We propose a multi-task learning architecture that jointly performs emotion and sentiment analysis across languages. The model leverages shared encoders and task-specific decoders, offering scalable performance and interpretability in noisy, short-text environments with cross-cultural linguistic diversity.

Laura Michelle Benson, Robert Anthony Knight, Benjamin Thomas Grayson, Emily Jane Waters, Sarah Victoria Douglas

This work introduces a hierarchical recurrent model for semantic segmentation of aerial urban data. It incorporates spatial reweighting layers that adjust feature importance based on context, improving segmentation granularity and stability in applications like mapping, planning, and infrastructure analysis.

Michael Daniel Roberts, Emily Katherine Harris, Olivia Ann Foster, Rachel Grace Townsend, Thomas Edward Young

We propose a contrastive pretraining framework for named entity recognition in legal texts. It aligns visual and textual embeddings across languages, enabling accurate NER performance in low-resource, specialized document formats without reliance on large annotated corpora.

Chen Yuan Zhi, Xu Tian Ming, Liu Bo Feng, Zhang Wei Jun, Gao Chang Hong

This paper introduces a meta-learning method for few-shot diagnosis prediction using variable patient histories. It enables generalization from minimal labeled samples, supporting robust clinical decision-making in sparse-record settings through context-adaptive learning and episodic training across heterogeneous input conditions.

Ravi Kiran Menon, Elena Lucia Rossi, Priya Venkatesh Nair, Carlos Miguel Ortega, Jean Jacques Blanc

This paper presents an end-to-end pipeline for building knowledge graphs from scientific literature. Combining distant supervision with logical rule learning, the system extracts structured insights from unstructured text, accelerating automated reasoning and discovery in academic and industrial research domains.

Chen Yong Jie, Liu Zhi Liang, Xu Hao Sheng, Zhang Min Rui, Gao Lin Fang, Huang Xiu Zheng