A Sentiment Aware Product Recommendation System Using Optimization Techniques and Deep Classifiers

Abstract

Electronic Commerce (e-Commerce) is growing rapidly due to the drastic growth of population and digitalization. Many people are interested to buy from their house itself without visit shops through online for affordable cost. Now a day, the people are saving their money and time with the help of this e-commerce. Moreover, many offers are provided by the e-commerce website for the various goods. People are saving their money and also getting lot of choices in their mobile phone and laptop itself. The data mining, Machine Learning and Deep Learning algorithms are helpful for categorizing the products or goods by analysing the product reviews. To analyse the product reviews, the Natural Language Processing is playing major role and perform the required data pre-processing. On the other hand, the sentiment analysis is also contributing more in the process of analysing the review comments effectively. This research work proposes a new Product Recommendation System (PRS) with the incorporation of data pre-processing, sentiment analysis, feature selection, optimization and classification with fuzzy temporal rules. This research work consists of six different works as six PRSs that are handling the various products and recommends the suitable products to the customers. newlineThe first work of this thesis, a new Content Aware Support Vector Machine (CASVM) has been built for identifying the co-relationships from sentiment to feature carefully by analysing the feature vectors of the input data with labelled class. In this work, a hyper plane is used in SVM for performing the classification and classified as two classes. Here, the margin becomes bigger and the risks will be lesser. The unprocessed text is extracted from X page and extracts the necessary feature vector for processing further. Generally, the classifier classifies the dataset as two classes but the sentiment aware classification process is comprised with three classes. For overcoming these challenges, the proposed CASVM uses the binary classification that is a one