Knowledge Graph Embedding Based Sentiment Analysis of Product Reviews using LSTM and Fuzzy Logic

Abstract

The popularity of online buying has skyrocketed, particularly during and after the COVID-19 era. Product reviews are a huge asset when making decisions about online purchases. Product reviews can also aid in product improvement from the standpoint of retailers and producers. It takes a lot of effort and time to read through each. With the aid of sentiment analysis techniques, researches have experimented with analysing product reviews. Intelligent systems that use machine learning and deep learning algorithms are in high demand because they allow customers to quickly uncover the trend in a product rather than having to read through many evaluations. The accuracy of these intelligent systems is still debatable, though. Different researchers have modified various strategies in the past to increase accuracy. This study is another effort in that similar approach. But this study takes a somewhat different track. Product review analysis is an NLP problem, therefore textual input is prepared before being fed into a deep learning model using knowledge graph embedding. We think that using a deep learning model like LSTM in conjunction with knowledge graphs can produce the much-needed system with increased recall, precision, and accuracy. The classification of three Amazon datasets from Kaggle into the four sentiment categories of "Positive", "Negative", "Highly Positive", and "Highly Negative" is the purpose of this study. Amazon product review benchmark datasets like "Customer reviews of Amazon Products," "Amazon Cell Phones Reviews," and "Amazon Fine Food Reviews" were used to test the suggested LSTM+KGemb model. Results are achieved with accuracy levels of 97%, 98%, and 96%, respectively.