Transformer-Based NLP Models for Predictive Consumer Behavior Analysis

The social media has transformed to be one of the top drivers of consumer perception and purchase intention especially in new luxury markets like India. In this paper, transformer-based Natural Language Processing techniques (NLP) will be suggested as predictive, AI-based methodology to model consumer behavior. Unlike the other past descriptive studies, we use a BERT based classifier of sentiment and intent to predict purchase forecast in social media content. It was further extended to 5,100 Instagram captions and 1,200 online product reviews each having the positive, neutral or negative sentiment labels and the binary purchase-intent indicators annotated on them individually. The models that were analyzed include Logistic Regression, Support Vector Machine (SVM), Long Short-Term Memory (LSTM) and BERT. According to the experimental outcomes, it has been established that BERT is always superior to the traditional machine learning and deep learning methods with an accuracy of 91.6, F1-score of 0.92 and an area under the ROC curve (AUC) of 0.94 in the prediction of purchase intent. Such results demonstrate that contextual language representations provide a better ability to capture the subtle emotional and aspiration signals in the luxury consumption environment compared to rule-based or association-mining systems. The study provides a predictive low scale alternative to Apriori based market basket analysis and has explicit applications to personalized marketing campaigns, recommendation systems and real-time social listening.