Simplified Approach for Predicting Emotions of Multi-Turn Textual Utterances

Sentiment analysis is a well-known topic in Natural Language Processing (NLP). After a lot of research in this sector, the Emotion Analysis of Conversations prompted out as a trending topic. The main idea behind the Emotion Analysis of Conversations is to extract and classify the emotions expressed through a conversation. The scope of this research is adopted from "EmoContext" (SemEval 2019 - Task 3), a competition conducted by Codalab - Microsoft. There are some models that achieve a significant accuracy in this section but they are too much complicated to understand. There is a need for a model that achieves significant accuracy with a simplified approach for many uses. Here, we present a model that has achieved a microF1 score of 0.742 with our simple CNN-BiLSTM model and customized FastText word embedding, which comfortably betters the 3rd Quartile value of 0.7317 and stands up into the top quarter of the leader board of EmoContext competition.