Modeling the impact of amino acid mutations on the binding affinity between antibodies and antigens plays a pivotal role in antibody therapeutics development. While numerous deep learning-based methods have shown promising results, accurately modeling the impact of mutations on binding interfaces remains a significant challenge, as even small changes can substantially affect binding affinity. To address this issue, we propose VQH-AM, a Vector Quantization-based Heterogeneous graph neural network for antibody Affinity Maturation prediction. Specifically, we construct atom-level heterogeneous graphs for both wild-type and mutant antibody-antigen complexes to model their binding interfaces. Atomic representations extracted by the heterogeneous graph neural networks are then quantized via a context-aware codebook designed to capture semantic atomic patterns. Finally, both the original and quantized atomic representations are integrated and used to predict the affinity maturation outcomes. Benchmark experiments demonstrate that VQH-AM consistently outperforms state-of-the-art methods across multiple evaluation metrics. The source code is available at: https://github.com/zwzhen-hnu/VQH-AM.

Ejection fraction (EF) is a key indicator of cardiac function, crucial for diagnosing heart failure and guiding treatment. Its estimation from echocardiography is challenged by morphological changes across cardiac phases and low-quality, noisy boundaries. We propose EFusionNet, a multimodal segmentation framework that integrates echocardiographic images with structured diagnostic text to enhance segmentation and EF assessment. Clinical phrases (e.g., “irregular boundary”) are embedded via a domain-specific language model into both input fusion and UNet skip connections, enabling phase-aware feature calibration. A feature fusion enhancement module (FFEM) refines spatial localization, while a multi-objective loss enforces uncertainty learning and semantic consistency. Evaluated on CAMUS and EchoNet-Dynamic datasets, EFusionNet achieves Dice scores of 91.2%/90.1% and EFMAEof 4.8/5.0, outperforming baselines and improving reliable, interpretable EF estimation.

Educational Data Mining (EDM) supports early detection of learning difficulties by predicting student performance. However, machine learning models often operate as black boxes. Explainable Artificial Intelligence (XAI) helps to explain why black-box models produce specific predictions. This paper systematically reviews the past five years of research on XAI applications for interpreting student performance prediction in Computer Science and STEM education. We found that behavioral and academic performance data were the most commonly used features, with the main prediction goals focused on course failure risk or grades. This study also examined the application areas of XAI, revealing that the most common uses were global feature importance analysis, individual prediction explanations, and supporting interventions and decision-making. Moreover, we found that SHapley Additive exPlanations (SHAP) were the most frequently utilized XAI technique, predominantly applied at the global level, with limited use at the individual level. Furthermore, a research gap was identified in utilizing XAI to support course improvements, customize visualizations, and generate personalized recommendations. Addressing this gap could enable educators to provide personalized, data-driven guidance to better support individual students.

Breast cancer poses a significant threat to women's health, and ultrasound plays a critical role in the assessment of breast lesions. This study introduces a prospective deep learning architecture, termed the "Multi-modal Multi-task Network" (3MT-Net), which integrates clinical data with B-mode and color Doppler ultrasound images. Specifically, an AM-CapsNet is employed to extract key features from ultrasound images, while a cascaded cross-attention mechanism is utilized to fuse clinical data. Moreover, an ensemble learning approach with an optimization algorithm is adopted to dynamically assign weights to different modalities, accommodating both high-dimensional and low-dimensional data. The 3MT-Net performs binary classification of benign versus malignant lesions and further classifies the pathological subtypes. Data were retrospectively collected from nine medical centers to ensure the broad applicability of the 3MT-Net. Two separate testsets were created and extensive experiments were conducted. Comparative analyses demonstrated that the AUC of the 3MT-Net outperforms the industry-standard computer-aided detection product, S-Detect, by 1.4% to 3.8%.

This Research Full Paper presents a systematic literature review on the impact of feedback features on students' subsequent learning strategies. Research indicates that providing quality feedback improves learning performance in higher education, namely in computing and engineering education. Framed by the self-regulated learning model, this enhancement results from the interplay of cognitive, metacognitive, motivational, and behavioral actions driven by feedback toward the learning goal. Such a combination of planned and selected actions is the learning strategy decision to direct the accomplishment of learning tasks. Insights of how feedback interventions lead to increased use of effective learning strategies have predominantly relied on qualitative data from self-reports. However, self-reports mainly reflect students' perceptions but cannot accurately capture the dynamic adjustments of learning strategies in the feedback process. With the growing use of learning management systems that can collect various learning analytics data, recent works have attempted to automatically generate personalized feedback based on mapping students' progress against pre-determined rules. Learning strategy alterations as a result of the various forms of feedback can be detected from the trace data. The findings of these studies provide evidence of how various feedback features are associated with the adjustment of learning strategies. This paper presents a systematic literature review that analyzes papers related to shifting learning strategies upon feedback provision to identify features that can trigger students' adoption of more effective learning strategies. The objective is to collect evidence to highlight feedback as more than information but a process to guide the proper use of learning strategies for better learning achievement. Our analysis shows a need for more studies to observe changes in learner actions due to feedback and discusses limitations in current works. With the rapid development of education data mining and deep learning models, the growing knowledge of feedback features can be potentially used with these computational models to generate learning advice to guide strategy changes for achieving better learning outcomes.

As the Artificial Intelligence (AI) industry rapidly expands in the Guangdong-Hong Kong-Macao Greater Bay Area (GBA) of China, the demand for AI talent continues to grow, higher educational institutions need to continuously update its AI education to adapt technological progress. However, there is currently a lack of understanding on AI education in the GBA. This work-in-progress research paper analyzes the current state of AI bachelor degrees within the GBA and compares it with the well-established curricula in the United States (US). The study selected a sample from the top 8 universities based on QS rankings and employed an enhanced AI course literacy categories model for the inductive and comparative analysis of AI course settings. The preliminary study highlights differences between universities in the GBA and the US in terms of course content depth and teaching strategies. The GBA universities emphasize a balanced and comprehensive curriculum that ensures all students have a solid foundation in both AI theories and technologies. In contrast, US universities prioritize a broader, interdisciplinary approach that fosters innovation and development of practical skills. The findings suggest that GBA universities could benefit from adopting some of the strategies used by US universities, such as adjusting the proportion of elective courses and increasing the availability of interdisciplinary courses. This would promote personalized student development and interdisciplinary learning, better preparing students to meet the challenges and opportunities of the AI era.

Molecular interaction prediction plays a crucial role in forecasting unknown interactions between molecules, such as drug-target interaction (DTI) and drug-drug interaction (DDI), which are essential in the field of drug discovery and therapeutics. Although previous prediction methods have yielded promising results by leveraging the rich semantics and topological structure of biomedical knowledge graphs (KGs), they have primarily focused on enhancing predictive performance without addressing the presence of inevitable noise and inconsistent semantics. This limitation has hindered the advancement of KG-based prediction methods. To address this limitation, we propose BioKDN (BiomedicalKnowledge GraphDenoisingNetwork) for robust molecular interaction prediction. BioKDN refines the reliable structure of local subgraphs by denoising noisy links in a learnable manner, providing a general module for extracting task-relevant interactions. To enhance the reliability of the refined structure, BioKDN maintains consistent and robust semantics by smoothing relations around the target interaction. By maximizing the mutual information between reliable structure and smoothed relations, BioKDN emphasizes informative semantics to enable precise predictions. Experimental results on real-world datasets show that BioKDN surpasses state-of-the-art models in DTI and DDI prediction tasks, confirming the effectiveness and robustness of BioKDN in denoising unreliable interactions within contaminated KGs.

Analyzing online consumer reviews is one of many popular applications of natural language processing in retail sectors. Text embedding models can transform the textual content of reviews into numerical representations for downstream analytic tasks and therefore they play an essential role in review analytics. As review analytics is increasingly used in the industries, more empirical research is needed to investigate how text embeddings perform in understanding the thoughts and attitudes of customers. In this study, we examined four commonly used text embeddings: namely TF-IDF, word2vec, sent2vec and BERT, to evaluate their performance in predicting the ratings and the sentiments of online consumer reviews. Drawn on the results, we highlight the strengths of these text embeddings and their desirable use cases. Our findings reveal that BERT and sent2vec can produce stable results in predicting the ratings of retail reviews in general. Besides, word2vec is more suitable for identifying negative sentiment within reviews. From a practical perspective, it is worth analyzing reviews from different product categories separately to achieve better results.

Treemaps have been used in information visualisation for over two decades. They make use of nested filled areas to represent information hierarchies such as file systems, library catalogues, etc. Recent years have witnessed the emergence of visualisations that resemble geographic maps. In this paper we present a study that compares the performance of one such map-like visualisation with the original two forms of the treemap, namely nested and non-nested treemaps. Our study focused on a quantitative evaluation of accuracy and speed. We found that accuracy was highest for the map-like visualisations, followed by nested treemaps and lastly non-nested treemaps. Task performance was fastest for nested treemaps, followed by non-nested treemaps, and then map-like visualisations. We conclude that the results regarding accuracy are promising for the use of map-like visualisations in tasks involving the visualisation of hierarchical information, even at the expense of somewhat longer performance times.

Maps have traditionally been used for representing the surface of the earth and for displaying geographical information. Apart from this obvious purpose, the metaphor of maps has been applied to other uses as well, such as information visualisation and novel user interfaces. Various methods exist for creating geographic map-like visualisations. Yet there is little understanding on how people perceive these graphical presentations as geographic maps, and how to make these information visualisations look like geographic maps. We attempt to find preliminary answers on these issues by conducting a user study with a series of map-like visualisations. In this paper, we report on the results of this study and reveal the factors that have an impact on the human perception of visualisations that are designed to resemble geographic maps. Based on this, we propose design suggestions for the future development of map-like visualisations.

Large data repositories such as electronic journal databases, document corpora and wikis often organise their content into categories. Librarians, researchers, and interested users who wish to know the content distribution among different categories face the challenge of analysing large amounts of data. Information visualization can assist the user by shifting the analysis task to the human visual sub-system. In this paper we describe three visualization methods we have implemented, which help users understand category hierarchies and content distribution within large document repositories, and present an evaluation of these visualizations, pointing out each of their relative strengths for communicating information about the underlying category structure.