Deep Learning-Based Approach to Classify Praises or Complaints from Customer Reviews

Abstract

Online reviews carry customers’ opinion about product or service and help the customers to make online purchase-related decisions. These reviews are analyzed by business organizations to understand customer sentiment w.r.t. product/service. The extreme opinions like praise and complaint sentences are a subset of positive and negative sentences and difficult to find. Praise sentences are more descriptive in nature. Praises contain more nouns, adjectives, and intensifiers as compared to plain positive sentences and complaint sentences contain more connectives, adverbs as compared to plain negative sentences. In the past machine, learning methods are used to identify extreme opinions but the accuracy of such methods is very limited. This paper proposes (1) linguistic feature-based approach for reviewing sentences filtering and (2) machine learning-based and deep learning-based approach to classifying review sentences as praises or complaints. These praise and complaint sentences can be further analyzed by business organizations to identify the reasons for customer satisfaction or dissatisfaction. It can also be used for creating automatic product description from online reviews in terms of pro and con of the product/service. The performance of the machine learning classifiers with proposed hybrid features and deep learning-based classifiers using dense neural network, CNN, and multichannel CNN was evaluated by training and testing the deep neural network with a set of important words such as nouns, adjectives, intensifiers, and verbs present in the sentence. Hotel domain reviews were evaluated using the parameters accuracy, precision, recall, and F1-score. The proposed method showed excellent results as compared to state-of–the-art classifiers.