A Comparative Study on the Performance of Deep Learning Algorithms for Detecting the Sentiments Expressed in Modern Slangs

摘要

Sentiment analysis is a text investigation technique that distinguishes extremity inside the text, regardless of whether an entire document, sentence, etc. Understanding individuals' feelings are fundamental for organizations since customers can communicate their considerations and emotions more transparently than any other time in recent memory. In this paper, the proposed model is the sentimental analysis on Twitter slangs, i.e., tweets that contain words that are not orthodox English words but are derived through the evolution of time. To do so, the proposed model will find the root words of the slangs using a snowball stemmer, vectorizing the root words, and then passing it through a neural network for building the model. Also, the tweets would pass through six levels of pre-processing to extract essential features. The tweets are then classified to be positive, neutral, or negative. Sentiment analysis of slangs used in 1,600,000 tweets is proposed using long short-term memory (LSTM) network, logistic regression (LR), and convolution neural network (CNN) algorithms for classification. Among these algorithms, the LSTM network gives the highest accuracy of 78.99%.