Research on Vehicle Retention Rate Prediction Combined with Pre-Trained Language Model

The used vehicle trading market is a huge market. In which the value retention rate is an important indicator that directly affects the transaction price of vehicles and the interests of buyers and sellers. With the continuous development of the used vehicle market, how to accurately predict the value retention rate of used vehicles has become a matter of great concern. It is of great significance to vehicle owners, buyers, financial institutions and other related aspects. The traditional prediction method of used vehicle value retention rate is mainly based on regression model and statistical analysis method. It is necessary to manually extract features and build models, which has problems such as low prediction accuracy and difficult to scale. The pre-trained language model can be trained through large-scale text data to learn the grammar, semantic and contextual information of the language, so as to generate high-quality text representation vectors that can better express the meaning and information of the text. Therefore, by integrating the pre-trained language model, the value retention rate prediction model will have better generalization ability and higher prediction accuracy. In this paper, BERT and ELECTRA pre-trained language models are used to train and fine-tune on self-constructed used vehicle related text datasets and resulting in high-quality text representation vectors. Based on the effect verification, this paper also uses traditional machine learning algorithms to predict the value retention rate, and compares the results of the three. Experimental results show that the method combined with pre-trained language model can significantly improve the accuracy of the value retention rate prediction. Especially it performs well when dealing with large amounts of complex text data such as vehicle reputation. Provide new ideas and methods for intelligent data analysis and application.