To Find the Best-Suited Model for Sentiment Analysis of Real-Time Twitter Data

This paper attempts to find the best-suited model for real-time sentiment analysis of tweets. A lot of research is being done on Twitter data and sentiment analysis, but all of them analyze the pre-existing datasets. This causes many problems as Twitter data is mostly dependent on time, location, topic, trends, etc., and the same trained model cannot be used for real-time analysis. Another problem is that a complex algorithm will take a lot of time, and some tweets can be missed. For example, we need to create a model to track tweets to predict terrorist attacks. A model like this should be fast so that we do not miss even a single tweet. Many other factors like traffic and the language of the tweet should also be kept in mind. This type of model cannot be trained on pre-existing datasets. In this paper, tweets are extracted from Twitter in real time and stored in a dataset. Feature extraction and NLP models are applied to the data, and then, machine learning models are run through the data. Support vector machine with bag of words provided the best results, whereas logistic regression with bag of words was the fastest.