Analysing the Sentiments in e-newspaper Contents using Novel Bidirectional Encoder Representation for Transformation - BERT over Linear Regression Algorithm

Abstract

Aim: The aim of the study is to detect sentiment analysis from the good ones and improve the false positivity rate by using the proposed Novel Bidirectional Encoder Representation for Transformers-BERT over Linear Regression Algorithm.
Materials and Methods: Sample groups that are considered in the project can be classified into two, one for Novel BERT over Linear Regression, which are tested using 0.80 for G-power to determine the sample size 20 and for t-test analysis. 23000 BBC e-news dataset that collected data from NewsBrief and MediSys.
Results: The automatic feature selection of the BERT algorithm splits the data with best fit, which has an average accuracy of 83.50%, which by far seems to be better than the Linear regression which gives around 77.80%. The significance is around 0.039 (p<0.05) and therefore there is a statistical insignificant difference among the study group.
Conclusion: Novel BERT seems to be better in finding the Sentiment in e-newspaper content of BBC e-news dataset over the Linear Regression Algorithm.