A Framework for Transfer Learning on Multi-lingual Text Data for Sentiment Classification
summary
Claims a practical method that merges GloVe/Stanford embeddings with custom Twitter embeddings to add Filipino vocabulary, trained on ~9 million scraped tweets (~6M unique). Uses off‑the‑shelf Vader labels on English and projects those sentiment signals into Tagalog by computing displacement vectors from shared reference words (a pragmatic, non‑rigorous rotation). Demonstrates emoji handling, some sarcasm detection, and a continually adaptive workflow that updates the embedding layer without retraining the full neural classifier.
AI-written summary · about
chapters (4)
- 0:00Start
- 5:36Advantages of using Word Embeddings
- 10:42Transfer Learning
- 23:18Steps for transfer and continuous learning by leveraging VADER-labeled data to build a model for Filipino phrases and sentences
related talks
22mMaking Natural Language Processing Robust to Sociolinguistic Variation
22mMultilingual embeddings to scale NLP models to multiple...
45mDeep Learning for Natural Language Processing
37mMultilingual sentiment analysis & emotion detection based on social media data
41mBetter Machine Learning with Less Data
31mAttacking Deep Learning-Based NLP Systems
41mSemantic Multimodal Multilingual Similarity Engine
43mDemocratizing NLP content modeling with transfer learning using GPUs