Energy vehicle user demand mining method based on fusion of online reviews and complaint information

Abstract

Understanding user demand is the basis for enterprises to improve their product quality and services. Online reviews have become a critical reflection of user demand. In particular, complaint information usually indicates defects in products or services. This study proposes a user demand mining method that integrates online reviews and complaint information. The term frequency-inverse document frequency (TF–IDF) method and expert consultations were adopted to generate a product attribute set. The latent Dirichlet allocation (LDA) method was also used to add more attributes to the product attribute set. This study analysed the aspect-level sentiment of each product attribute using the fine-tuning bidirectional encoder representation from transformers (BERT) model. The labelled dataset was used to train the BERT classification model to classify the product complaint information. Finally, the sentiment analysis results of online reviews and the classification results of complaint information were comprehensively analysed to obtain more accurate and comprehensive user demand elements. An example of a car product verified the effectiveness of the proposed method in the current paper. The results reveal that the method proposed in this study can integrate online reviews and complaint information to improve the accuracy and comprehensiveness of data mining results, thereby playing an auxiliary decision-making role for enterprises to accurately obtain user demand information and formulate product development strategies.