대화의 문맥을 분석해 감정 상태를 판단한다.
turn1, trun2, turn3 세 개의 대화를 분석하여 감정 상태를 판단한다.
https://competitions.codalab.org/competitions/19790
| turn1 | turn2 | turn3 | label |
|---|---|---|---|
| Hahah i loved it | Yay! Glad you loved it X3 😅 | You always make us happy | happy |
| Say something about you dear | You Know | Yes | others |
| See the thing is i love a girl but i am afraid to confess that to her | "go ahead. You feel good and are matured enough to manage relationships, ga." | But the thing is she doesnt even wants to talk with me :/ | sad |
| What is there for Lunch..?? | Coffee coffee and more coffee | Fuck off fuck off and more fuck off | angry |
| who are they | i'm in.! Whats up.? | nothing up only down | sad |
| My one main friend backstabbed me a few weeks ago | Awesome. What house did you get in? | Excuse I just told you a sad story and your gonna say awesome wtf is wrong with you | angry |
| Do u know cooking ? | Ofcourse Bro... I simply love cooking. ☺😊👍😍 | Nice | others |
- label은 turn3를 기준으로 매겨졌다.
- label은 총 4개로 happy, sad, angry, others로 나누어진다.
-
BI_LSTM
- CuDNN 사용한 Bi LSTM 레이어 2개 쌓았다.
- Glove Pre-trained Embedding 사용
-
TextCNN
- filter_size : 3, 4, 5
- filter_num : 128
- Glove Pre-trained Embedding 사용
-
ELMo
- tf-hub에서 제공하는 embedding matrix 사용
- embedding size가 1024 차원이다.