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

TL;DR

This paper presents an ensemble neural network approach combining character- and word-level models for predicting emotion intensity in tweets, achieving competitive results in the EmoInt shared task.

Contribution

It introduces a novel ensemble of character- and word-level neural models with lexicon integration for tweet emotion intensity estimation.

Findings

Achieved 4th place in full intensity range

Achieved 3rd place in 0.5-1 intensity range

Demonstrated effectiveness of combined neural models

Abstract

The WASSA 2017 EmoInt shared task has the goal to predict emotion intensity values of tweet messages. Given the text of a tweet and its emotion category (anger, joy, fear, and sadness), the participants were asked to build a system that assigns emotion intensity values. Emotion intensity estimation is a challenging problem given the short length of the tweets, the noisy structure of the text and the lack of annotated data. To solve this problem, we developed an ensemble of two neural models, processing input on the character. and word-level with a lexicon-driven system. The correlation scores across all four emotions are averaged to determine the bottom-line competition metric, and our system ranks place forth in full intensity range and third in 0.5-1 range of intensity among 23 systems at the time of writing (June 2017).