Multi-domain Emotion Detection using Transfer Learning

Anonymous

Abstract: The task of emotion detection in text, particularly in informal and spontaneous messaging, such as email, posts, or tweets, varies in its scope and depth depending upon the requirements of the end application as well as the domain of use. The most popular emotion categories reported in research include the Ekman’s or Plutchik’s emotion models, but often the application domain requires a more specialized emotion categorization, for which there are insufficient annotated datasets available for training. It is additionally complicated by different perceptions and definitions of emotion labels in different domains. The popularity of empathetic systems across a wide range of industries and applications has given rise to the the task of multi-domain emotion detection to increase its adaptability and resiliency across domains. In this paper, we present a generalized approach of emotion detection that can be adapted to any domain and any set of emotion labels with minimal loss in performance. The multi-domain-emotion model could be plugged into any emotion detection application without any further training or fine-tuning. We show the zero-shot and few-shot performance of our approach on the publicly available SemEval 2018 dataset and also a new dataset consisting of tweets related to the French elections in 2017. This approach demonstrates good performance in predicting emotion categories previously unseen to the model, including domains different than those on which the model was originally trained. We further propose a few ways to boost the model performance with the availability of a small annotated dataset in the target domain.

Paper Type: long

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining