A Recommender System for Developing Marketing Strategies for Online Products Based on Sentiment Analysis of User Reviews Using Deep Learning Algorithms

Abstract

This study aimed to develop a recommender system for formulating marketing strategies for online products based on sentiment analysis of user reviews and to compare the performance of deep learning algorithms in predicting product sales. This applied, descriptive-analytical study was based on textual data obtained from users of an online store. User reviews of products were employed as the primary input data, and the sentiments expressed in the reviews were classified into positive, negative, and neutral categories. Three deep learning algorithms—Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM)—were implemented for sentiment analysis and sales prediction. The model incorporated product- and review-related features, including user sentiment, price, brand, quality, delivery, packaging, support, warranty, and other product characteristics. Model performance was comparatively evaluated using accuracy, precision, F1-score, recall, confusion matrices, and receiver operating characteristic (ROC) curves. Comparative results demonstrated that CNN achieved the strongest overall performance, followed by LSTM and RNN. Accuracy values were 0.966, 0.960, and 0.953 for CNN, LSTM, and RNN, respectively. Corresponding precision values were 0.956, 0.954, and 0.944; F1-scores were 0.931, 0.917, and 0.904; and recall values were 0.964, 0.957, and 0.949. The confusion-matrix and ROC analyses further supported the relative superiority of CNN in classification and prediction. Feature-importance analysis based on the CNN model indicated that other users’ comments had the greatest contribution to classification, followed by user sentiment, delivery, brand, and packaging. Sentiment analysis of user reviews using deep learning algorithms, particularly CNN, can provide an effective analytical basis for predicting product sales and developing recommender systems that support product-specific marketing strategy formulation in online retail environments.