# 5/12 報告
## [PPT](https://docs.google.com/presentation/d/1C_2ZTH-PspbBcG3FpEzHab_VcRkMYB37/edit?usp=sharing&ouid=113619581856615547902&rtpof=true&sd=true)

## 項目
1. Show your data again. (New version or the same.)
2. What have you done? (Show your method or some results.)
3. What will you do next? (Including the Gantt Chart would be better.)
4. If Professor Tsai gave you the suggestion, what's the suggestion? and what changes did you make?

## Review
### 專題是要做什麼？
- 利用圖論上的方法，引用外部知識來讓立場分類的性能有所提升。
### 要達成什麼目標？
- 了解目前 Graph 的方法能否提升或改變立場檢測任務的準確率。

## Part 1: Data
1. [SemEval-2016 Task 6](https://alt.qcri.org/semeval2016/task6/)
- 提供測試資料
- 資料集下載 [Link](http://alt.qcri.org/semeval2016/task6/data/uploads/stancedataset.zip)
2. [ConceptNet](https://conceptnet.io/)
- python [API](https://pypi.org/project/ConceptNet/)
- 建圖(*)
- 提供外部知識

## Part 2: Done
## 已達成
- 處理資料集
### Baseline
- word2vec 250 維 (DL)
| 項目 | Test Acc. | Rank |
| -------------------------- | ------ | ---- |
| Deep Neural Network | 67.10% | 3 |
| Convolution Neural Network | 67.17% | 4 |
| Support Vector Machine | 66.32% | 2 |
| Random Forest Classifier | 63.6% | 1 |

## Part 3: Next
### 預計
- 實作簡易 baseline model : BERT
- 資料集資料處理
- ConceptNet 學習
- 資料集與 ConceptNet 的搭配
- 實作先前 paper 的 model

## Part 4: Suggestion