infini-news
Anushka BisneyFake News Detection System (Trending) You already mentioned this — great choice 👍 💡 Idea Detect whether news is fake or real using AI. ⚙️ Tech Stack Frontend: React Backend: Node.js / Django AI: NLP (BERT / Logistic Regression) API: News APIs 🚀 Features Paste news → get authenticity score Highlight misleading words Show source credibility Chrome Extension support 🌍 Real Impact Helps fight misinformation (huge problem today)
Comments (0)
Sign in to leave a comment
Model Manager
Manage AI models and update them for improved accuracy across all analysis pipelines.
Active Filters
Filter by Status
Filter by Model Type
Sort By
Select a model and click View to inspect its full details here.
Deployment Status
BERT Fake News Detector — Model v3.2.1
Target Environment
Version History
All released versions, accuracy changes, and deployment status
BERT-Large fine-tuned on expanded Indian regional news corpus; improved clickbait headline detection.
Release Notes
- Re-trained on 1.2M additional labeled articles from regional Indian news outlets.
- Clickbait classifier threshold lowered to 0.62, reducing false negatives by 18%.
- Inference latency reduced from 340ms to 210ms via quantization.
- Cross-lingual support added for Hindi and Tamil news text.
- Precision improved to 94.7%, Recall to 92.1%.
Hotfix for bias in political satire classification; source credibility scoring recalibrated.
Release Notes
- Fixed systematic misclassification of political satire articles as fake news.
- Source credibility weights updated using Jan 2026 IFCN database snapshot.
- Edge case handling for paywalled article snippets improved.
- Logistic Regression fallback model accuracy stabilized at 87.3%.
Introduced ensemble model combining BERT + Logistic Regression for dual-layer verification.
Release Notes
- Ensemble architecture combining BERT embeddings with Logistic Regression signal.
- Overall accuracy jumped from 89.4% to 92.5% on internal test set.
- Word cloud weighting algorithm v2 deployed with this release.
- API response schema updated; Chrome Extension v2.3+ required.
- Memory footprint increased by ~120MB; infrastructure scaled accordingly.
Patch: resolved token overflow errors on long-form investigative articles exceeding 512 tokens.
Release Notes
- Sliding window tokenizer introduced for articles > 512 BERT tokens.
- Minor accuracy regression on short news blurbs noted and logged.
- Rollback safeguard scripts added to deployment pipeline.
Major release: BERT-Base replaced with BERT-Large; full retraining on infini-news v2 dataset.
Release Notes
- Migrated from BERT-Base to BERT-Large for deeper contextual understanding.
- infini-news v2 dataset: 4.8M labeled articles, up from 1.9M in v1.
- Authenticity score gauge updated with smoother animation curve.
- Admin dashboard model cards redesigned to surface accuracy deltas.
- Model size: 340MB. Requires GPU inference for <250ms SLA compliance.
No comments yet. Be the first!