Dongxu Zhang 2016-03-14

Last Week

---

• evaluate p@1. (origin 2000-800, 600-600; co-occur, low-freq)
• LSTM compare.
• pickup test samples that most distant from training data and test models.

This Week

---

• LSTM + similar pair.
• pickup words not in training data but in similar pair to sample test data.
• POS similar pair.