Implementation of EM Algorithm in Opinion Mining Movies Review Case Studies

Abstract

Movies are very familiar to everyone, from children, adolescents to adults, whether just because they want to watch, a hobby, or fill their spare time. Movies that used to be watched only on television and had to wait months after release or directly to the cinema, with the development of technology, of course, it is increasingly easier for everyone to enjoy movies, now they can be watched through paid television services to smartphones. One of the websites that viewers often use to review movies they have watched is IMDb. The data review can be used to get an opinion or opinion mining from the audience, whether the title of the movie being reviewed is good or not. One of the algorithms that are often used is Naïve Bayes, apart from being easy to implement, Naïve Bayes is also known to be very fast and easy to use to predict classes on a test dataset. The purpose of this study is to see how much influence the Expectation-Maximization to increase accuracy on implementation of Expectation-Maximization algorithm in opinion mining movies review case studies. From the results of this study using the Expectation-Maximization method, it was found that the accuracy increased by 4% compared to using only Naïve Bayes.

A model-based algorithm for quantity and parameters of clusters discovery

“…Jeff Wu gave a proof of the convergence of EM algorithm outside the exponential family distribution [11]. For GMM with two components, Daskalakis et al [12]; Xu et al [13] provide a global analysis of EM for the mixture of two Gaussians and deliver results that guarantee convergence of EM for this specific problem from a random initialization. Kwon et al show here that EM converges for mixed linear regression with two components (it is known that it may fail to converge for three or more), and moreover that this convergence holds for random initialization [14], and completely characterize the convergence behavior of EM, and show that the EM algorithm achieves minimax optimal sample complexity under all SNR regimes [15].…”