A novel method for consumer preference extraction based on perceived usefulness and de-neutral sentiment

Abstract

The accurate assessment of consumer preferences has become increasingly crucial for effective business decision-making. However, merchants face significant challenges in accurately analyzing these preferences due to the overwhelming volume of online consumer feedback, the growing complexity of customer preferences, and the rapid evolution of market trends. Current methodologies are often hampered by various errors and biases, such as the misleading influence of irrelevant information within overloaded data, challenges in extracting latent features from big data, and issues related to neutral sentiment factors. Additionally, the usefulness of information is frequently dependent on voting mechanisms, which are susceptible to biases like the early bird effect. To address these challenges, we propose a novel preference extraction method based on perceived useful information. This approach integrates sentiment analysis rooted in deep learning with neutral sentiment processing and combines multi-criteria decision-making to extract hidden features while mitigating the misleading impact of irrelevant word frequencies on preference calculations. Furthermore, our method employs the principle of information entropy to extract information utility, thereby avoiding common biases associated with traditional voting methods. Experimental results demonstrate the superiority of our method across two case studies: for search-based products, the method achieved an F1 score of 89.6 % and an AUC of 79.4 %, while for experience-based products, it recorded 84.6 % and 80.6 %, respectively. The primary contribution of this research lies in providing a systematic approach to uncover product features and accurately analyze consumer preferences, offering valuable insights for business decision-making. This has significant theoretical and practical implications for product development, marketing, and customer service.