GradAscent at EmoInt-2017: Character- and Word-Level Recurrent Neural Network Models for Tweet Emotion Intensity Detection

The WASSA 2017 EmoInt shared task hasthe goal to predict emotion intensity val-ues of tweet messages. Given the text ofa tweet and its emotion category (anger,joy, fear, and sadness), the participantswere asked to build a system that assignsemotion intensity values. Emotion inten-sity estimation is a challenging problemgiven the short length of the tweets, thenoisy structure of the text and the lackof annotated data. To solve this problem,we developed an ensemble of two neuralmodels, processing input on the charac-ter. and word-level with a lexicon-drivensystem. The correlation scores across allfour emotions are averaged to determinethe bottom-line competition metric, andour system ranks place forth in full inten-sity range and third in 0.5-1 range of in-tensity among 23 systems at the time of writing (June 2017).

@InProceedings{LBW17, 
 	 author =  {Lakomkin, Egor and Bothe, Chandrakant and Wermter, Stefan},  
 	 title = {GradAscent at EmoInt-2017: Character- and Word-Level Recurrent Neural Network Models for Tweet Emotion Intensity Detection}, 
 	 booktitle = {Proceedings of the 8th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis at EMNLP-2017},
 	 number = {},
 	 volume = {},
 	 pages = {169--174},
 	 year = {2017},
 	 month = {Sep},
 	 publisher = {ACL},
 	 doi = {10.18653/v1/W17-5222}, 
 }