Structural Information Research Articles

Open-Weight Language Models and Retrieval Augmented Generation for Automated Structured Data Extraction from Diagnostic Reports: Assessment of Approaches and Parameters.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop and evaluate an automated system for extracting structured clinical information from unstructured radiology and pathology reports using open-weights language models (LMs) and retrieval augmented generation (RAG) and to assess the effects of model configuration variables on extraction performance. Materials and Methods This retrospective study utilized two datasets: 7,294 radiology reports annotated for Brain Tumor Reporting and Data System (BT-RADS) scores and 2,154 pathology reports annotated for IDH mutation status (January 2017 to July 2021). An automated pipeline was developed to benchmark the performance of various LMs and RAG configurations for structured data extraction accuracy from reports. The impact of model size, quantization, prompting strategies, output formatting, and inference parameters on model accuracy was systematically evaluated. Results The best performing models achieved up to 98% accuracy in extracting BT-RADS scores from radiology reports and over 90% for IDH mutation status extraction from pathology reports. The best model was medical finetuned llama3. Larger, newer, and domain fine-tuned models consistently outperformed older and smaller models (mean accuracy, 86% versus 75%; P < .001). Model quantization had minimal impact on performance. Few-shot prompting significantly improved accuracy (mean increase: 32% ± 32%, P = .02). RAG improved performance for complex pathology reports +48% ± 11% (P = .001), but not for shorter radiology reports-8% ± 31% (P = .39). Conclusion This study demonstrates the potential of open LMs in automated extraction of structured clinical data from unstructured clinical reports with local privacy-preserving application. Careful model selection, prompt engineering, and semiautomated optimization using annotated data are critical for optimal performance. ©RSNA, 2025.

ANALYSIS OF FARMERS INCOME SELLING IN SUCO LOUR ZUMALAI ADMINISTRATIVE POST-COVALIMA MUNICIPALITY

The research aims to analyze farmers' income selling Corn in Suco Lour, Zumalai Administrative Post, Covalima Municipality. The research employs descriptive methods, utilizing a simple random sampling technique to select a sample of 100 farmers from the population. The data collection involves three methods: interviews, which facilitate direct communication with the farming community; observation to witness farming activities firsthand; and questionnaires designed to gather structured information. The researcher analyzed with the results of counting through economic analysis equipment, responded with significant results as follows: (i) the total number of farmer family members responded in fifty (50), discovered farmers Secondary Education with its minimum frequency of two (2) people, 4 percent, discovered non-alphabetic farmers got the highest frequency of twenty-one (21) people, with 42%. (ii) The impact of technology on productivity: Investigate how the adoption of modern agricultural technologies influences corn production efficiency and income levels among farmers. The study highlights the factors influencing earnings, including production costs, market access, and pricing strategies. This research can inform policymakers and agricultural stakeholders about the economic challenges faced by local farmers, guiding the development of targeted interventions to enhance profitability and sustainability in the agricultural sector. Furthermore, the findings may serve as a basis for further studies on income diversification and agricultural practices in similar regions, ultimately contributing to improved livelihoods for farming communities.

WPR-Net: A Deep Learning Protocol for Highly Accelerated NMR Spectroscopy with Faithful Weak Peak Reconstruction.

Multidimensional NMR spectroscopy contains a large amount of molecular-level species and structure information, which is of great significance in various disciplines; however, it is unfortunately limited by lengthy acquisition times. Undersampling signals accompanied by spectral reconstruction provide a powerful and efficient way to accelerate its implementation. However, the accurate reconstruction of weak peaks remains a crucial issue to compromise the reconstruction performance. In this work, we introduce a deep learning architecture for highly accelerated NMR spectroscopy along with the reliable reconstruction of weak peaks. This deep learning protocol allows one to eliminate undersampled artifacts and reconstruct high-quality multidimensional NMR spectroscopy signals, even under the conditions of highly sparse sampling density or in the presence of severe noise. Therefore, this study provides a powerful tool for fast multidimensional NMR spectroscopy and presents meaningful application prospects toward broader chemical and biological applications.

Exploring the role of large language model in collaborative travel planning task

Introduction. This study explores how generative AI, specifically ChatGPT, influences collaborative travel planning. Understanding its effect on tasks like trip planning reveals insights into human-AI collaboration, particularly how AI tools support decision-making and streamline information gathering. Method. Twenty participants (10 pairs) planned a 1-night, 2-day trip under two conditions: (1) using only Google, and (2) using Google with ChatGPT. This within-subject study measured completion time, satisfaction, and plan quality via questionnaires and observation to capture task performance, user behaviour, and collaboration dynamics. Analysis. Data was analysed using t-tests and Wilcoxon signed-rank tests to compare completion times, satisfaction, and plan quality. Analyses of conversation volume and ChatGPT logs provided insights into AI-assisted collaboration dynamics and interaction patterns. Results. No significant difference in task completion time was found. However, plans made with ChatGPT were more complete and aligned with requirements. Participants found information more easily with ChatGPT, but satisfaction levels remained similar, suggesting that easier information access did not translate to higher satisfaction. Conclusions. Generative AI improves collaborative search task quality but does not enhance efficiency or satisfaction. AI tools like ChatGPT are effective for providing structured information and are best used as complementary resources alongside traditional search engines in planning tasks.

Weighted twin support vector machine with rescaled hinge loss

Abstract Weighted twin support vector machine (WTSVM) has been proved to be effective for classification problems. However, it is sensitive to noises, especially for data corrupted by outliers. In this paper, we propose an improved classifier termed as weighted twin support vector machine with rescaled hinge loss (RHWTSVM). Similar to WTSVM, it uses the intra-class KNN technique to extract structural information in the same class. It uses the inter-class KNN technique to reduce the redundant constraints to improve the computational speed. Furthermore, we introduce the regularization term into the objective function to make the proposed RHWTSVM implement the principles of structural risk minimization and empirical risk minimization simultaneously. Besides, we use the rescaled hinge loss function which is a monotonic, bounded, and nonconvex loss to replace the traditional hinge loss function in WTSVM to make the proposed classifier more robust. Therefore, the RHWTSVM is less sensitive to outliers. Because the model is a nonconvex optimization problem, we use the half-quadratic optimization method to solve it and find that the new method is equivalent to an iterative WTSVM. Numerical experiments on datasets with various levels of noise demonstrate that RHWTSVM is reasonable and effective.

GraphSleepFormer: a multi-modal graph neural network for sleep staging in OSA patients

Obstructive sleep apnea (OSA) is a prevalent sleep disorder. Accurate sleep staging is one of the prerequisites in the study of sleep-related disorders and the evaluation of sleep quality. We introduce a novel GraphSleepFormer (GSF) network designed to effectively capture global dependencies and node characteristics in graph-structured data. The network incorporates centrality coding and spatial coding into its architecture. It employs adaptive learning of adjacency matrices for spatial encoding between channels located on the head, thereby encoding graph structure information to enhance the model's representation and understanding of spatial relationships. Centrality encoding integrates the degree matrix into node features, assigning varying degrees of attention to different channels. Ablation experiments demonstrate the effectiveness of these encoding methods. The Shapley Additive Explanations (SHAP) method was employed to evaluate the contribution of each channel in sleep staging, highlighting the necessity of using multimodal data. We trained our model on overnight polysomnography data collected from 28 OSA patients in a clinical setting and achieved an overall Accuracy of 80.10%. GSF achieved performance comparable to state-of-the-art methods on two subsets of the ISRUC database. The GSF Accurately identifies sleep periods, providing a critical basis for diagnosing and treating OSA, thereby contributing to advancements in sleep medicine.

Discovery of mono-ADP ribosylating toxins with high structural homology to Pseudomonas exotoxin A

Mono-ADP-ribosyl transferase (mART) proteins are secreted virulence factors produced by several human pathogens, the founding member being diphtheria toxin (DT). Pseudomonas aeruginosa can also secrete a mART toxin, known as exotoxin A (PE), but with an organization of its three functional domains (receptor, translocation, and enzymatic elements) that is opposite to DT. Two additional PE-like toxins (PLTs) have been identified from Vibrio cholerae and Aeromonas hydrophila, suggesting more PLT family members may exist. Database mining discovered six additional putative homologues, considerably extending this group of PLTs across a wide range of bacterial species. Here, we examine sequence and structural information for these new family members with respect to previously identified PLTs. The X-ray crystal structures of four new homologues show the conservation of critical features responsible for structure and function. This study shows the potential of these newly described toxins for the development of novel drug delivery platforms. Additionally, genomic analysis suggests horizontal gene transfer to account for the wide distribution of PLTs across a range of eubacteria species, highlighting the need to monitor emerging pathogens and their virulence factors.

Utility of UAS-LIDAR for estimating forest structural attributes of the Miombo woodlands in Zambia.

The ability to collect precise three-dimensional (3D) forest structural information at a fraction of the cost of airborne light detection and ranging (lidar) makes uncrewed aerial systems-lidar (UAS-lidar) a remote sensing tool with high potential for estimating forest structural attributes for enhanced forest management. The estimation of forest structural data in area-based forest inventories relies on the relationship between field-based estimates of forest structural attributes (FSA) and lidar-derived metrics at plot level, which can be modeled using either parametric or non-parametric regression techniques. In this study, the performance of UAS-lidar metrics was assessed and applied to estimate four FSA (above ground biomass (AGB), basal area (BA), diameter at breast height (DBH), and volume (Vol)) using multiple linear regression (MLR), a parametric technique, at two wet Miombo woodland sites in the Copperbelt province of Zambia. FSA were estimated using site-specific MLR models at the Mwekera and Miengwe sites and compared with FSA estimates from generic MLR models that employed combined data from the two sites. The results revealed that the model fit of site-specific MLR models was marginally better (Adj-R2: AGB = 0.87-0.93; BA = 0.88-0.89; DBH = 0.86-0.96; and Vol = 0.87-0.98 than when using a generic combined data model (AGB = 0.80; BA = 0.81; DBH = 0.85; and Vol = 0.85). However, the rRMSE (2.01 - 20.89%) and rBias (0.01-1.03%) of site specific MLR models and combined data model rRMSE (3.40-16.71%) and rBias (0.55-1.16%) were within the same range, suggesting agreement between the site specific and combined data models. Furthermore, we assessed the applicability of a site-specific model to a different site without using local training data. The results obtained were inferior to both site-specific and combined data models (rRMSE: AGB = 36.29%-37.25%; BA = 52.98-54.52%; DBH = 55.57%-64.59%; and Vol = 26.10%-30.17%). The results obtained from this indicate potential for application in estimating FSA using UAS-lidar data in the Miombo woodlands and are a stepping stone towards sustainable local forest management and attaining international carbon reporting requirements. Further research into the performance of UAS-lidar data in the estimation of FSA under different Miombo vegetation characteristics, such as different age groups, hilly terrain, and dry Miombo, is recommended.

Multi-order asymptotic expansion of blow-up solutions for autonomous ODEs: II. Dynamical correspondence

Abstract In this paper, we provide a natural correspondence of eigenstructures of Jacobian matrices associated with equilibria for appropriately transformed two systems describing finite-time blow-ups for ODEs with quasi-homogeneity in an asymptotic sense. As a corollary, we see that asymptotic expansions of blow-ups proposed in Part I themselves provide a criterion of the existence of blow-ups with an intrinsic gap structure of stability information among two systems. Examples provided in Part I are revisited to show the above correspondence.

A Fusion Deep Learning Model for Predicting Adverse Drug Reactions Based on Multiple Drug Characteristics

Artificial intelligence (AI)-assisted prediction of adverse drug reactions (ADRs) has significant potential for improving drug safety and reducing financial costs. Early studies often relied on limited dimensions such as the molecular structure of drugs or interactions with biomolecules. In contrast, integrating these characteristics provides valuable insights into ADR predictions from multiple perspectives, enhancing the comprehensiveness and accuracy of the prediction models. In addition, previous studies have focused on whether a specific adverse drug reaction occurs with a particular drug, ignoring the fact that multiple adverse drug reactions may occur concurrently with a single drug. To address these, we developed a predictor that identifies ADRs early in drug discovery, using a deep learning model designed to fuse multiple drug characteristics. Our approach employed four modules to extract one- and two-dimensional sequence structure information of drug molecules, drug–protein interaction data, and drug similarity. A fusion model integrated these characteristics to predict the precise probability of ADRs. The receiver operating characteristic–area under curve (ROC-AUC), area under precision–recall curve (AUPR), and F1 scores on the benchmark dataset are 0.7002, 0.6619, and 0.6330, respectively. The AUPR is significantly improved compared to the conventional multi-label classifier (from 64.02% to 66.19%). In addition, we compared the results with the state-of-the-art methods on LIU’s dataset and the AUPR increased from 34.65% to 68.82%, which shows that our model outperforms them in terms of accuracy and robustness. Ablation experiments further validated the effectiveness of the individual modules. This model accurately predicted the probability of various ADR classes by integrating comprehensive information, thereby offering significant value in enhancing monitoring measures for new drug development and clinical use.

Classification and Carbon-Stock Estimation of Mangroves in Dongzhaigang Based on Multi-Source Remote Sensing Data Using Google Earth Engine

Changes in carbon stock are a key indicator for assessing the carbon-pool function and the impact of regional carbon cycling on climate. Mangroves, as an essential component of coastal ecosystems, play a critical role in carbon sequestration. However, traditional carbon-sink assessments often overlook biophysical differences between mangrove species and depend on field measurements, which are constrained by the challenging geographical conditions of the intertidal zone. This leads to biases and operational difficulties in estimating carbon stock. To address these challenges, this study proposes a fine-scale method for inter-species carbon-stock assessment, integrating mangrove three-dimensional structural information and spectral characteristics through Google Earth Engine (GEE). By combining GEDI data and Sentinel-2 imagery, this approach incorporates both vertical structure and spectral characteristics, overcoming the limitations of traditional models that neglect inter-species differences and vertical structural information. As a result, the accuracy of carbon-stock estimation is significantly improved. Unlike previous studies, this paper achieves a fully remote sensing-based partial carbon-stock assessment for mangrove species and quantifies carbon stock using the InVEST model, addressing the limitations of previous carbon-sink models. Specifically, on the GEE platform, Sentinel-2 imagery is used for inter-species classification through a random forest (RF) model, while a relationship model between canopy height and biomass is established using GEDI data to estimate biomass. To optimize feature selection, this study introduces a forward feature selection (FFS) approach, which incrementally selects the most predictive features, enhancing the stability and accuracy of the model. By combining biomass and classification results, a remote sensing-based carbon-sink assessment for mangroves is achieved. The study quantifies and visualizes the carbon stock of different mangrove species in Dongzhaigang, revealing that the region’s annual carbon stock totals 302,558.77 t. This validates the superiority and accuracy of the proposed method.

Physical Basics of Scanning Electron Microscopy in Volume Electron Microscopy.

Volume electron microscopy (vEM) has become a widely adopted technique for acquiring three-dimensional structural information of biological specimens. In addition to the traditional use of transmission electron microscopy (TEM), recent advances in the resolution of scanning electron microscopy (SEM) made it suitable for vEM application. Currently, various types of SEM with different advantages have been utilized. For selecting the appropriate type of SEM to obtain optimal vEM images for the purpose of individual research, it is important to understand the physics underlying each SEM technology. This article aims to explain the physics for signal electron generation, various objective lens configurations, and detection systems, employed in SEM to enhance high-resolution imaging and improve signal detection conditions.

MMFmiRLocEL: A multi-model fusion and ensemble learning approach for identifying miRNA subcellular localization using RNA structure language model.

MiRNA subcellular localizations (MSLs) are essential for uncovering and understanding miRNA functions in various biological processes. Several computational methods have been proposed for measuring MSL. However, existing methods only rely on manually crafted features based on sequence without considering RNA 3D structure information, and most methods often rely on single-model approaches, which fail to capture the full complexity of biological systems, further hindering predictive accuracy and performance. In this study, we introduce a deep learning-based approach, MMFmiRLocEL, which integrates multi-model fusion and ensemble learning for MSL identification. To the best of our knowledge, MMFmiRLocEL is the first method to combine sequence, structure, and function three information for MSL prediction. Specifically, it employs RNA 3D structure generated by the predicted structural model to construct a structure-based approach for MSL prediction. It also develops a sequence-based prediction method using sequence features and convolutional neural networks, while constructing a function-based prediction method using miRNA-disease association networks and deep residual neural networks. Furthermore, a multi-model fusion approach, employing weighted ensemble strategies, integrates sequence, structure, and function models to enhance the robustness and accuracy of MSL identification. Experimental results demonstrate that MMFmiRLocEL outperforms existing state-of-the-art methods, and then ablation analysis confirmed the significant contribution of the multi-model fusion mechanism to improve the prediction performance.

Virtual and Real Occlusion Processing Method of Monocular Visual Assembly Scene Based on ORB-SLAM3

Addressing the challenge of acquiring depth information in aero-engine assembly scenes using monocular vision, which complicates mixed reality (MR) virtual and real occlusion processing, we propose an ORB-SLAM3-based monocular vision assembly scene virtual and real occlusion processing method. The method proposes optimizing ORB-SLAM3 for matching and depth point reconstruction using the MNSTF algorithm. MNSTF can solve the problems of feature point extraction and matching in weakly textured and texture-less scenes by expressing the structure and texture information of the local images. It is then proposed to densify the sparse depth map using the double-three interpolation method, and the complete depth map of the real scene is created by combining the 3D model depth information in the process model. Finally, by comparing the depth values of each pixel point in the real and virtual scene depth maps, the virtual occlusion relationship of the assembly scene is correctly displayed. Experimental validation was performed with an aero-engine piping connector assembly scenario and by comparing it with Holynski’s and Kinect’s methods. The results showed that in terms of virtual and real occlusion accuracy, the average improvement was 2.2 and 3.4 pixel points, respectively. In terms of real-time performance, the real-time frame rate of this paper’s method can reach 42.4 FPS, an improvement of 77.4% and 87.6%, respectively. This shows that the method in this paper has good performance in terms of the accuracy and timeliness of virtual and real occlusion. This study further demonstrates that the proposed method can effectively address the challenges of virtual and real occlusion processing in monocular vision within the context of mixed reality-assisted assembly processes.

Dynamic multimodal information encryption combining programmable structural coloration and switchable circularly polarized luminescence.

Multimodal optical-materials are highly desirable due to their advantages in enhancing information security, though independent modulation is challenging, especially accurately controlling the orthogonal relationship between the structural coloration (SC) and fluorescence (FL) pattern. Herein, we report a strategy which combines programmable structural coloration and switchable circularly polarized luminescence (CPL) for multimodal information encryption. Using photomask with aligned grating, programmable periodic patterns are fabricated on a polydiacetylene (PDA) gel film, which produces image in tunable structural colors. Introducing a chiral fluorescence layer containing perovskite nanocrystals and twisted-stacking Ag nanowires (NWs) bilayers, which provides stimuli-responsive FL and CPL with high dissymmetry factor (glum, up to 1.3). Importantly, the structural coloration information and FL patterns (including CPL pattern) can be independently modulated without mutual interference, even selectively concealed or exposed by varying microstructure design of the cross-linked PDA gel or by acetonitrile treatment.

Nickel-Dithiolene Cofactors as Electron Donors and Acceptors in Protein Hosts.

Metal dithiolene compounds are attracting considerable attention in the field of molecular electronics, particularly as constituents of materials with high charge-carrier mobilities. Recent experiments on cable bacteria that perform centimeter-scale charge transport suggest that Ni-bis(dithiolene) cofactors are important components of the bacterial conductive network. Further, current-voltage experiments of cable-bacteria-conductive sheaths have measured high conductivity values as compared to other electron-transfer bacteria. An important question is how the Ni-bis(dithiolene) structures participating as electron donors/acceptors contribute to the high conductivity. Currently, the protein and cofactor structures of these bacterial networks are largely unknown. Given this limitation, in this work, we explore the more general question of how Ni-bis(dithiolene) molecules would perform as electron donor and acceptor centers in protein-mediated charge transfer. Our aim is to deduce order-of-magnitude higher bounds for charge-transfer rates in such systems as a function of donor-acceptor distance, protein-bridge (amino acid) sequence, cofactor size, and redox state. These bounds are useful for predicting charge-transfer mechanisms and estimating rates in the absence of detailed structural information on protein wires that may use Ni-bis(dithiolene) redox cofactors. Our analysis is also relevant to the design of artificial Ni-bis(dithiolene) protein wires.

Optimal control-based nuclear spin cross-polarization in the presence of complicating anisotropic interactions.

Cross-polarization is an indispensable part of solid state nuclear magnetic resonance spectroscopy to enhance sensitivity and extract structural information. However, the presence of certain anisotropic interactions, including chemical shift anisotropy and quadrupolar coupling, makes the inter-nuclear spin correlation experiments challenging. This impedes characterization of numerous materials and pharmaceutical compounds containing isotopes, such as 19F with large chemical shift anisotropy and 6/7Li, 23Na, 27Al, etc., with quadrupolar coupling. To address this problem, we introduce a new optimal control simulation-generated pulse sequence for Optimal Polarization Transfer In the presence of Anisotropic Nuclear Spin interactions (OPTIANS). Numerical simulations show high efficiency and robustness against experimental imperfections under a broad range of anisotropic interaction strengths for 19F-7Li, 19F-23Na, 19F-27Al, and 19F-13C polarization transfers. The polarization transfer curves show transient oscillations, which make the pulse sequence a quantitative method for dipolar coupling measurements. Experiments on a multi-metal fluoride system validate the predictions of the simulations by showing efficient PT in three spin pairs at varying experimental conditions. Remarkably, this method shows 50% better 19F-7Li PT efficiency at 14.1 T compared to the ramped cross-polarization experiment. The underlying polarization transfer mechanism is analyzed using the Fourier transform of the polarization transfer curves revealing that this optimal control method utilizes the chemical shift anisotropy and quadrupolar coupling to facilitate robust and efficient cross-polarization.

Protein Aging Dynamics: a Perspective from Non-equilibrium Coarse-grained Models

Abstract The aging of biomolecular condensates has been implicated in the pathogenesis of various neurodegenerative diseases, characterized by a transition from a physiologically liquid-like state to a pathologically ordered structure. However, the mechanisms governing the formation of these pathological aggregates remain poorly understood. To address this, the present study utilizes coarse-grained molecular dynamics simulations based on Langevin dynamics to explore the structural, dynamical, and material property changes of protein condensates during the aging process. Here, we further developed a non-equilibrium simulation algorithm that not only captures the characteristics of time-dependent amount of aging beads but also reflects the structural information of chain-like connections between aging beads. Our findings reveal that aging induces compaction of the condensates, accompanied by a decrease in diffusion rates and an increase in viscosity. Further analysis suggests that the heterogeneous diffusivity within the condensates may drive the aging process to initiate preferentially at the condensate surface. Our simulation results align with the experimental phenomena and provide a clear physical picture of the aging dynamics.

Streamline the information phase of the value engineering workshop through a web-based platform integrating BIM and GIS

PurposeThe traditional information phase of value engineering (VE) workshops often involves face-to-face meetings and physical site visits, which can be time-consuming, costly and impractical due to various constraints. This paper introduces a web-based platform integrating BIM and GIS to streamline this phase by systematically gathering, processing and visualizing multi-format information in a virtual environment.Design/methodology/approachA comprehensive literature review was conducted to analyze the potential of BIM and GIS integration in the VE context. A three-step information structure generation method was employed to outline the information gathering needs during the VE workshops, incorporating a review of multi-source references. A framework including information gathering, processing and visualization was proposed to streamline the information phase with BIM and GIS integration as the core. Subsequently, a web-based prototype platform was developed utilizing the three-layer architecture, with the case study to demonstrate its application. A comprehensive evaluation involving black box testing, user interface testing and focus group discussion was conducted to validate the functionality, usability, potential and limitations.FindingsThe platform can support the systematic collection, processing, organization and visualization of multi-format information during the information phase, enable virtual meetings and virtual site visits and facilitate VE participants’ review, understanding and analysis of such information and the documentation of their comments and suggestions. The black box testing validated its functionality, with necessity, effectiveness and convenience confirmed by VE practitioners. Its potential was approved due to clear information structure, web-based virtual review and logical multi-format information organization.Originality/valueThe generalized four-category information structure and proposed three-layer framework streamline the information phase of VE practices. The developed platform allows the VE team to review multi-format information virtually, simultaneously, freely and as many times as necessary.

Aspect-Based Sentiment Analysis Through Graph Convolutional Networks and Joint Task Learning

Aspect-based sentiment analysis (ABSA) through joint task learning aims to simultaneously identify aspect terms and predict their sentiment polarities. However, existing methods face two major challenges: (1) Most existing studies focus on the sentiment polarity classification task, ignoring the critical role of aspect term extraction, leading to insufficient performance in capturing aspect-related information; (2) existing methods typically model the two tasks independently, failing to effectively share underlying features and semantic information, which weakens the synergy between the tasks and limits the overall performance of the model. In order to resolve these issues, this research suggests a unified framework model through joint task learning, named MTL-GCN, to simultaneously perform aspect term extraction and sentiment polarity classification. The proposed model utilizes dependency trees combined with self-attention mechanisms to generate new weight matrices, emphasizing the locational information of aspect terms, and optimizes the graph convolutional network (GCN) to extract aspect terms more efficiently. Furthermore, the model employs the multi-head attention (MHA) mechanism to process input data and uses its output as the input to the GCN. Next, GCN models the graph structure of the input data, capturing the relationships between nodes and global structural information, fully integrating global contextual semantic information, and generating deep-level contextual feature representations. Finally, the extracted aspect-related features are fused with global features and applied to the sentiment classification task. The proposed unified framework achieves state-of-the-art performance, as evidenced by experimental results on four benchmark datasets. MTL-GCN outperforms baseline models in terms of F1ATE, accuracy, and F1SC metrics, as demonstrated by experimental results on four benchmark datasets. Additionally, comparative and ablation studies further validate the rationale and effectiveness of the model design.