Tag: Natural Language Processing

Researchers at Anhui University have developed KVCB, a key-value cache merging method that compresses large language model memory by fusing ...

Researchers have developed a few-shot AI framework that combines agentic retrieval-augmented generation with multi-agent data augmentation to detect coded dog-whistle ...

A new study shows that linking skills, occupations and qualifications in job advertisements requires construct-specific matching strategies, yielding significant accuracy ...

A new multimodal deep learning framework fuses Log-Mel spectrograms and BERT embeddings through self-attention to automatically score spoken English fluency ...

Researchers converted patent abstracts into semantic embeddings to measure technological similarity between firms and found the resulting network predicts subsequent ...

Researchers developed EntQA, a reference-free, entity-centric metric that scores how faithfully large language models preserve clinically relevant concepts in medical ...

Researchers show that an unmodified large language model can screen and extract data from digital health trial reports with human-level ...

University of Michigan researchers show that large language models like Claude and ChatGPT can mine millions of comments on citizen-science ...

A new hybrid deep learning model tuned by a snake-inspired optimization algorithm achieves 98.13 percent accuracy in classifying sentiment from ...

Researchers have built a hierarchical attention network that detects fake news in English, Hindi, and Marathi, reaching 98.9 percent accuracy ...

A new book by University of Illinois professor John Gallagher draws on interviews with more than 100 AI and machine ...

A new study in the Journal of Big Data shows that large language models can outperform traditional machine learning in ...

Researchers have unveiled byteification, a two-stage conversion method that retrofits existing subword-based large language models into byte-level models matching their ...

A new generative protocol called BridgeProt achieves top accuracy and perfect structural validity in emotion-cause pair extraction by verifying each ...

Researchers have developed ValuePEFT, a parameter-efficient fine-tuning method that steers large language models toward explicitly specified human values with measurable ...

A new multi-domain Arabic dataset and benchmark shows that deepfake-text detectors excel within familiar domains but falter sharply on government ...

A new 3.5 million euro Horizon Europe project coordinated by Johannes Gutenberg University Mainz will test twenty existing European counter-disinformation ...

Researchers in South Korea have developed RECUR, a retrieval-based framework using nonparametric masked language modeling that outperforms conventional methods in ...

A new study shows that lightweight transformer models can automatically classify the roles participants play in cyberbullying episodes, setting a ...

A new Riemannian framework called HyperSpectrum Geometry lets text classifiers bend the curvature of their semantic space, revealing that emotionally ...