Enhancing Marketing Strategies through AI-Powered Sentiment Analysis: A Comparative Study of BERT, LSTM, and Sentiment Lexicon Approaches

Abstract

This research paper investigates the effectiveness of integrating AI-powered sentiment analysis into marketing strategies, focusing on a comparative analysis of three prominent approaches: Bidirectional Encoder Representations from Transformers (BERT), Long Short-Term Memory (LSTM) networks, and traditional sentiment lexicon methods. With the rising importance of understanding consumer emotions in digital marketing, this study seeks to determine which technique offers superior accuracy and actionable insights. The research involves the collection and preprocessing of extensive social media datasets, followed by the implementation of each sentiment analysis method. BERT, with its contextual understanding capabilities, LSTM, known for handling sequential data, and sentiment lexicons, which provide rule-based sentiment categorization, were evaluated on metrics such as precision, recall, F1-score, and computational efficiency. The results indicate that BERT consistently outperformed the others in nuanced sentiment detection, while LSTM demonstrated strengths in handling longer text sequences. Sentiment lexicons, although less adaptable to context, offered simplicity and speed, making them suitable for quick, preliminary analyses. Based on these findings, the paper discusses the potential of each approach to refine marketing strategies by accurately gauging consumer sentiment, ultimately enhancing brand positioning and customer engagement. The study concludes by recommending a hybrid approach, utilizing the strengths of all three methods to optimize marketing strategies tailored to specific data characteristics and business objectives.