Abstract: In this study, we propose a hybrid CNN-BiLSTM model for sentiment analysis of customer reviews, specifically designed to address the challenges of precision marketing in e-commerce, particularly in the context of customer churn. One of the main issues businesses face is the inability to fully understand evolving customer needs, often relying on static profiles that lead to poor decision-making and reduced customer satisfaction, ultimately resulting in higher churn rates. The primary objectives of this study are threefold: to predict customer satisfaction based on product reviews, to identify key factors that enhance satisfaction for more targeted marketing efforts, and to develop a scalable approach capable of processing large volumes of customer feedback data. By combining the strength of convolutional neural networks (CNN) for feature extraction with bidirectional long short-term memory (BiLSTM) networks for capturing contextual dependencies, the model aims to deliver a deeper understanding of customer sentiment. This integrated approach enhances the accuracy of recommendations and marketing strategies, helping to reduce churn. The proposed architecture consists of three main stages: data preprocessing using Doc2Vec for document embedding, feature extraction through CNN and max-pooling layers, and sentiment classification via BiLSTM and fully connected layers, culminating in a final classification of positive, negative, or neutral sentiment.

Keywords: Artificial intelligence; precision marketing; sentiment analysis; Natural Language Processing; E-commerce