---
tags: meeting
---

# 12-23 meeting

## Model
- 目前是五層全linear

- **加入abstract與title**

1. LINE embedding:


loss相較之前沒有abstract與title上升且卡住

2. DeepWalk embedding:


**使用DeepWalk當作target**

## TODO
- [ ] 比較GraphSAGE的成效
- [ ] 修改model(deeper & Normalization)
- [ ] 看2018AAAI做testing的loss
- [ ] 計算MAP