High-dimensional Feature Research Articles

BCED-Net: Breast Cancer Ensemble Diagnosis Network using transfer learning and the XGBoost classifier with mammography images.

Breast cancer poses a significant global health challenge, characterized by complex origins and the potential for life-threatening metastasis. The critical need for early and accurate detection is underscored by the 685,000 lives claimed by the disease worldwide in 2020. Deep learning has made strides in advancing the prompt diagnosis of breast cancer. However, obstacles persist, such as dealing with high-dimensional data and the risk of overfitting, necessitating fresh approaches to improve accuracy and real-world applicability. In response to these challenges, we propose BCED-Net, which stands for Breast Cancer Ensemble Diagnosis Network. This innovative framework leverages transfer learning and the extreme gradient boosting (XGBoost) classifier on the Breast Cancer RSNA dataset. Our methodology involved feature extraction using pre-trained models-namely, Resnet50, EfficientnetB3, VGG19, Densenet121, and ConvNeXtTiny-followed by the concatenation of the extracted features. Our most promising configuration combined features extracted from deep convolutional neural networks-namely Resnet50, EfficientnetB3, and ConvNeXtTiny-that were classified using the XGBoost classifier. The ensemble approach demonstrated strong overall performance with an accuracy of 0.89. The precision, recall, and F1-score values, which were all at 0.86, highlight a balanced trade-off between correctly identified positive instances and the ability to capture all actual positive samples. BCED-Net represents a significant leap forward in addressing persistent issues such as the high dimensionality of features and the risk of overfitting.

Prediction of PM10 Concentration in Dry Bulk Ports Using a Combined Deep Learning Model Considering Feature Meteorological Factors

Accurate prediction of PM10 concentration is important for effectively managing PM10 exposure and mitigating health and economic risks posed to humans in dry bulk ports. However, accurately capturing the time-series nonlinear variation characteristics of PM10 concentration is challenging owing to the specific intensity of port operation activities and the influence of meteorological factors. To address such challenges, a novel combined deep learning model （CLAF） was proposed, merging cascaded convolutional neural networks （CNN）, long short-term memory （LSTM）, and an attention mechanism （AM）. This integrated model aimed to forecast hourly PM10 concentration in dry bulk ports. First, using the random forest characteristic importance algorithm, the distinct meteorological factors were identified among the selected five meteorological factors. These selected factors were incorporated into the prediction model along with the PM10 concentration. Subsequently, the CNN layer was employed to extract high-dimensional time-varying features from the input variables, while the LSTM layer captured sequential features and long-term dependencies. In the AM layer, different weights were assigned to the output components of the LSTM layer to amplify the effects of important information. Finally, three evaluation metrics were applied to compare the performance of the CLAF model with three basic models and three commonly used prediction models. Real-case data was collected and used in this study. Comparison results demonstrated that considering the meteorological factors could improve the prediction accuracy and fitting performance of PM10 concentration in ports. The CLAF model reduced the mean absolute error statistic by 13.92%-56.9%, decreased the mean square error statistic by 45.99%-81.02%, and improved the goodness-of-fit statistic by 3.2%-15.5%.

Refined Prior Guided Category-Level 6D Pose Estimation and Its Application on Robotic Grasping

Estimating the 6D pose and size of objects is crucial in the task of visual grasping for robotic arms. Most current algorithms still require the 3D CAD model of the target object to match with the detected points, and they are unable to predict the object’s size, which significantly limits the generalizability of these methods. In this paper, we introduce category priors and extract high-dimensional abstract features from both the observed point cloud and the prior to predict the deformation matrix of the reconstructed point cloud and the dense correspondence between the reconstructed and observed point clouds. Furthermore, we propose a staged geometric correction and dense correspondence refinement mechanism to enhance the accuracy of regression. In addition, a novel lightweight attention module is introduced to further integrate the extracted features and identify potential correlations between the observed point cloud and the category prior. Ultimately, the object’s translation, rotation, and size are obtained by mapping the reconstructed point cloud to a normalized canonical coordinate system. Through extensive experiments, we demonstrate that our algorithm outperforms existing methods in terms of performance and accuracy on commonly used benchmarks for this type of problem. Additionally, we implement the algorithm in robotic arm-grasping simulations, further validating its effectiveness.

Prior knowledge embedding convolutional autoencoder: A single-source domain generalized fault diagnosis framework under small samples

The proposed transfer learning-based fault diagnosis models have achieved good results in multi-source domain generalization (MDG) tasks. However, research on single-source domain generalization (SDG) is relatively scarce, and the limited availability of small training samples is seldom considered. Therefore, to address the insufficient feature extraction capability and poor generalization performance of existing models on single-source domain small sample data, a novel single-source domain generalization fault diagnosis (SDGFD) framework, the prior knowledge embedded convolutional autoencoder (PKECA), is proposed. During the training phase, first, single-source domain data are used to construct prior features based on the time domain, frequency domain, and time-frequency domain. Second, a prior knowledge embedding structure based on the convolutional autoencoder is built, which compresses the prior knowledge and original vibration data into a high-dimensional space of consistent dimensions, embedding the prior knowledge into the features corresponding to the vibration data using a mean squared error loss function. Subsequently, the proposed centroid-based self-supervised learning (CBSSL) strategy further constrains high-dimensional features, improving the generalization ability. The designed sparse regularized activation (SRA) function significantly enhances the regularization effect on features. During the testing phase, it is only necessary to input the data from the unknown domain to identify the fault types. The experimental results show that the proposed method achieves superior performance in fault diagnosis tasks involving cross-speed, time-varying speed, and small sample data in SDGFD, demonstrating that PKECA has strong generalizability. The code can be found here: https://github.com/John-520/PKECA. © 2024 Elsevier Science. All rights reserved

High-dimensional Iterative Causal Forest (hdiCF) for Subgroup Identification Using Health Care Claims Data.

We recently developed a machine-learning subgrouping algorithm, iterative causal forest (iCF), to identify subgroups with heterogeneous treatment effects (HTEs) using predefined covariates. However, such predefined covariates may miss or poorly define important features leading to inaccurate subgrouping. To address such limitations, we developed a new semi-automatic subgrouping algorithm, hdiCF, which adapts methodology from high-dimensional propensity score for feature recognition in claims data. The hdiCF algorithm has 3 steps: 1) high-dimensional feature identification by International Classification of Diseases, Current Procedural Terminology, and Anatomical Therapeutic Chemical codes (in/outpatient diagnoses, procedures, prescriptions) and creation of ordinal variables by frequency of occurrence; 2) propensity score trimming and high-dimensional feature preparation; 3) iCF implementation to identify subgroups. We applied hdiCF in a 20% random sample of fee-for-service Medicare beneficiaries who initiated sodium-glucose cotransporter-2 inhibitors (SGLT2i) or glucagon-like peptide-1 receptor agonists to identify subgroups with HTEs for incidence of hospitalized heart failure. HdiCF findings were consistent with studies suggesting SGLT2i to be more beneficial for patients with pre-existing heart failure or chronic kidney disease. HdiCF is not dependent on prior hypotheses about HTEs and identifies subgroups with markers for potential HTEs in real-world evidence studies where active-comparator, new-user study designs limit the potential for unmeasured confounding.

Scene understanding method utilizing global visual and spatial interaction features for safety production

Risk identification in power operations is crucial for both personal safety and power production. Existing risk identification methods mainly use target detection models to identify the common risks but the scene specificity of risk occurrence. For example, not wearing a safety harness, not wearing insulated gloves, etc. Since most methods for detecting safety gears make sense only under specific scene. But the power electric work is a complex object involving many elements such as personnel, equipment and safety tools. Therefore, this paper proposes a scene understanding method that integrates visual features and spatial relationship features among scene elements. This method constructs a scenean undirected scene graph to represent the interactive relationship among the elements, extracts the interactive features by using a graph encoder-decoder convolution module, and fuse perceived high-dimensional visual features and spatial topological features for scene recognition, in order to effectively solve addressing the power operation scene understanding problem under multi-element interaction. Finally, a power inspection operation scenario was chosen as the test case. The outcome of the evaluation indicates results indicate that the proposed approach suggested in this study exhibits superior precision in scene identification and shows ademonstrates strong generalization ability.

Data-Driven Clustering and Feature-Based Retail Electricity Pricing with Smart Meters

The adoption of smart meters and dynamic pricing programs is rapidly increasing among electric utility companies. In “Data-Driven Clustering and Feature-Based Retail Electricity Pricing with Smart Meters,” Keskin, Li, and Sunar analyze how utility companies should use smart meter data for better pricing decisions. Utility companies typically have access to consumption patterns and high-dimensional features on customer characteristics and exogenous factors. The authors identify that such feature data can exhibit different forms of heterogeneity—over time and over customers. They show that the different forms of feature heterogeneity significantly worsen the best profit performance that can be achieved by a data-driven dynamic pricing policy. The authors also develop a policy based on joint spectral clustering and contextual dynamic pricing and prove that this policy achieves near-optimal profit performance.

ECF-Net: Enhanced, Channel-Based, Multi-Scale Feature Fusion Network for COVID-19 Image Segmentation

Accurate segmentation of COVID-19 lesion regions in lung CT images aids physicians in analyzing and diagnosing patients’ conditions. However, the varying morphology and blurred contours of these regions make this task complex and challenging. Existing methods utilizing Transformer architecture lack attention to local features, leading to the loss of detailed information in tiny lesion regions. To address these issues, we propose a multi-scale feature fusion network, ECF-Net, based on channel enhancement. Specifically, we leverage the learning capabilities of both CNN and Transformer architectures to design parallel channel extraction blocks in three different ways, effectively capturing diverse lesion features. Additionally, to minimize irrelevant information in the high-dimensional feature space and focus the network on useful and critical information, we develop adaptive feature generation blocks. Lastly, a bidirectional pyramid-structured feature fusion approach is introduced to integrate features at different levels, enhancing the diversity of feature representations and improving segmentation accuracy for lesions of various scales. The proposed method is tested on four COVID-19 datasets, demonstrating mIoU values of 84.36%, 87.15%, 83.73%, and 75.58%, respectively, outperforming several current state-of-the-art methods and exhibiting excellent segmentation performance. These findings provide robust technical support for medical image segmentation in clinical practice.

Magnetic Resonance-Guided Cancer Therapy Radiomics and Machine Learning Models for Response Prediction.

Magnetic resonance imaging (MRI) is known for its accurate soft tissue delineation of tumors and normal tissues. This development has significantly impacted the imaging and treatment of cancers. Radiomics is the process of extracting high-dimensional features from medical images. Several studies have shown that these extracted features may be used to build machine-learning models for the prediction of treatment outcomes of cancer patients. Various feature selection techniques and machine models interrogate the relevant radiomics features for predicting cancer treatment outcomes. This study aims to provide an overview of MRI radiomics features used in predicting clinical treatment outcomes with machine learning techniques. The review includes examples from different disease sites. It will also discuss the impact of magnetic field strength, sample size, and other characteristics on outcome prediction performance.

The superior value of radiomics to sonographic assessment for ultrasound-based evaluation of extrathyroidal extension in papillary thyroid carcinoma: a retrospective study.

Extrathyroidal extension was related with worse survival for patients with papillary thyroid carcinoma. For its preoperative evaluation, we measured and compared the predicting value of sonographic method and ultrasonic radiomics method in nodules of papillary thyroid carcinoma. Data from 337 nodules were included and divided into training group and validation group. For ultrasonic radiomics method, a best model was constructed based on clinical characteristics and ultrasonic radiomic features. The predicting value was calculated then. For sonographic method, the results were calculated using all samples. For ultrasonic radiomics method, we constructed 9 models and selected the extreme gradient boosting model for its highest accuracy (0.77) and area under curve (0.813) in validation group. The accuracy and area under curve of sonographic method was 0.70 and 0.569. Meanwhile. We found that the top-6 important features of xgboost model included no clinical characteristics, all of whom were high-dimensional radiomic features. The study showed the superior value of ultrasonic radiomics method to sonographic method for preoperative detection of extrathyroidal extension in papillary thyroid carcinoma. Furthermore, high-dimensional radiomic features were more important than clinical characteristics.

HEnsem_DTIs: A heterogeneous ensemble learning model for drug-target interactions prediction

Drug discovery is the process by which a drug is discovered. Drug-target interactions prediction is a major part of drug discovery. Unfortunately, producing new drugs is time-consuming and expensive; Because it requires a lot of human and laboratory resources. Recently, predictions have been made using computational methods to solve these problems and prevent blindly examining all interactions. Various experiences using computational methods show that no single algorithm can be suitable for all applications; Hence, ensemble learning is expressed. Although various ensemble methods have been proposed, it is still not easy to find a suitable ensemble method for a particular dataset. In general, the existing algorithms in aggregation and combination method are selected manually based on experience. Reinforcement learning can be one way to meet this challenge. High-dimensional feature space and class imbalance are among the challenges of drug-target interactions prediction. This paper proposes HEnsem_DTIs, a heterogeneous ensemble model, for predicting drug-target interactions using dimensionality reduction and concepts of recommender systems to address these challenges. HEnsem_DTIs is configured with reinforcement learning. Dimensionality reduction is applied to handle the challenge of high-dimensional feature space and recommender systems to improve under-sampling and solve the class imbalance challenge. Six datasets are used to evaluate the proposed model; Results of the evaluation on datasets show that HEnsem_DTIs works better than other models in this field. Results of evaluation of the proposed model on the first dataset using 10-fold cross-validation experiments show the amount of sensitivity 0.896, specificity 0.954, GM 0.924, AUC 0.930 and AUPR 0.935.

PSEENet: A Pseudo-Siamese Neural Network Incorporating Electroencephalography and Electrooculography Characteristics for Heterogeneous Sleep Staging.

Sleep staging plays a critical role in evaluating the quality of sleep. Currently, most studies are either suffering from dramatic performance drops when coping with varying input modalities or unable to handle heterogeneous signals. To handle heterogeneous signals and guarantee favorable sleep staging performance when a single modality is available, a pseudo-siamese neural network (PSN) to incorporate electroencephalography (EEG), electrooculography (EOG) characteristics is proposed (PSEENet). PSEENet consists of two parts, spatial mapping modules (SMMs) and a weight-shared classifier. SMMs are used to extract high-dimensional features. Meanwhile, joint linkages among multi-modalities are provided by quantifying the similarity of features. Finally, with the cooperation of heterogeneous characteristics, associations within various sleep stages can be established by the classifier. The evaluation of the model is validated on two public datasets, namely, Montreal Archive of Sleep Studies (MASS) and SleepEDFX, and one clinical dataset from Huashan Hospital of Fudan University (HSFU). Experimental results show that the model can handle heterogeneous signals, provide superior results under multimodal signals and show good performance with single modality. PSEENet obtains accuracy of 79.1%, 82.1% with EEG, EEG and EOG on Sleep-EDFX, and significantly improves the accuracy with EOG from 73.7% to 76% by introducing similarity information.

Seagrass classification using unsupervised curriculum learning (UCL)

Seagrass ecosystems are pivotal in marine environments, serving as crucial habitats for diverse marine species and contributing significantly to carbon sequestration. Accurate classification of seagrass species from underwater images is imperative for monitoring and preserving these ecosystems. This paper introduces Unsupervised Curriculum Learning (UCL) to seagrass classification using the DeepSeagrass dataset. UCL progressively learns from simpler to more complex examples, enhancing the model's ability to discern seagrass features in a curriculum-driven manner. Experiments employing state-of-the-art deep learning architectures, convolutional neural networks (CNNs), show that UCL achieved overall 90.12 % precision and 89 % recall, which significantly improves classification accuracy and robustness, outperforming some traditional supervised learning approaches like SimCLR, and unsupervised approaches like Zero-shot CLIP. The methodology of UCL involves four main steps: high-dimensional feature extraction, pseudo-label generation through clustering, reliable sample selection, and fine-tuning the model. The iterative UCL framework refines CNN's learning of underwater images, demonstrating superior accuracy, generalization, and adaptability to unseen seagrass and background samples of undersea images. The findings presented in this paper contribute to the advancement of seagrass classification techniques, providing valuable insights into the conservation and management of marine ecosystems. The code and dataset are made publicly available and can be assessed here: https://github.com/nabid69/Unsupervised-Curriculum-Learning—UCL.

Non-Intrusive Load Monitoring Based on Dimensionality Reduction and Adapted Spatial Clustering

Non-invasive load monitoring (NILM) deduces changes in energy consumption patterns and operational statuses of electrical equipment from power signals in the feed line. With the emergence of fine-grained power load distribution, the importance of utilizing this technology for implementing demand-side energy management in smart grid development has become increasingly prominent. To address the issue of low load identification accuracy stemming from complex and diverse load types, this paper introduces a NILM method based on uniform manifold approximation and projection (UMAP) reduction and enhanced density-based spatial clustering of applications with noise (DBSCAN). Firstly, this paper combines the characteristics of user load under transient and steady-state conditions and selects data with significant differences to construct a load-characteristic database. Additionally, UMAP is employed to reduce the dimensionality of high-dimensional load features and rebuild a load feature database. Subsequently, DBSCAN is utilized to categorize typical user loads, followed by a correlation analysis with the load-characteristic database to determine the types or classes of loads that involve switching actions. Finally, this paper simulates and analyzes the proposed method using the electricity consumption data of industrial users from the CER–Electricity–Data dataset. It identifies the electricity load data commonly utilized by users in a specific area of Zhejiang Province in China. The experimental results indicate that the accuracy of the proposed non-invasive load identification method reaches 95%. Compared to the wavelet transform, decision tree, and backpropagation network methods, the improvement is approximately 5%.

Intrusion Event Classification of a Drainage Tunnel Based on Principal Component Analysis and Neural Networking

Drainage tunnel stability is crucial for engineering project safety (e.g., mine engineering and dams), and rockfall events and water release are key indicators of drainage tunnel stability. To address this, we developed a monitoring system to simulate drainage tunnel intrusions based on distributed acoustic sensing (DAS), and we obtained typical characteristics of events like rockfall events and water release. Given the multitude of DAS signal feature parameters and challenges, such as high-dimensional features impacting the classification accuracy of machine learning, we proposed an identification method for drainage tunnel intrusion events using principal component analysis (PCA) and neural networks. PCA reveals that amplitude-related parameters—amplitude, mean amplitude, and energy—significantly contribute to DAS signal classification, reducing the feature parameter dimensions by 54.8%. The accuracy of intrusion event classification improves with PCA-processed data compared to unprocessed data, with overall accuracy rates of 79.1% for rockfall events and 72.7% for water release events. Additionally, the artificial neural network model outperforms the Bayesian and logistic regression models, demonstrating that ANN has advantages in handling complex models for intrusion event classification.

Interpretable semi-supervised clustering enables universal detection and intensity assessment of diverse aviation hazardous winds

The identification of aviation hazardous winds is crucial and challenging in air traffic management for assuring flight safety, particularly during the take-off and landing phases. Existing criteria are typically tailored for special wind types, and whether there exists a universal feature that can effectively detect diverse types of hazardous winds from radar/lidar observations remains as an open question. Here we propose an interpretable semi-supervised clustering paradigm to solve this problem, where the prior knowledge and probabilistic models of winds are integrated to overcome the bottleneck of scarce labels (pilot reports). Based on this paradigm, a set of high-dimensional hazard features is constructed to effectively identify the occurrence of diverse hazardous winds and assess the intensity metrics. Verification of the paradigm across various scenarios has highlighted its high adaptability to diverse input data and good generalizability to diverse geographical and climate zones.

An adaptive detection framework based on artificial immune for IoT intrusion detection system

Given the continual evolution of new network attack methodologies, defenders face the imperative of constantly upgrading security defenses. Current security technologies, albeit effective against known threats, often fall short in handling the intricacies of diverse and novel attacks. Artificial immunity-based network anomaly detection offers a promising avenue by dynamically adapting to evolving threats. However, prevailing algorithms in this domain suffer from low detection rates, limited adaptability, and extended detector generation times. This study aims to tackle these challenges by introducing a high-efficiency network anomaly detection framework, emphasizing both high-dimensional feature selection and adaptive detector generation. Our approach begins with an enhanced dual-module hybrid high-dimensional feature selection method, leveraging evolutionary principles. Furthermore, we introduce a self-sample clustering algorithm based on fuzzy clustering during the tolerance stage, enhancing detector tolerance efficiency. Additionally, an adaptive detector generation scheme is devised. It divides the non-boundary sub-population based on non-self differences and evolution, while employing the red fox optimization algorithm in the boundary region. This adaptive approach dynamically adjusts detector positions and radii to derive optimal detectors. Through comprehensive validation using well-established IoT and network anomaly datasets, our proposed artificial immunity-based IoT intrusion detection framework exhibits superior performance. It achieves higher classification accuracy and lower error rates compared to current state-of-the-art machine learning and artificial immunity algorithms.

Cross-domain manifold structure preservation for transferable and cross-machine fault diagnosis

To address the decline or failure in the autonomous learning capability of traditional transfer learning methods when training and test samples come from different machines, resulting in low cross-machine fault diagnosis rates, we propose a cross-domain manifold structure preservation (CDMSP) method for diagnosing rolling bearing faults across machines. The CDMSP method can induce the manifold space projection matrices of the source and target domains more effectively. This method maps high-dimensional features into a low-dimensional manifold, preserving non-linear relationships and aligning distribution differences while maintaining cross-domain manifold structure consistency. Additionally, highly confidently labeled target domain samples are selected from each mapping result and added to the training dataset to enhance subspace learning in subsequent iterations. The CDMSP method is both simple and effective at capturing the underlying structures and patterns in the data. The CWRU dataset and our self-built test platform dataset were used to validate this method. Experimental results show that CDMSP, as a non-deep domain adaptation method of transfer learning, outperforms similar methods in cross-machine fault identification, achieving a maximum fault identification accuracy of 100 % with excellent convergence performance. Furthermore, simulated diagnostic experiments under noise interference indicate that CDMSP maintains high fault identification accuracy, even in noisy environments. Overall, CDMSP is an efficient and reliable new method for diagnosing cross-machine bearing faults.

Wushu Movement Recognition System Based on DTW Attitude Matching Algorithm

Motion recognition technology is widely used in intelligent video surveillance, human-computer interaction and other fields. With the development of computer vision technology, improving the accuracy and efficiency of motion recognition has become the focus of research. The purpose of this study is to improve the performance of Wushu movement recognition through improved dynamic time warping algorithm and hierarchical model. Firstly, a high-dimensional feature vector is constructed by using the position, velocity and Angle changes of human bone joints. The actions are subdivided by the hierarchical model, and matched and recognized by the max-minimum dynamic time regularization model. Meanwhile, the K-class mean algorithm is combined to optimize the type of tree core, improve the performance of the model, reduce the interference of noise nodes, and effectively classify Wushu actions. Experimental verification was carried out on four public data sets of KTH, Olympic Sports, Hollywood2 and HMDB51. The experimental results showed that the recognition rate of the proposed model in KTH data set was 95.2%, and that in Olympic Sports data set was 91.4%. The Hollywood2 dataset was 66.7%, and the HMDB51 dataset was 61.2%. Comparing the results of different algorithms, the proposed method improved the recognition performance by 10% compared with long short-term memory network and gated cycle unit. Compared with one-dimensional convolutional neural network, the time of the proposed method was 15s longer, but the recognition rate was 1.6% higher. The results showed that the proposed method had significant performance advantages in diverse and complex action recognition tasks. Meanwhile, the results emphasized the factors to be considered in the design of the model, demonstrating its effectiveness in the application of Wushu movement recognition.

ERT-GFAN: A multimodal drug–target interaction prediction model based on molecular biology and knowledge-enhanced attention mechanism

In drug discovery, precisely identifying drug–target interactions is crucial for finding new drugs and understanding drug mechanisms. Evolving drug/target heterogeneous data presents challenges in obtaining multimodal representation in drug–target prediction(DTI). To deal with this, we propose ‘ERT-GFAN’, a multimodal drug–target interaction prediction model inspired by molecular biology. Firstly, it integrates bio-inspired principles to obtain structure feature of drugs and targets using Extended Connectivity Fingerprints(ECFP). Simultaneously, the knowledge graph embedding model RotatE is employed to discover the interaction feature of drug–target pairs. Subsequently, Transformer is utilized to refine the contextual neighborhood features from the obtained structure feature and interaction features, and multi-modal high-dimensional fusion features of the three-modal information constructed. Finally, the final DTI prediction results are outputted by integrating the multimodal fusion features into a graphical high-dimensional fusion feature attention network (GFAN) using our innovative multimodal high-dimensional fusion feature attention. This multimodal approach offers a comprehensive understanding of drug–target interactions, addressing challenges in complex knowledge graphs. By combining structure feature, interaction feature, and contextual neighborhood features, ‘ERT-GFAN’ excels in predicting DTI. Empirical evaluations on three datasets demonstrate our method’s superior performance, with AUC of 0.9739, 0.9862, and 0.9667, AUPR of 0.9598, 0.9789, and 0.9750, and Mean Reciprocal Rank(MRR) of 0.7386, 0.7035, and 0.7133. Ablation studies show over a 5% improvement in predictive performance compared to baseline unimodal and bimodal models. These results, along with detailed case studies, highlight the efficacy and robustness of our approach.