emotionX-2018

This is source code for our paper EmotionX-DLC: Self-Attentive BiLSTM for Detecting Sequential Emotions in Dialogues

Note: We accidently used dropout duirng testing. Setting dropout rate 0 could improve performance that is reported in our original paper.

Dataset

Please go to the share task webpage: http://doraemon.iis.sinica.edu.tw/emotionlines/challenge.html to download or request data needed for training and testing. You need to create a directory EmotionPush to store EmotionPush data (which you need to make a request to the organizer). The saved files should be renamed as emotionpush_{train, dev, test}. For testset either of Friends or EmotionPush, you need to make a formal request to the organizer.

Glove

Please download pre-trained Glove data with 300d size (glove.840B.300d.txt) or 50d size (glove.twitter.27B.50d).

Model training

python(3) train.py or ./train.sh

Model testing

python(3) test.py