Understanding the Impact of Data Parallelism on Neural Network Classification
S. Starlin Jini, N. Chenthalir Indra · Optical Memory and Neural Networks · 2022
Abstract Social Networks have become a platform to express each moment of a person via texts in widespread. With the help of a lot of words, ideas, thoughts and good memories are shared. Twitter is a social network platform that contains a rich source of data targeted by different organizations to analyze people’s emotions, sentiments, and opinions. Previously different big data clustering methods such as semantic driven subtractive clustering method and metaheuristic clustering method have been presented. These methods failed to perform the classification of emotions in a simple and cost-effective manner. In order to overcome the issues, efficient clustering-based classification of emotions of tweets has been presented. In this paper, after pre-processing the obtained dataset, a step of clustering has been performed using the Bayesian Mixture Model. By performing this clustering, an efficient classification result is obtained. After clustering, feature extraction processes collects both semantic and sentimental features then give it as input to Probabilistic Neural Network (PNN) for training purpose. The data parallelism concept is integrated so as to portion the training data in order to minimize training time and make the system efficient. The training phase gets completed so as to predict the testing results. In the testing results, the classification of tweets with respect to their probability of emotions is obtained. This method was then implemented in the MATLAB tool platform, and the analysis results with accuracy and of error. Comparing this analysis of proposed method to previous method, it has been proved that the proposed model is efficient and overwhelmed conventional methods.