LSTM-based Deep Learning Model for Emotion Intensity Level by Enhanced Sentiment Classification

Abstract

In last decades, social networking sites like Twitter and Facebook has provided a
platform to express people opinions about product, politics, sports, entertainment and so
on. At the same time, emotion and sentiment classification problem also attracted several
researchers to analyze the state of mind of the people, which finds useful to improve the
quality of the product or services. The recent developments of deep learning (DL)
techniques started to be used for emotion and sentiment classification of social
networking data. This paper presents an effective DL based Long Short-Term Memory
(LSTM) model for emotions with intensity level sentiment classification called LSTMEISC for Twitter text. The proposed LSTM-EISC model follows a two-step procedure,
namely preprocessing and classification. The LSTM model is applied to classify multiple
intensities of Twitter data such as Fear, Anger, Joy, and Sadness. A series of experiments
takes place on SEMEVAL2018 Task-1Emotion Intensity Ordinal Classification dataset to
ensure the effective performance of the LSTM-EISC model. The simulation outcome
pointed out the effective classification results of the LSTM-EISC model over the existing
methods under several aspects.