EXPLOITING QUALITY OF PRODUCTS TO IDENTIFY CUSTOMER’S ATTITUDE USING MACHINE LEARNING METHODS

Abstract

With the rise of e-commerce and online shopping, consumers are increasingly relying on the reviews and feedback of other customers to make informed decisions about the products they are considering purchasing. While product descriptions, pictures, and videos can be helpful, they may not provide enough detail or nuance for consumers to fully evaluate a product's quality and complexity. As a result, online shops and independent review websites are collecting and aggregating customer feedback to provide a more comprehensive picture of a product's strengths and weaknesses, which can be especially helpful for consumers who are unable to see or handle a product in person before making a purchase. The proposed work is focused on using natural language processing and machine learning methods to automatically identify the quality and complexity of products using consumers' reviews.  The proposed study involves a semantic and contextual analysis of consumers' reviews using an appraisal framework. This approach aims to identify and evaluate the opinions and attitudes expressed by consumers in their reviews of products. The appraisal framework is a linguistic approach that considers how language is used to convey judgments, feelings, and evaluations, and can be used to analyze the opinions expressed in written or spoken text. Using machine learning to analyze customer feedback can be an effective way to extract valuable insights and identify patterns that might be difficult to discern using manual methods. The main contribution of this work is preparation of manually annotated datasets for machine learning methods to automatically identify product quality and complexity.  The datasets were tested using machine learning methods: Logistic Regression (LR), Support Vector Machine (SVM), Multinomial Naïve Bayes (NB) and Bernoulli NB. In experiments Multinomial NB with over sampling technique ADASYN is compared with other models and achieving average macro F1 score of 0.77 in respect of quality of product.