User Evaluation Sentiment Analysis Model Based on Machine Learning

Abstract

In order to obtain a correlation between the online evaluation of three online sales products provided by Amazon and their corresponding star rating, we use the tf-idf algorithm for text mining on three online product evaluations, extract each product based on the AFINN sentiment dictionary of sentiment words with tf-idf values included in consumer reviews, and add their intersections into a thesaurus. We also extract the brand information implied in the information and observe a certain difference in the consumer evaluation of each brand. Based on our established thesaurus, we conduct a sentiment analysis of consumer text reviews, assign corresponding sentiment scores, and obtain the corresponding review score variables. Then, we find a significant correlation between the consumer sentiment score and its corresponding star rating. Based on the star rating, verified purchase, and helpful votes variables in the data, we use machine learning models to learn the sample characteristics of each star rating. The final three-product decision tree model has an accuracy of 78.33%. The accuracy rate of the 5-star evaluation is 99.41%, which can better distinguish the extreme evaluations of consumers to a certain extent. An evaluation model based on product brand, consumer star rating, and market share of product brand is established to predict the potential success or failure of the product to a certain extent.