Consumer interaction intention recognition based on transformer architecture

Abstract

In the current digital business environment, massive online consumer reviews have become an important source of information for understanding user behavior. Accurately identifying the interactive intentions contained in reviews plays a key role in improving customer service quality, optimizing product development and formulating precise marketing strategies. Traditional analysis methods are difficult to effectively distinguish different types of intentions expressed by consumers. In this study, a hybrid model combining particle swarm optimization algorithm and Transformer architecture is proposed to automatically identify four typical intentions in consumer reviews: information provision, emotional expression, help seeking and suggestion/ recommendation. Through the verification on the public Amazon review data set, the method is superior to the traditional machine learning and benchmark deep learning model in many indicators, such as accuracy, F1 score, and shows a stronger ability to distinguish intentions. This technology provides a feasible solution for enterprises to realize the intelligent processing of consumer feedback.