SemEval-2019-task3-EmoContext
The code of SemEval-2019 task3 EmoContext: Multi-Step Ensemble Neural Network for Sentiment Analysis in Textual Conversation
The training of Elmo word vectors used in EmoContext is in the Elmo_pre_train file, where elmo_result.py is the training code. There is no sentence training, but word-by-word training, which lacks contextual information about words.
Training of word vectors in EmoContext in char-embeddings.
Instructions are available in the Elmo Pre-training and Word Vector Training folders. Check it out yourself. And I have written the sample code.
ELMO pre-training and word vector training use other people's projects, please pay attention to the source when using.
The system architecture.
The result of system.
enviroment
- Win 10
- tensorflow-gpu == 1.8.0
- keras == 2.2.4
- gensim == 3.6.0
- ekphrasis == 0.5.1