Certain Investigation On Cause Analysis Of Accuracy Metrics In Sentimental Analysis On News Articles
Prabhu Ram Nagarajan, T. Meeradevi, Shri Ram A.M Hari, K Kavippriya., Madhumitha G · 2021 5th International Conference on Electronics, Communication and Aerospace Technology (ICECA) · 2021
In the modern world, news is reaching the people through online mode. News readers are inclined to news based on interactivity and immediacy. The current technology facilitates all users a direct news on any event in real world. The sentimental analysis can be used to determine an emotional rating of text data sources from news articles such as Times of India, India Today etc., TFIDF methodology of word vectorization and Word Embedding of word vectorization is used to analyze the effectiveness in accuracy of models. The experiments have been performed on datasets using linear SVM and Gaussian Naive Bayes classifiers, in which consideration in decision making is based maximum scores acquired at certain class is ranked to be an important news to public. The accuracy for unigram by TFIDF vectorization is 62%, which is lesser than the accuracy of bigram by TFIDF vectorization which is 65% using SVM classifier. For Gaussian Naive Bayes classifier, the accuracy of unigram model is 61 %. Even though the accuracy of Gaussian Naive Byes classifier is less, independent feature considering nature of this classifier makes this model more effective in predicting the socially important news than other classifiers and also common occurrence of token of word between each class will be cause of affect the F1 score of models.