关键词: 交通运输经济, 乘用车需求趋势, LDA主题模型, 在线评论, 情感倾向

Abstract: In the Internet environment, online reviews reflect consumers’ most intuitive feelings and needs, which will affect consumers’ cognition of products and their decision-making behavior. Mining the potential value of online reviews has become an effective way to obtain intelligence, which is conducive to passenger car manufacturers and sales companies to analyze market demand in multiple dimensions and improve product design and competition strategies. Transforming massive information containing real needs into structured data to extract valuable intelligence information is an urgent problem to be solved. Based on this, this paper uses the implicit Dirichlet assignment (LDA) topic model to learn and obtain document topics from the passenger car review dataset, and sorts the subject words extracted from the documents. The word frequency-inverse document algorithm (TF-IDF) is used to calculate the sentiment value of the subject words. Finally, a data panel related to sales volume is constructed to transform consumers’ real needs into design problems, which is applied to the development of passenger car products. From the subject characteristics and emotional attitude characteristics, the emotional tendency of different model review topics is estimated, and the panel data model of user preference analysis is constructed corresponding to the passenger car sales volume in 2009-2019.
Based on the LDA model, this paper conducts cluster analysis of the topic characteristics of passenger car consumers’ attention, then uses the TF-IDF algorithm to calculate the weight value and the sentiment tendency analysis method to carry out statistical analysis of the comment data to determine the topic of consumer attention,and finally analyzes the tendency change in consumers’ attention topic characteristics through the measurement method and constructs a consumer attention model based on LDA-TFIDF panel data. With the help of the Octopus web page data collection device, this paper conducts online reviews and sales data collection of passenger car products on the three major automobile consumption websites “Autohome”, “Pacific Automobile Network” and “Sohu Auto Channel” with the highest customer engagement rate in China. The Lagrange multiplier method and dual method are used to clean word segmentation and denoising the data, and the LDA topic model is used to analyze the processed data. The main check is the number of words with the same information in the document, and then the TF-IDF algorithm is used to combine the sentiment tendency analysis to assign the characteristics of the comments.
In order to better explore the impact of consumer sentiment on the market share of passenger cars, this study uses the TFIDF value of the previous year’s topic attention in the model to predict the current market share and uses the generalized least squares method (GLS) for estimation. In order to reduce the degree that the prediction results are affected by extreme values, the paper estimates all variables by focusing on the median value of sales weight, then interprets the coefficient of the linear term as the elasticity of passenger cars with median characteristics to market share, and finally uses the linear term to interact with the year to analyze the change in consumer sentiment. In the study, the market share of passenger cars is taken as the dependent variable, and the user’s emotional inclination value as the independent variable.In order to avoid pseudo-regression phenomenon and ensure the stability of the variables, the ADF test should be used to perform unit root test or cointegration test on the variables before establishing the econometric model. The experiments show that: this method can effectively extract online review intelligence information; through the interaction with the year to analyze the productdemand trend, consumers only pay attention to fuel consumption by 8.64% in the past 11 years; among them, the increase in indicators such as appearance, interior and comfort, which are more closely related to the quality of life, is much higher than that in traditional indicators such as power and space. This shows that fuel economy is becoming less and less of a concern and is no longer a decisive factor for consumers to buy passenger cars.
For the past 11 years, consumers’ sensitivity to the fuel economy of passenger cars has gradually decreased, and high-power and large-space models are more popular, but the trend of more pollution emission is increasing. The increase in residents’ income has upgraded consumption,and the appearance, interior and comfort have a more significant impact on consumers’ purchasing decisions and will replace fuel consumption as the main factor affecting consumers’ purchasing decisions. Therefore, passenger car production and sales enterprises should actively use online review analysis results to track consumer preferences and adjust product development directions in a timely manner to enhance market competitiveness. Government should increase policy efforts to guide energy conservation and emission reduction in order to achieve the goals of “peak carbon dioxide emission” and “carbon neutrality” in the transportation sector.

Key words: transportation economy, passenger car demand trend, LDA topic model, online reviews, emotional tendency