Spaces:
Running
Running
Commit History
-
Add LDA topic inference — show detected topic label + defining words in UI Running
Ryan Christian D. Deniega +1: Claude Sonnet 4.6LDAFeatureClassifier gains get_topic_info() which reuses the already-trained _lda and _count_vec to infer the dominant topic for any new text. Returns a human-assigned label (Health & Conspiracy, Breaking News, etc.), the top 6 words that define that topic, and the probability confidence. _run_comparison now returns (comparison_entries, LDATopicResult) tuple. VerificationResponse gains lda_topic field. The LDA column in the comparison panel expands to show the topic label in cyan, top-word chips, and match %. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>affe2db -
feat: extension button placement, text extraction, OCR display + ML improvements
Ryan Christian D. Deniega +1: Claude Sonnet 4.6Extension: - Fix verify button placement to appear below Like/Comment/Share bar on all page types - Fix walk-up logic to find correct post container (message container > article > innerText>100) - Fix [dir="auto"] filter to work when postElement is a sub-section of the article - Add obfuscated text detection to skip Facebook's character-by-character spans - Show full caption text (no truncation) and extracted image in verification modal - Add IMAGE TEXT (OCR) section to modal using backend ocr_text field - Fix "Extension context invalidated" error to show refresh prompt instead of retry - Add webNavigation SPA detection for Facebook pushState navigation API: - Add image_url field to TextVerifyRequest — runs Tesseract OCR alongside caption text - Add ocr_text field to VerificationResponse for image OCR results ML/NLP: - Add Naive Bayes, BoW, LDA, and ensemble classifiers - Add Tagalog RoBERTa fine-tuned model - Improve preprocessor, NER, claim extractor, and scoring engine - Add bench…c78c2c1
-
fix: cold start 502, favicon, verify state persistence
Ryan Christian D. Deniega- Dockerfile: pre-download HuggingFace models to eliminate cold start 502s - scoring/engine.py: cache NLP instances per-process (_nlp_cache) - Cloud Run: set min-instances=1 in deploy.sh - frontend: Lucide Radar icon favicon (matches navbar) - frontend/VerifyPage: sessionStorage persistence + Verify Again button - .gitignore: exclude ml/models/xlmr_model/ and *.safetensorsb1c84b5
-
feat: PhilVerify Phase 1-3 — FastAPI backend, NLP pipeline, TF-IDF classifier (23/23 tests)
Ryan Christian D. Deniega6c9b8f1