IMS at EmoInt-2017: Emotion Intensity Prediction with Affective Norms, Automatically Extended Resources and Deep Learning.

摘要: Our submission to the WASSA-2017 shared task on prediction of emotion intensity in tweets is a supervised learning method with extended lexicons affective norms. We combine three main informa- tion sources random forrest regressor, namely (1), manually created resources, (2) automatically lexicons, and (3) output neural network (CNN-LSTM) for sentence regression. All feature sets perform similarly well isolation (≈ .67 macro average Pearson correlation). The combination achieves .72 official test set (ranked 2nd out 22 participants). analysis reveals that performance increased by providing cross-emotional predictions. automatic extension lexicon features benefit from domain specific embeddings. Complementary ratings norms increase impact features. resources (ratings 1.6 million twitter words) our imple- mentation publicly available at http: //www.ims.uni-stuttgart.de/ data/ims_emoint.