Large language models (LLMs) make it possible to build knowledge graphs from professional literature at scale, but how such graphs fail at that scale is rarely measured. We built \sys{}, a cross-media knowledge graph of swine, aquaculture, tomato, maize and rice production, and analysed its 199,418 entities and 277,053 relations together with the web system that serves it. An LLM extracted typed triples from the professional literature of each sector; 74.9% of the relations carry an evidence record of document, page and excerpt, and a multimodal embedding model links 25,158 entities to images and 14,132 to video clips. We identified five failure modes of construction. First, the sector schem... More >

Fine-grained plant disease classification requires separating classes of the same crop that often differ only in small, sparse lesion patterns. Looped Vision Transformers (LoopViTs) obtain depth by applying a small stack of weight-tied blocks repeatedly, yet each block rebuilds its attention graph from scratch at every iteration, although the same block has just computed a graph over the same tokens. We propose LoGA, a looped vision transformer with Loop-Carried Graph Attention, in which the attention graph of each block at iteration $t$ is biased by the graph the same block produced at iteration $t-1$, used as a tempered prior. In its log-prob form the update $A_t \propto \exp(QK^\top/\sqrt... More >

To solve the problems faced by Chinese character recognition, such as complex shapes and diverse structures, this paper adopts the Visual Geometry Group 16 (VGG-16) model for feature extraction and introduces a two-layer bidirectional Long Short-Term Memory (LSTM) network. It improves the Faster Region-based Convolutional Neural Network (Faster R-CNN) by using a Region Proposal Network (RPN) to extract candidate boxes and adjust the positions of candidate regions. The improved model, namely Faster BLSTM-CNN, is tested through three types of experiments: validation of feature extraction effectiveness, comparative analysis of the algorithm before and after improvement, and comparison with trad... More >

Analyzing heterogeneous online reviews from multiple stakeholder groups represents an emerging challenge in AI-driven service intelligence. This study proposes a two-stage sentiment correlation detection framework and applies it to customer and employee reviews of the Istanbul Marriott \c{S}i\c{s}li hotel. During dataset construction, a semi-supervised domain-specific blacklisting approach was developed alongside standard preprocessing steps to improve sentiment signal quality. In the first phase, customer and employee reviews were treated as separate datasets, and 5-fold cross-validation was applied using TF-IDF, BOW, and Word2Vec representations with multiple classifiers, achieving 99.8% a... More >

Molecular property prediction is a fundamental task in drug discovery and materials science, yet most high-performing approaches depend on large-scale pretraining that demands substantial computational resources. This work proposes a pretraining-free ensemble framework that trains multiple Transformer-based architectures—BERT, RoBERTa, and XLNet—from random initialization using the Atom-in-SMILES (AIS) molecular representation, which provides richer atomic-level semantics than conventional SMILES. The three Transformer encoders are coupled with BiLSTM prediction heads and integrated via a BaggingRegressor to reduce variance and improve generalization. Experiments on the ZINC250k and ZINC... More >

This perspective introduces the range of ethical concerns entailed by the widespread adoption of AI, particularly as they impact human health. It begins by (1) illustrating risks associated with all large-scale AI systems, then moves to (2) corporate and governmental applications of AI that affect human health. It overviews the ways (3) that patient usage of AI has affected human health; (4) that “passive” medical AI (like recording documents) and (5) “active” medical AI (like diagnosing and prescribing) may affect human health. It concludes with (6) reflections on reporting, responsibility, and regulation, wherein international cooperation and governance systems appear essential for... More >

Traditional methods for classifying plant diseases usually depend on manual observation, which is time-consuming, labor-intensive, and prone to human error. The rise of deep learning has greatly advanced this field by enabling more accurate and efficient classification techniques. In this paper, we introduce a novel lightweight deep learning framework that builds on the RegNetY convolutional neural network architecture by incorporating a modified Efficient Channel Attention module. This enhancement is specifically designed to improve the classification of various rice leaf diseases. Our experiments on a publicly available dataset show that the proposed approach not only boosts classification... More >

Breast ultrasound imaging plays a crucial role in early breast cancer screening and diagnosis due to its noninvasive nature and cost-effectiveness. However, accurate lesion segmentation remains challenging because of severe speckle noise, low contrast, and blurred tumor boundaries. To address these issues, this paper proposes SEFF-Net, a novel edge-aware feature fusion network with a U-shaped encoder–decoder architecture to capture multi-level semantic representations for breast ultrasound image segmentation task. To enhance boundary perception, a Self-learning Edge Enhancement Module is embedded in the shallow encoding stages, while a Spatial Feature Fusion Module is introduced to effecti... More >