Ontology Framework Turns Concrete Pouring Radio Chatter into Safety Records
A proof-of-concept ontology framework classifies 25 simulated concrete pouring radio utterances into semantic tags, hazard signals and situation states, with ...
Tag: Natural Language Processing
AI Agents That Debate Grades: New System Scores Essays Trait by Trait Without Training
A new multi-agent AI system called MATE uses specialized debating agents and a coordinating agent to score essays across multiple ...
BERT-EmoShiftNet Fusion Enhances Stress Dialogue Recognition
A new BERT-EmoShiftNet fusion framework achieves 85.0% accuracy in recognizing psychological stress by integrating semantic, speech, and visual data in ...
AI System Turns Hotel Guest Reviews into Automated Management Decisions
A new study combines whole-word-masked BERT, a bidirectional LSTM and deep reinforcement learning to turn Chinese hotel guest reviews directly ...
LLM Domain Adaptation Optimizes Radiology Impression Generation in Low-Data Regimes
Researchers evaluated domain adaptation and fine-tuning of five large language models for radiology impression generation across 3.1 million reports, finding ...
Ensemble AI model flags and rewrites media bias in news text
A new ensemble AI system detects media bias in news text with 77.5 percent accuracy and rewrites flagged passages while ...
Verb-Driven Task Networks Turn Static Epidemic Plans into Executable Playbooks
Researchers in China have built an HTN-Verb framework that converts static public health emergency plans into executable task networks, with ...
Most COPD hospital records lack captured pulmonary function reports, audit finds
A retrospective audit of an inpatient EHR export found pulmonary function report rows captured for only 10.8% of patients with ...
Research report: A collaborative multi-LLM framework with self-reflection iterative prompting for automated essay scoring
A new multi-LLM framework that lets three AI models iteratively refine scoring rubrics before grading reached an average Quadratic Weighted ...
Hybrid model detects suicidal ideation in Spanish social media posts
A hybrid BERT, BiLSTM and BiTCN model classified suicidal ideation in Spanish social media posts with F1-scores of 93.51% and ...
ViolenceBanglaBERT improves Bangla violence-text classification in controlled benchmark
ViolenceBanglaBERT combined contextual Bangla representations, a lightweight classifier and back-translation augmentation to reach a 0.75 macro F1-score in a controlled ...
AI Learns to Summarize Indonesian Friday Sermons Without Human Labels
Researchers built the first extractive summarization system for Indonesian Friday sermon transcripts by fine-tuning IndoBERT on automatically generated pseudo-labels, achieving ...
BERT-guided nested Dirichlet forests sharpen hierarchical topic modeling
A new hierarchical topic model, DF_sLNDA, replaces LDA's flat Dirichlet prior with nested Dirichlet distributions guided by BERT-derived semantic constraints, ...
Machine learning with patient words sharpens emergency triage classification
A retrospective cohort study of 64,382 adult emergency department patients in Brazil found that adding natural language processing of chief ...
Adaptive Temperature Distillation Shrinks Multilingual AI Models for Classroom Use
A new distillation framework that adjusts its temperature per language and sample compresses a 560-million-parameter multilingual model into a 66-million-parameter ...
AI Moves Into Vascular Surgery, But Most Models Still Await Real-World Proof
A new review of thirty studies finds artificial intelligence showing strong potential across minimally invasive vascular surgery, from risk prediction ...
Wolf-pack algorithm with multiple leaders flags deceptive statements in executive background checks
A hybrid wolf-pack optimization framework combining multi-leadership search with BERT-based and statistical text features achieved an F1 score of 0.8733 ...
Hybrid Optimizer Lifts AraBERT Accuracy in Arabic Sentiment Analysis
A hybrid metaheuristic algorithm combining the Aquila Optimizer, the Sine Cosine Hyperbolic Optimizer, and opposition-based learning tuned eight hyperparameters of ...
Automated de-identification preserves relationships in child protection reports
An evaluation of 10,000 South Australian child protection reports found that automated relationship-preserving de-identification achieved high performance across most tested ...
BiGRU-CRF system reaches 96.52% accuracy in power-grid text
A 2026 computational study found that a domain-specific BiGRU-CRF pipeline achieved 96.52% entity-level exact-match accuracy across 6,000 Chinese power-grid text ...
POPULAR NEWS
-
CRISPR Beta-Cell Regeneration Shows Early Promise for Long-Term Diabetes Management
-
Sudan Study Reveals Gaps in Kidney Stone Prevention Care
-
Machine Learning Model Predicts Pediatric Sepsis Risk Using Routine Clinical Indicators
-
ChatMPC Integrates Natural Language Processing with Predictive Control for Autonomous Driving