Customer Satisfaction Assessment System on Transactions E-commerce Product Purchases Using Sentiment Analysis

Abstract

Currently, more products appear, and various services that offer similar products make it difficult for buyers to decide to buy before seeing reviews from other users. The growth of different e-commerce platforms exacerbates this. Users spent more time choosing products on each platform with many alternative considerations, such as looking at ratings, prices, and reviews from other buyers. This study conducted the optimization process of selecting e-commerce products so that users do not have to spend a long time reading every review when they want to buy a product. This research is expected to provide a comprehensive assessment of the purchase transaction of a product from the reviews provided. The data is sourced from product reviews on e-commerce in Indonesia, which are then classified into positive, negative, and neutral sentiments. The data is divided into 10 folds of data using stratified k-fold cross-validation, consisting of training and testing data with ratios of 90% and 10% of the total data. Our research proposed a system that implemented our modified Naive Bayes model to calculate a product's Customer Satisfaction (CSAT) score and compare it with the Google Cloud NLP model. In our model, the log prior and log-likelihood formulas are modified in the algorithm, adding the prefix "NOT_" after the negation words in the preprocessing. This doubled our model’s F1 score and increased the accuracy by 32%, from 59% to 91%, when compared to the Naive Bayes algorithm without modification.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint