Robust Review Rating Prediction Model based on Machine and Deep Learning: Yelp Dataset

Public reviews for a business are very important and help the business to measure the quality and excellence in different directions which leads to predict the worth of a business in the market. In other words, reviews have a very high impact on business revenue. In this paper, we focus on reviews for all kinds of restaurants business and have proposed a sentiment analysis and opinion mining model to perform the classification on business reviews. In order to achieve robust results both binary and multilabel classification are used used by using a large and rich text reviews dataset provided by Yelp Dataset Challenge round -13. Extensive and series of experiments have been done and compare the results of a machine learning based algorithm “Multinomial Naive Bayes” and deep learning algorithm “convolution Long Short Term Memory'” (CLSTM) with word2vec and Global Vector (Glove). After analyzing the performance of each model with different metrics, it has been observed that the best model for classifying the review ratings is CLSTM. We have also found the role of bias in the machine and its importance in explaining the performance differences observed on specific problems.