Audio Based Sentiment Prediction Model
When something gets newly introduced among the people, it is important to know about people’s opinion on that and this is where sentiment analysis comes into play. Model that’s built for sentiment analysis helps to predict the sentiment of the content. The audios that we have collected are taken as the dataset for the system and our proposed model mainly overcomes the language barrier as the system involves converting of other language audio reviews to English text. The preprocessing, feature extraction are applied on the text before training the model. The data is split into testing and training data based on testsize parameter. Initially, we have carried out training phase using three supervised machine learning algorithms i.e., SVM, Random Forest, Multinomial NB and one neural network based deep learning algorithm LSTM used with RMSprop, Adamax optimizers. The training and stratified sampling is done and we have chosen the best algorithm based on the average of the accuracy results produced by system when performed against various test samples. SVM, Random Forest, Multinomial NB, LSTM with RMSprop, LSTM with Adamax have given accuracies of 80.02%, 77.08%, 76.49%, 72.36%, 68.24% on ration dataset and 86.67%, 81.67%, 81.67%, 74.17%, 60.84% on JVD dataset respectively. As SVM algorithm gave the best accuracy, we have selected it as the algorithm for our sentiment prediction model.