〰️ Word embedding
= the representation of a 🈁 Word as a vector, which encodes its meaning, such as that similar words are close to each other 1
Types
Evaluation
Extrinsic
- e.g. compare different embedding models
Intrinsic
Computing similarity
- ⚫️ Dot-product
- Normalized (by 5️⃣ Relative frequency):
- 📐 Cosine similarity
- result: [0, 1], because frequency >= 0
Sentiment analysis
→ investigate the evolution of word sentiments
Cultural biases
→ study cultural biases (e.g. gender, ethnicity…)