal feature fusion,this paper proposes a multimodal sentiment analysis model MIF(Modal Interactive Feature Encoder For Multimodal Sentiment Analysis).First,the global features of three modalities are obtained through unimodal feature extraction networks.Second,the inter-modal interactive feature encoder and the intra-modal interactive feature encoder extract similarity features between modal-ities and intra-modal special features separately.Finally,unimodal special features and the interaction information between modalities are decoded to get the fusion features and predict sentimental polarity results.We conduct extensive experi-ments on three public multimodal datasets,including one in Chinese and two in English.The results show that the performance of our approach is significantly improved compared with benchmark models.
文摘Correction to:GraphFM:Graph Factorization Machines for Feature Interaction Modelling DOI:10.1007/s11633-024-1505-5 Authors:Shu Wu,Zekun Li,Yunyue Su,Zeyu Cui,Xiaoyu Zhang,Liang Wang The article GraphFM:Graph Factorization Machines for Feature Interaction Modelling,written by Shu Wu,Zekun Li,Yunyue Su,Zeyu Cui,Xiaoyu Zhang,Liang Wang,was originally published without Open Access.After publication,the authors decided to opt for Open Choice and to make the article an Open Access publication.
基金supported by the National Science Foundation of China(No.62141608).
文摘Factorization machine(FM)is a prevalent approach to modelling pairwise(second-order)feature interactions when dealing with high-dimensional sparse data.However,on the one hand,FMs fail to capture higher-order feature interactions suffering from combinatorial expansion.On the other hand,taking into account interactions between every pair of features may introduce noise and degrade the prediction accuracy.To solve these problems,we propose a novel approach,the graph factorization machine(GraphFM),which naturally represents features in the graph structure.In particular,we design a mechanism to select beneficial feature interactions and formulate them as edges between features.Then the proposed model,which integrates the interaction function of the FM into the feature aggregation strategy of the graph neural network(GNN),can model arbitrary-order feature interactions on graph-structured features by stacking layers.Experimental results on several real-world datasets demonstrate the rationality and effectiveness of our proposed approach.The code and data are available at https://github.com/CRIPAC-DIG/GraphCTR.
基金supported by National Key R&D Program of China under Grant No.2022YFB3305700Shanghai Science Innovation Action Plan under Grant No.21511104302.
文摘Object detection serves as a challenging yet crucial task in computer vision.Despite significant advancements,modern detectors remain struggling with task alignment between localization and classification.In this paper,Global Collaborative Learning(GCL)is introduced to address these challenges from often-overlooked perspectives.First,the essence of GCL is reflected in the label assignment of the detector.Adjusting the loss function to transform samples with strong localization yet weak classification into high-quality samples in both tasks,provides more effective training signals,enabling the model to capture key consistent features.Second,the spirit of GCL is embodied in the head design.By enabling global feature interaction within the decoupled head,the approach ensures that final predictions are made more comprehensively and robustly,thereby preventing the two independent branches from converging into suboptimal solutions for their respective tasks.Extensive experiments on the challenging MS COCO and CrowdHuman datasets demonstrate that the proposed GCL method substantially enhances performance and generalization capabilities.
基金co-supported by the National Natural Science Foundation of China(U22A20176)Guangdong Basic and Applied Basic Research Foundation,China(2022B1515120078)+2 种基金the Guangdong Basic and Applied Basic Research Foundation,China(2021A1515110898)GDAS’Project of Science and Technology Development,China(2022GDASZH-2022010108)the Key Areas R&D Program of Dongguan City,China(20201200300062).
文摘With the rapid advancement of mechanical automation and intelligent processing technology,ac-curately measuring the surfaces of complex parts has emerged as a significant research challenge.Robotic measurement technology plays a crucial role in facilitating rapid quality inspections during the manufacturing process due to its inherent flexibility.However,the irregular shapes and viewpoint occlusions of complex parts complicate precise measurement.To address these challenges,this work proposes a point cloud registration network for robotic scanning systems and introduces a DBR-Net(Dual-line Registration Network)to overcome the issues of low overlap rates and perspective occlusion that currently impede the registration of certain workpieces.First,feature extraction is performed using a bilinear encoder and multi-level feature interactions of both point-wise and global features.Subsequently,the features are sampled through unanimous voting and fed into the RANSAC(Random Sample Consensus)algorithm for pose estimation,enabling multi-view point cloud registration.Experimental results demonstrate that this method significantly outperforms many existing techniques in terms of feature extraction and registration accuracy,thereby enhancing the overall performance of point cloud registration.
基金support of National Key Research and Development Program of China(2023YFE0120700)National Natural Science Foundation of China(52274041 and 52304023)+2 种基金Distinguished Young Sichuan Science Scholars(2023NSFSC1954)Natural Science Foundation of Chongqing(CSTB2022NSCQ-MSX0403)Chongqing Municipal Support Program for Overseas Students Returning for Entrepreneurship and Innovation(2205012980950154).
文摘CO_(2) injection not only effectively enhances oil recovery(EOR)but also facilitates CO_(2) utilization and storage.Rapid screening and optimization of CO_(2)-EOR operations is urgently needed for unconventional reservoirs.However,it remains challenging due to a limited understanding of fluid flow in multiscale porous media and the problem complexity invoked by numerous factors.This work developed a new interpretable machine learning(ML)framework to specifically address this issue.Three different methods,namely random forest(RF),support vector regression(SVR),and artificial neural network(ANN),were used to establish proxy models using the data from a specific unconventional reservoir,and the RF model demonstrated a preferable performance.To enhance the interpretability of the established models,the multiway feature importance analysis and Shapley Additive Explanations(SHAP)were proposed to quantify the contribution of individual features to the model output.Based on the results of model interpretability,the genetic algorithm(GA)was coupled with RF(RF-GA model)to optimize the CO_(2)-EOR process.The proposed framework was validated by comparing the GA-RF predictions with simulation results under different reservoir conditions,which yielded a minimum relative error of 0.34%and an average relative error of 5.3%.The developed interpretable ML method was capable of rapidly screening suitable CO_(2)-EOR strategies based on reservoir conditions and provided a practical example for field applications.
基金supported by the National Natural Science Foundation of China(62201454,62306235)the Xi'an Municipal Science and Technology Program Projects(23SFSF0004)the Yunnan Key Laboratory of Unmanned Autonomous Systems(202501YB01)。
文摘Object detection in unmanned aerial vehicle(UAV)images plays an important role in environmental monitoring,post-disaster rescue and other fields.However,the complex context and small size of typical objects in UAV images pose great challenges to existing object detection algorithms.To address these issues,a new object detection method based on improved YOLOv7 with higher-order feature interaction(HOFI-YOLOv7)is proposed for UAV aerial images.Firstly,a new higher-order feature interaction(HOFI)module was constructed to promote the interaction of feature information.And then,a feature fusion network with effective channel attention(ECA)module was designed to improve the focus of the method on small objects.Finally,a focal extended intersection over union(Focal-EIOU)loss function is utilized to improve the detection accuracy of small objects.The experimental results on the VisDrone2019 dataset show that compared with the original model,the mAP50 and mAP50∶5∶95 of the improved model are increased by 3.37%and 1.77%,respectively.Furthermore,the proposed method exhibits superior generalization performance on the self-picked dataset.
基金supported by the National Key Research and Development Program of China(No.2023YFB2504704)the Natural Science Foundation in Hunan Province of China(No.S2023JJQNJJ1969).
文摘Real-time conflict prediction at signalized intersections is crucial for urban road safety management.This study developed a real-time conflict prediction framework for signal-ized intersections using real-time video data recognition technology and deep learning techniques,incorporating lane-level information and feature interactions.The modeling framework consists of three stages:real-time video data extraction and processing,the development of a deep and cross network(DCN)-based real-time traffic conflict prediction model,and conflict-driven factor interpretability analysis through Shapley additive expla-nations(SHAP).In the first stage,an efficient automated trajectory extraction system is designed to obtain vehicle trajectories in real time for dynamic traffic parameters and con-flict frequency identification.In the second stage,a DCN model is developed to construct the relationships between dynamic traffic parameters,including their interactions,and traffic conflicts.In the third stage,SHAP explores the impact mechanisms of different dynamic traffic parameters on traffic conflicts.The model’s predictive performance and interpretability are evaluated using intersection video data from Changsha City,China.The results are as follows.(1)In real-time traffic conflict prediction at signalized intersec-tions across different modified time-to-conflict thresholds(1.5 s and 3.0 s),the DCN model consistently outperformed statistical and machine learning models.(2)High traffic flows on main and secondary roads at signalized intersections significantly increase the com-plexity and frequency of conflicts,with varying sensitivity depending on the interaction of traffic flow,speed,and platoon length.(3)The proposed framework provides a safety measurement standard for data-driven road safety management methods.
