The Impact of Consumer Sentiment on Sales of New Energy Vehicles: Evidence from Textual Analysis

: The advancement of new energy vehicles (NEVs) represents a strategic initiative to combatting climate change, mitigating the energy crisis, and fostering green growth. Using provincial panel data from China between 2017 and 2022, in this study, we applied machine learning techniques for sentiment analysis of textual reviews, used word frequency statistics to explore consumers’ views on the attributes of new energy vehicles, and constructed a consumer sentiment index to study the impact of consumer sentiment on NEV sales. Considering the dependence of NEVs on a charging station, this paper explores the nonlinear impact of the popularity of charging stations on the relationship between consumer sentiment and sales of new energy vehicles. The findings indicate the potential for enhancement in the areas of space, interior design, and comfort of NEVs. Additionally, consumer sentiment was found to facilitate the diffusion of NEVs, with this effect being heterogeneous across different educational backgrounds, income levels, and ages. Furthermore, the availability of per capita public charging stations was shown to significantly reduce range anxiety and encourage consumer purchasing behavior.


Introduction
The continuous increase in global greenhouse gas emissions, along with the frequent occurrence of extreme weather events and rising sea levels, have had unprecedented impacts on both natural ecosystems and human socioeconomic activities [1].The transportation sector is a major contributor to global carbon emissions, with energy-related carbon emissions from the global transport sector accounting for approximately 24% of the total emissions in 2019, according to the International Energy Agency (IEA).Among transportation modes, road transport is the largest source of emissions and is responsible for more than half of all transport-related greenhouse gas emissions [2].Therefore, addressing the rapid growth of carbon emissions from road transport is crucial to achieving global carbon peaks and carbon neutrality goals.The development of strategic emerging industries, particularly the promotion of NEVs, plays a vital role in effectively alleviating energy and environmental pressures [3].With widespread recognition of the importance of environmental protection, a low-carbon economy, and energy conservation, the transformation of the automotive industry is a prevailing trend.Governments worldwide have shown strong support for NEVs due to their pollution-free nature and use of renewable power fuels.
In China, the carbon emissions of the transportation sector cannot be overlooked.In 2019, China's transportation industry emitted 1.274 billion tons of CO 2 , ranking second only to the United States.This accounted for 12.42% of the country's total CO 2 emissions and 14.82% of the world's total CO 2 emissions from transportation [4].Among transportation-related carbon emissions, China's road transport sector is the primary contributor, accounting for 87% of the total.Passenger car emissions alone make up 33% of these emissions.A recent study conducted in Saudi Arabia on carbon emissions from electric vehicles revealed that substituting 1% of fuel vehicles with electric vehicles could reduce carbon emissions by 0.5-0.9%[5].However, by the end of 2022, the ownership of NEVs in China constituted less than 5% of the total [6].Thus, it is evident that there is still significant untapped potential for the development of NEVs in China.
Existing research has demonstrated that psychological factors play a fundamental role in influencing consumers' decisions to purchase NEVs [7].These psychological factors not only directly impact consumers' willingness to adopt NEVs but also influence objective factors [8][9][10].The emotion-cognition-behavior framework proposes that people's emotion and cognition will affect their behavior, and it can be understood that potential consumers will have different cognition according to the emotions expressed by purchased consumers, which will affect their purchase behavior.Among them, consumers' cognition is affected by their personal experience, knowledge level, information acquisition, and other factors.Nevertheless, previous studies have not comprehensively explored how consumer sentiment affects NEV sales.Moreover, most of the existing research relies on questionnaire surveys to study consumers' willingness to adopt NEVs.The limitations of this methodology, such as in sample size, timeliness, cycle, spatial span, richness, and objectivity [11], make them challenging to effectively determine the influence of psychological factors on consumers' purchasing behavior.Therefore, this paper presents a study that constructed a consumer sentiment index using the text analysis method.The objective is to investigate the impact of consumer sentiment on the diffusion of NEVs.Considering the differences in consumers' cognition, this paper discusses the differences in NEV consumer groups from the perspective of consumers' personal characteristics.
The main contributions of this paper are as follows.First, it provides a new research perspective.With reference to investor sentiment on the stock market, this paper introduces machine learning and text analysis methods into the research and constructs a new consumer sentiment indicator to estimate consumer sentiment in the NEV market.Second, employing the two-way fixed effects model, the heterogeneous impacts of consumer sentiment on NEV sales were examined by educational background, income level, and age.Finally, the nonlinear association between consumer sentiment and NEV sales was investigated considering the infrastructure construction level, which can provide valuable insight for the formulation of differentiated infrastructure investment policy.
The structure of this paper is as follows.Section 2 provides a comprehensive review of the relevant literature.Section 3 introduces the model's construction and data sources utilized in our study.Section 4 presents the results of the empirical analysis.Section 5 is a discussion of the findings.Finally, Section 6 concludes our research and provides suggestions for future studies.

Influencing Factors of NEV Development
The NEV industry has developed rapidly under the policy support of the central government.Developing NEVs plays an important role in alleviating energy shortages and improving environmental quality.Numerous studies in recent years have revealed that mileage, charging time, and charging infrastructure are the primary barriers to NEVs' adoption.Among these challenges, the lack of a well-developed public charging infrastructure has become the biggest obstacle hindering the popularity of NEVs [12].The availability of a public charging infrastructure is crucial to facilitating the widespread adoption of NEVs [12], and the location of charging infrastructure greatly impacts the demand for charging facilities.Wirges [13] proposed a dynamic spatial model for Stuttgart, Germany, suggesting that the installation of public charging infrastructure should be concentrated in the largest urban centers within the region.In real-life scenarios, consumers often have irrational expectations regarding mileage, and their charging behavior is influenced by anxiety over their vehicle's range [12].Vehicle range anxiety impedes the widespread adoption of NEVs, and a well-planned distribution of charging infrastructure can help alleviate this concern.Currently, China still has issues pertaining to infrastructure construction, and there is a need for greater attention to the development of customized governmental policies aimed at promoting the charging infrastructure market.

Public Perception of NEVs and NEV Sales
It is important not only to address the technical limitations of electric vehicles, such as their battery capacity and weight, but also to investigate consumers' personal preferences [14] and psychological factors.Factors such as perceived benefits, risk perception, emotion, and trust can significantly impact public acceptance and willingness to purchase electric vehicles [15].Research conducted by Plötz et al. [16], and others has shown that attitude is the most influential predictor of consumers' intention to adopt NEVs.Attitude includes both cognitive and emotional components [17], and emotions can also play a significant role.Consumers express different emotions when using NEVs, with those experiencing positive emotions being more inclined to adopt battery electric vehicles, and vice versa [8,18].Therefore, it is necessary to examine consumers' perceptions of NEVs and their emotional preferences to determine their likelihood of purchasing NEVs and to identify the various factors that drive different emotional responses.In addition to the objective and subjective factors, individual characteristics such as gender, age, education level, and income are significant considerations, and these factors will affect consumers' cognition.Existing research indicates that well-educated young and middle-aged consumers tend to have a stronger intention to purchase [6,19,20].Furthermore, personal income serves as a factor that influences consumers' decisions to acquire NEVs [21].By exploring variations in consumers' concerns and emotional preferences regarding NEVs, governments and manufacturers can develop more tailored strategies for different consumer segments.However, the literature does not effectively describe consumers' concerns and emotional preferences toward NEVs.Therefore, this study aims to analyze consumer comment information to gauge consumer sentiment and bridge this research gap.

Measurement Methods of Consumer Sentiment
The methods commonly utilized to gather consumers' opinions or emotions include questionnaires [22,23] and interviews [24].However, traditional questionnaires and interviews have limitations in accurately reflecting consumers' emotions [25], particularly in the context of national issues where sample questionnaires can only make a small contribution [26].The widespread development of the mobile internet has achieved an unprecedented scale and professional autoplatforms such as Autohome have accumulated a wealth of real user comments, which serve as valuable after-sales feedback containing both positive and negative emotions from different consumer groups.These comments have strong potential to influence future consumer purchasing behavior [27].Therefore, consumers' opinions or emotions can be obtained based on online comments.This will also create a new problem of how to process this information so that it can reflect the emotions of consumers.At present, the research on consumer psychological factors in the NEV market needs to be further deepened.We found that the measurement method of consumer psychological factors in the stock market is relatively mature.The measurement methods of investor sentiment are divided into three categories: a direct measurement method, an indirect measurement method, and the construction of an investor sentiment index based on social media.Among them, the construction method of investor sentiment index based on social media is our focus.Scholars build investor sentiment indicators based on the classification of posts, divide posts into positive emotions and negative emotions, and calculate the difference between the number of posts with positive emotions and negative emotions, which is compared with the sum of positive and negative emotions and multiplied by the total number of posts.The bullish index obtained is used as investor sentiment to express the relative bullish degree of investors on the stock market (Antweiler and Frank) [28].Based on this, in this study, we aim to employ crawler technology to mine the word-of-mouth comments from Autohome to obtain real information and use the measurement method of the bullish index to construct a consumer sentiment index and explore its impact on the NEV market.
In general, scholars have extensively researched the psychological factors influencing the purchase intentions of NEVs.However, there are still some limitations in these studies.First, the research on consumer psychology heavily relies on questionnaires, with limited empirical research on consumer psychological indicators based on big data text analysis.Additionally, only a few articles in the NEV market established relevant indices for extracted consumer sentiment and examined their mathematical impact on NEV sales.Furthermore, most studies analyzing the heterogeneity of NEV consumer groups were based on regionspecific questionnaire surveys.In contrast, this paper examines the characteristics of consumer groups across multiple provinces from a national perspective.Finally, while many studies have considered infrastructure construction to be a core influencing factor affecting NEV purchase intentions or sales, the actual impact of infrastructure construction on consumer behavior has not been well explored.Consequently, this paper employs partial linear functional coefficient dynamic panel data models to examine the changes in the association between consumer purchase intention and behavior caused by infrastructure construction.

Data Collection
The data source used in this study was obtained from consumer comments on Autohome (https://www.autohome.com.cn), a professional automobile BBS (Bulletin Board System).The word-of-mouth reviews on Autohome cover 11 attributes: satisfaction, dissatisfaction, space, driving experience, endurance, appearance, interior decoration, cost performance, intelligence, control, and comfort.For text analysis, in addition to these 11 attributes, the evaluator's geographic position and release time were also collected via a crawler program, and more than 260,000 comments in total were collected.
The specific crawling process is as follows.Firstly, obtain the URL of the post that needs crawling.Secondly, SeleniumLibrary based on Python crawls the comments of mobile users on the reputation of a car home.Finally, call Lxml and XPath to parse the crawled content and obtain the required data information; the data acquisition process is shown in Figure 1.

Text Analysis
Text analysis in this study aims to investigate the public's sentiment orientations toward NEVs.The concept of sentiment analysis, which involves extracting emotions or opinions from words, was initially proposed by Nasukawa and Yi [29].Presently, approaches to sentiment analysis can be grouped into rule-based methods and machine

Text Analysis
Text analysis in this study aims to investigate the public's sentiment orientations toward NEVs.The concept of sentiment analysis, which involves extracting emotions or opinions from words, was initially proposed by Nasukawa and Yi [29].Presently, approaches to sentiment analysis can be grouped into rule-based methods and machine learning (data-driven)-based methods.Machine learning has proven to be the most effective approach for analyzing "big data".Consequently, as shown in Figure 2, the overall text analysis process based on machine learning was adopted to explore consumers' sentiments toward NEVs.

Text Analysis
Text analysis in this study aims to investigate the public's sentiment orientations toward NEVs.The concept of sentiment analysis, which involves extracting emotions or opinions from words, was initially proposed by Nasukawa and Yi [29].Presently, approaches to sentiment analysis can be grouped into rule-based methods and machine learning (data-driven)-based methods.Machine learning has proven to be the most effective approach for analyzing "big data".Consequently, as shown in Figure 2, the overall text analysis process based on machine learning was adopted to explore consumers' sentiments toward NEVs.(

1) Data Preprocessing
To perform a thorough analysis of the textual data and obtain more accurate results, it was necessary to preprocess the word-of-mouth data collected from Autohome.The preprocessing procedures included data cleaning, Chinese word segmentation, and stop word removal.First, duplicate textual data and short comments with insufficient content were deleted.Second, the cleaned text underwent word segmentation.To enhance the accuracy of this process, relevant words in the automotive field were added to a userdefined dictionary based on information related to NEVs.The Jieba word segmentation tool was then utilized to segment the comment text.Third, the text was cleansed for stop words using the stop word list provided by the Harbin Institute of Technology.(

1) Data Preprocessing
To perform a thorough analysis of the textual data and obtain more accurate results, it was necessary to preprocess the word-of-mouth data collected from Autohome.The preprocessing procedures included data cleaning, Chinese word segmentation, and stop word removal.First, duplicate textual data and short comments with insufficient content were deleted.Second, the cleaned text underwent word segmentation.To enhance the accuracy of this process, relevant words in the automotive field were added to a userdefined dictionary based on information related to NEVs.The Jieba word segmentation tool was then utilized to segment the comment text.Third, the text was cleansed for stop words using the stop word list provided by the Harbin Institute of Technology.
(2) Text Vectorization After performing text cleaning, the comments from the most satisfied and least satisfied attributes of the 11 attributes were considered to be positive emotional comments and negative emotional comments, respectively.As the data used for subsequent model training, a total of 17,733 positive emotional comments and 17,565 negative emotional comments were obtained.To facilitate the training of the subsequent model, the comments of the above two attributes were converted into vectors, and we used the continuous bag of words (CBOW) to process the text.The CBOW function is calculated by using the following formula: where C denotes the corpus; W represents the central word; and "Context" represents the context of the central word.The CBOW model was used to predict the likelihood that the target would appear in the output.There are three distinct stages that generate a network: the input, projection, and output layers. (

3) Modeling Training
The vectors transformed by the CBOW model were then used to train various models, including Random Forest, AdaBoost, SVM, Logistic Regression, KNN, and Bayes.A portion of the data (80%) was allocated to the training set, and the remaining 20% was for testing.The test set data were used for prediction and analysis, and the performance of different models is presented in Table 1.It was observed that SVM performed the best among them, indicating its stronger generalization ability and better classification results when processing experimental datasets.Hence, this paper utilized the trained SVM emotion classifier to predict the emotions of unlabeled text.

(4) Text Classification
Based on the reviews of nine categories, including space, driving experience, endurance, appearance, interior decoration, cost performance, intelligence, control, and comfort, the SVM sentiment classifier was employed to predict the sentiment, and the sentiment tendency of each comment was obtained, along with the sentiment prediction values for comments across different attributes.
Figure 3 shows the emotional distributions of NEVs across nine attributes.These distribution results effectively represent consumers' overall satisfaction with NEVs.Notably, consumers expressed particularly positive feedback regarding driving experience, appearance, and cost performance, and negative comments were prevalent in the attributes of space, interior decoration, and comfort.Specifically, Figures 4 and 5 show the top ten words' frequencies in positive and negative comments, respectively.Consumers recognized the power, cost performance, and appearance of new energy vehicles and expressed strong dissatisfaction with the storage space, riding space, interior decoration, and noise of new energy vehicles.The exceptional driving experience was attributed to the swift motor response and abundant low-speed torque of NEVs.Additionally, NEVs exhibit superior driving performance compared to traditional fuel vehicles.The implementation of purchase subsidies and tax reduction policies has significantly reduced the cost of NEVs for consumers, further contributing to their positive perception of their cost performance.The scientific and technological design of NEVs, coupled with the use of advanced materials, has not only resolved energy-related issues but also attracted consumers through an appealing appearance.Conversely, some models sacrifice storage space to accommodate electrical components at the chassis and batteries at the bottom of the boot, leading to dissatisfaction among consumers regarding storage space.Consumers have also reported that the interior lacks texture and exhibits a strong sense of plasticity.Although NEVs do not generate engine noise, the low-frequency noise produced by the motor is an annoyance for some consumers.

Econometric Models (1) Baseline Model
To explain the impact mechanism of consumer sentiment on the NEV market at a mathematical level, this study employed a two-way fixed effects model and robust clustering to address heteroscedasticity in the model.The benchmark regression model was set as the following: where it NEV  The exceptional driving experience was attributed to the swift motor response and abundant low-speed torque of NEVs.Additionally, NEVs exhibit superior driving performance compared to traditional fuel vehicles.The implementation of purchase subsidies and tax reduction policies has significantly reduced the cost of NEVs for consumers, further contributing to their positive perception of their cost performance.The scientific and technological design of NEVs, coupled with the use of advanced materials, has not only resolved energy-related issues but also attracted consumers through an appealing appearance.Conversely, some models sacrifice storage space to accommodate electrical components at the chassis and batteries at the bottom of the boot, leading to dissatisfaction among consumers regarding storage space.Consumers have also reported that the interior lacks texture and exhibits a strong sense of plasticity.Although NEVs do not generate engine noise, the low-frequency noise produced by the motor is an annoyance for some consumers.

Econometric Models (1) Baseline Model
To explain the impact mechanism of consumer sentiment on the NEV market at a mathematical level, this study employed a two-way fixed effects model and robust clustering to address heteroscedasticity in the model.The benchmark regression model was set as the following: where NEV it is the logarithm of NEV sales; CS it is the core explanatory variable and the consumer sentiment index (see Section 3.4 for the calculation process); and X it is the control variable group.α represents the intercept term, β represents the coefficient parameter corresponding to the variable, u i represents the individual effect that captures the difference between various brands of electric vehicles, v t represents the time effect that captures the missing variable that does not change with individuals but changes with time, and ε it is the random disturbance term.
(2) Partial Linear Functional Coefficient Dynamic Panel Data Model Infrastructure development can mitigate consumers' concerns about mileage, enhance their intention to purchase, and drive the growth of NEV sales.Therefore, the differences caused by infrastructure may alter the association between consumer sentiment and NEV sales, implying the presence of heterogeneity.To measure this heterogeneity, this paper adopted partial linear functional coefficient dynamic panel data models to estimate the impact of consumer sentiment on NEV sales.This approach not only avoids biased estimation and modeling errors but also quantifies the heterogeneous impact of infrastructure development.Given that the use of a private charging station is exclusive, this study focused on the impact of public charging station when discussing the availability of new charging station.Specifically, we assumed that the response of consumer sentiment to NEV sales was a function of the popularity of a public charging station, denoted as β 1 = G(PCP it ).By adding β 1 into the benchmark model, partial linear functional coefficient dynamic panel data models were obtained as the following: where PCP it is the penetration rate of a public charging station.The availability of public charging points is defined as the average number of public charging points per 10,000 people in that year [30] while G(PCP it ) is an unknown function used to measure the impact of purchase intentions on the purchase gap.In this paper, we used the sequence estimation method to estimate the unknown function [31].The specific steps were as follows.
First, the variable coefficient function G(PCP it ) was estimated by the linear combination of sieve basis functions in Equation (4): where ′ is a p × 1 sequence of the basis function and was transformed into Equation ( 5), as the following: where v it = ε it + κ it and κ it = G(PCP it ) − h(PCP it ) ′ η denotes screening approximation error.Second, after the first-order differential elimination of the fixed effects of Equation (3), we obtained the following model: If all explanatory variables were exogenous, Equation ( 6) could be estimated by the least square method, as the following: where Finally, the functional coefficients G(PCP it ) could be estimated by the following formula: The consistency and asymptotic normality of the above estimators were verified under certain regular assumptions [32].

Variable Description and Data Introduction
(1) Explained Variable At present, NEVs are the focus of the country and also the development direction of manufacturers.This paper examined the logarithm of NEV sales as the independent variable (China's NEVs were mainly electric vehicles).In 2017, there were frequent adjustments to the promotional policy of NEVs, resulting in increased consumer attention and an expansion of the overall NEV market.As a result, this study focused on the sales volume of NEVs in various provinces from 2017 to 2022, with data primarily obtained from the China Passenger Car Association.Due to the lack of core explanatory variable data for certain provinces, samples from Hainan, Heilongjiang, Tibet, and Qinghai were excluded, leaving 27 provinces as the sample for the explanatory variable analysis.Figure 6 shows the sales of NEVs in China's Eastern China, Western China, Central China, and Northeastern China, and China's new sales have shown an upward trend in recent years.Specifically, the sales of NEVs in China in 2022 was mainly concentrated in the eastern region, accounting for 63% (see Figure 7).
(2) Core Explanatory Variable The core explanatory variable of this paper was the consumer sentiment index (CS), which was determined using text sentiment analysis techniques.To measure the sentiment of the nine attributes, the construction method of investor sentiment proposed by Antweiler and Frank [28] was employed, combined with the prediction results of the SVM model.For instance, to calculate the emotional score of the spatial dimension, we utilized the following formula: where CS refers to the emotional index of consumers' evaluation of space in province i at time t; M pos is the number of comments with positive attitudes in the spatial dimension; M neg is the number of comments with negative attitudes in the spatial dimension; and B is the degree of consumers' emotions for space, ranging from −1 to 1.The degree of emotion and the number of comments together constitute the spatial-level consumer emotion index CS.By following this process, emotional indices for all nine dimensions were calculated, and the data were weighted accordingly based on the number of comments.The comprehensive emotional index of consumer sentiment toward NEVs was subsequently determined.This comprehensive emotional index plays a vital role in understanding and analyzing the overall views and attitudes of consumers toward NEVs. Figure 8 shows the spatial distribution of the consumer sentiment index.Consumer sentiment in the central and coastal regions was relatively high, while that in the northern region was relatively low.

Variable Description and Data Introduction
(1) Explained Variable At present, NEVs are the focus of the country and also the development direction of manufacturers.This paper examined the logarithm of NEV sales as the independent variable (China's NEVs were mainly electric vehicles).In 2017, there were frequent adjustments to the promotional policy of NEVs, resulting in increased consumer attention and an expansion of the overall NEV market.As a result, this study focused on the sales volume of NEVs in various provinces from 2017 to 2022, with data primarily obtained from the China Passenger Car Association.Due to the lack of core explanatory variable data for certain provinces, samples from Hainan, Heilongjiang, Tibet, and Qinghai were excluded, leaving 27 provinces as the sample for the explanatory variable analysis.Figure 6 shows the sales of NEVs in China's Eastern China, Western China, Central China, and Northeastern China, and China's new sales have shown an upward trend in recent years.Specifically, the sales of NEVs in China in 2022 was mainly concentrated in the eastern region, accounting for 63% (see Figure 7).(2) Core Explanatory Variable The core explanatory variable of this paper was the consumer sentiment index (CS), which was determined using text sentiment analysis techniques.To measure the sentiment of the nine attributes, the construction method of investor sentiment proposed by Antweiler and Frank [28] was employed, combined with the prediction results of the SVM  (

3) Control Variable
Five control variables were considered as the following: the national policy, economic environment, market size, infrastructure construction, and price of NEVs [33][34][35][36].The national policy primarily considered the state subsidy (SB), which was indicated by the subsidy standard of 300-400 pure trams.The economic environment was measured by per capita disposable income (PCDI); market size was evaluated based on family size (FS).Infrastructure construction was assessed by the availability of cumulative charging points (CCPs), with the ratio of the amount of cumulative infrastructure to the regional population serving as an indicator of regional cumulative charging point availability.As the No. 1 brand in domestic sales, BYD's sales price was representative to a certain extent.So, the price of NEVs (NP) was determined by the average annual sales price of BYD's Song MAX.
To sum up, Table 2 presents the descriptive statistics of each variable.Logarithmic processing was performed for both the explanatory variable and the control variables.From the values of the mean and variance, it can be seen that the sales gap of new energy vehicles (NEVs) and consumer sentiment (CS) fluctuations in various provinces was not large, and the penetration rate of charging stations (CPCs) varied greatly between different provinces.The data were collected from sources such as the China Statistical Yearbook, China Charging Alliance, China Government Network, China Passenger Car Association, and Autohome.(3) Control Variable Five control variables were considered as the following: the national policy, economic environment, market size, infrastructure construction, and price of NEVs [33][34][35][36].The national policy primarily considered the state subsidy (SB), which was indicated by the subsidy standard of 300-400 pure trams.The economic environment was measured by per capita disposable income (PCDI); market size was evaluated based on family size (FS).Infrastructure construction was assessed by the availability of cumulative charging points (CCPs), with the ratio of the amount of cumulative infrastructure to the regional population serving as an indicator of regional cumulative charging point availability.As the No. 1 brand in domestic sales, BYD's sales price was representative to a certain extent.So, the price of NEVs (NP) was determined by the average annual sales price of BYD's Song MAX.
To sum up, Table 2 presents the descriptive statistics of each variable.Logarithmic processing was performed for both the explanatory variable and the control variables.From the values of the mean and variance, it can be seen that the sales gap of new energy vehicles (NEVs) and consumer sentiment (CS) fluctuations in various provinces was not large, and the penetration rate of charging stations (CPCs) varied greatly between different provinces.The data were collected from sources such as the China Statistical Yearbook, China Charging Alliance, China Government Network, China Passenger Car Association, and Autohome.

Benchmark Regression Results
Following the data selection process, empirical studies were conducted using the ordinary least square (OLS), fixed effects, and two-way fixed effects models to examine the influence of consumer sentiment on the sales of NEVs.According to model (3) in Table 3, consumer sentiment had a positive effect on the sales of NEVs (β = 0.196, p < 0.01).Generally, as positive information about the NEV market continues to increase, it may lead to herd behavior among potential consumers, continuously attracting new buyers.Conversely, an influx of negative information in the NEV market could restrain the purchase intentions of potential consumers, resulting in a decrease in sales.In order to explore the heterogeneous effect of nine attributes, we established an Autoregressive Distributed Lag Model (ARDL Model) to identify the long-term and shortterm impacts of consumer sentiment (CS) on sales of NEVs from nine attributes, such as space and driving experience.The results in Table A1 show that the space, driving experience, and intelligent consumer sentiment of new energy vehicles are the long-term influencing factors affecting the development of NEVs.

Robustness Check
To validate the reliability of the impact of consumer sentiment on the diffusion effect of NEVs, we conducted several robustness tests in this study.
(1) Replace the core explanatory variable.To further determine whether positive comments enhanced NEV sales, the proportion of positive comments in consumer evaluations was used as a proxy variable for the consumer sentiment index in the benchmark regression model.This analysis investigated the influence of positive comments on NEV sales.The regression results are displayed in model (1) of Table 4, affirming that positive evaluations (PCs) significantly stimulated NEV sales.These findings further support the regression results of the benchmark model, thus confirming the robustness of the analysis and the reliability of the conclusion.(2) Modify the approach used to estimate the core explanatory variables.Due to the potential classification errors that different emotional analysis methods may produce, it is important to ensure the accuracy of the estimation results.To address this issue, this paper adopted the emotional dictionary method to assess consumer sentiment and incorporate the findings into the benchmark model to assess its impact on NEV sales.The specific outcomes can be found in model ( 2) of Table 4.The findings suggest that consumer sentiment (CS_1) calculated by the sentiment dictionary method still had a positive influence on NEV sales, with the coefficient displaying similar significance and magnitude as the results of the benchmark regression model.This further validates the precision of the benchmark model's regression findings, thereby solidifying the reliability of the analysis and its conclusions.
(3) Adjust the estimation model.As the static panel model neglected the dynamic influence of lag periods on the explained variable in the current period, its estimation results may have been more biased than those of the dynamic panel model.Given the evident inertia effect observed in the consumption of NEVs, overcoming the limitations of the static panel model was crucial.To achieve this, the dynamic panel Gaussian Mixture Model (GMM) was introduced to test the robustness of the conclusions.Specifically, the lag term of NEV sales was added to the benchmark model.The estimated results are shown in model (3) of Table 4.The dynamic panel GMM model successfully passed both the autocorrelation and overidentification tests.This finding implies that the analysis of the impact trend of consumer sentiment on NEV sales possessed strong robustness, thus enhancing the reliability of the conclusions.

Heterogeneity Analysis
The previous analysis indicated that consumer sentiment plays a crucial role in driving the sales of NEVs.Potential buyers of these vehicles often rely on feedback and evaluations posted online by existing owners to inform their purchasing decisions.Therefore, enhancing consumer sentiment can significantly influence the intentions of potential buyers and consequently boost the NEV market.However, it was essential to further explore the specific characteristics of potential buyers to effectively leverage consumer sentiment to promote market development.To gain a deeper understanding of the association between consumer sentiment and the development of NEVs, this study aimed to identify which groups were more susceptible to influence from consumer sentiment.The objective was to analyze the variations in purchase intentions among different consumer groups and to comprehensively comprehend how consumer sentiment drove the development of NEVs.
In a study conducted by Hidrue et al. [37], it was revealed that early adopters of electric vehicles were generally young or middle-aged, possessed a bachelor's degree or higher, and their income was not a significant determining factor.In contrast, research by Plötz [16] suggested that early adopters of electric vehicles typically had incomes greater than average and were predominantly male.However, other scholarly research has shown that gender is not a significant influencing factor in the purchase intention of NEVs [38].It was necessary to provide further evidence to determine whether gender and income impact the purchase intentions of NEVs.To address this issue, this study tested whether income and gender had a significant impact on purchase intentions through personal income and the proportion of males and reaffirmed the previous findings by examining the influence of education level and the proportion of young and middle-aged people.This investigation helped uncover any differences in consumer sentiment and the promotion of NEVs.Consumer groups were divided into high-level and low-level groups based on national averages for education level (EL), personal income (PI), male proportion (MP), and the proportion of young and middle-aged people (YM).
Table 5 presents the results of the regression analysis.Among the young and middleaged consumer groups with a high education, high personal income, and strong willingness to purchase NEVs, no significant gender differences were found.In the current Chinese market for NEVs, their prices are generally higher than those of traditional fuel vehicles.Furthermore, recent announcements from major automobile companies have highlighted issues such as chip shortages, declining national new energy policies, and rising prices of raw materials for power batteries.These factors have contributed to an overall increase in the price of NEVs.Consequently, considering the price factor, consumers with lower incomes tend to be less inclined to purchase NEVs.On the other hand, with increasing economic independence and autonomy, women are demonstrating greater purchasing power and demand for cars.The 21st Century Business Herald, the 21st Century New Automobile Research Institute, and Nielsen conducted a survey among women using data from the "2021 China Automobile Consumption Trend Survey" and compared it with a male sample.The results indicated that female users displayed a preference for NEVs, exhibited greater brand recognition for Chinese brands, were more open to online vehicle purchases, and had greater financial penetration.As the market share of female consumers continues to increase, gender differences no longer significantly affect purchase intentions.

Moderating Effect
Furthermore, this study analyzed the moderating effect of a per capita public charging station.As mentioned earlier, the development of charging infrastructure helps alleviate consumers' "mileage anxiety" regarding limited driving range.Considering the diminishing marginal utility, it is reasonable to assume that the psychological effect of charging infrastructure construction on consumers may be dynamic.To investigate this issue, a partial linear functional coefficient dynamic panel data model was used to estimate the moderating effect of a charging station on the association between consumers' purchase intentions and NEV purchase behavior.
The empirical regression results consist of two main components.The first part includes the linear regression findings of the control variables presented in Table 6, while the second part involves the estimation results of the nonparametric aspects depicted in Figure 9. minishing marginal utility, it is reasonable to assume that the psychological effect of charging infrastructure construction on consumers may be dynamic.To investigate this issue, a partial linear functional coefficient dynamic panel data model was used to estimate the moderating effect of a charging station on the association between consumers' purchase intentions and NEV purchase behavior.
The empirical regression results consist of two main components.The first part includes the linear regression findings of the control variables presented in Table 6, while the second part involves the estimation results of the nonparametric aspects depicted in Figure 9.  Table 6 reveals that the coefficients of per capita disposable income (PCDI) and governmental subsidies (SBs) were both statistically significant at the 1% level, indicating a positive impact on consumers' propensity to purchase NEVs.This aligns with our expectations.Additionally, Figure 9 illustrates that the marginal influence of consumer sentiment on the sale of NEVs was contingent upon the extent of charging pile coverage (CPC).Once the prevalence of charging stations reaches a certain threshold, their increased availability effectively mitigates consumers' concerns regarding "mileage anxiety".Table 6 reveals that the coefficients of per capita disposable income (PCDI) and governmental subsidies (SBs) were both statistically significant at the 1% level, indicating a positive impact on consumers' propensity to purchase NEVs.This aligns with our expectations.Additionally, Figure 9 illustrates that the marginal influence of consumer sentiment on the sale of NEVs was contingent upon the extent of charging pile coverage (CPC).Once the prevalence of charging stations reaches a certain threshold, their increased availability effectively mitigates consumers' concerns regarding "mileage anxiety".
This research underscores the significance of well-established supporting infrastructure as a crucial prerequisite for the adoption of NEVs.Adequate charging pile popularity is essential in stimulating consumers' purchasing inclination and converting their intentions into tangible automobile consumption.

Discussion
Discussions of research perspective and research methods were carried out herein, and practical implications of the research results are given.

Comparison of Research Perspectives
The existing research mainly considered the brand of new energy vehicles and established the public attribute perception of new energy vehicles.For example, Yang [39] found that the comfort, handling, cost, and spatial attribute perception of electric vehicles had important impacts on the sales of electric vehicles.In terms of the development of NEVs, automobile brands have an impact, but the regional differentiated policies should be more worthy of consideration.Therefore, this paper established a dual fixed effects model, in which individual fixed effects capture region-generated differences such as the heterogeneity of policies, temperatures, and so on.In addition, we further established a comprehensive consumer sentiment index based on the public perception attribute to further verify the impact of the emotional factors indicated in public comments on the sales of NEVs.

Comparison of Research Methods
In the research on the influencing factors of NEVs, the charging pile was also the focus of scholars' attention.Due to the immaturity of battery technology for new energy vehicles, consumers will have "mileage anxiety".Scholars used the questionnaire survey method and found that mileage anxiety will hinder consumers' purchase behavior to a certain extent.In addition, many scholars have studied the linear impact of the popularity of charging piles on the sales of NEVs.The popularization of charging stations can be seen as a policy tool to promote the development of NEVs, and any policy has two sides, with its effectiveness also decreasing with the promotion of policies.However, the vast majority of existing research has not studied the nonlinear effects of charging piles.In this paper, the penetration rate of charging piles, consumer sentiment, and sales of NEVs were put into a model.Considering diminishing marginal utility, we had reason to assume that the psychological impact of charging infrastructure construction on consumers may have been dynamic.Therefore, a partial linear functional coefficient dynamic panel data model was used to estimate the moderating effect of a charging station on the association between consumers' purchase intentions and NEV purchase behavior.

Practical Implications
Our findings have several important practical implications.First, consumer sentiment can influence NEV sales.Therefore, automobile manufacturers are suggested to follow consumers' feedback on NEVs, analyze their concerns by browsing their comments, and thus improve their NEV services so as to enhance consumers' purchase intentions and further fill the gap between purchase intention and behavior.Based on the findings of this study, consumers are dissatisfied with the design of NVEs in terms of their space, interior decoration, and comfort.Therefore, in future development, NEV manufacturers should endeavor to satisfy the needs of consumers for product appearance design and performance requirements such as the insufficient range of electric vehicles.
Second, in terms of sellers, they can promote consumption by diversifying sales channels and developing unique sales plans.This study showed that the young consumer groups with a high education level had a higher intention to buy NEVs, so a differentiated customized marketing scheme for the target group can improve publicity efficiency.It can inform a sales strategy based on online sales and supplemented by offline sales to improve the popularity and recognition of automotive products.It should also explore innovative methods, such as using social platforms and short videos for promotion.
Third, reasonable infrastructure construction can promote consumers' purchase intention and then affect consumers' purchase behavior.At present, China's infrastructure development is unbalanced.For most provinces and cities, the government should prioritize infrastructure construction and optimize the charging environment of NEVs.For provinces and cities with abundant charging piles such as Beijing and Shanghai, we should focus on making the facilities more user-friendly by providing a pleasant charging experience and improving the efficiency of distribution and utilization, as well as avoiding duplication and disorderly construction.On the basis of data analysis and scientific evaluation, the government should strengthen information sharing and effective scheduling, realize the balanced distribution and efficient utilization of charging stations, and support the healthy development of the NEV industry.
On the basis of existing research, this article constructed a consumer sentiment index based on consumer comments on the various attributes of NEVs and to study the impact of consumer sentiment on their sales in the NEV market.Furthermore, we also investigated the nonlinear moderating effect of the popularity of charging piles on the relationship between consumer sentiment and NEV sales, which was a supplement to existing research.In further research, we plan to refine the categories of NEVs and enrich the construction methods of consumer sentiment.In addition to considering consumer comments, the inherent attributes of NEVs can also be incorporated into the construction methods of consumer sentiment.

Conclusions
The promotion of NEVs serves as a pivotal measure for reducing carbon emissions within the domain of road transportation.Drawing upon consumer comments, as well as regression results derived from the two-way fixed effects model, robustness tests, heterogeneity tests, and the partial linear functional coefficient panel data model, we arrived at the following conclusions: (1) Consumer comments predominantly highlight positive experiences in driving, appearance, and cost performance.Negative feedback mainly revolves around space limitations, interior design quality, and noise levels, particularly related to inadequate storage, a plastic-like decor impression, and noise.(2) The consumer sentiment index significantly influences NEV sales.Higher consumer sentiment strengthens potential buyers' willingness to make purchases, thus bolstering sales.(3) Individuals' purchasing intentions exhibit varied effects on their actual purchasing behavior based on factors such as education level, personal income, and age.(4) A higher per capita public charging pile availability assuages consumers' concerns over a limited driving range, consequently promoting their purchasing behavior.
Our study has several limitations.First, starting from the consumer's perspective, this paper explored the impact of consumer sentiment on the diffusion of the new energy vehicle market.It only considered the significance of the carbon emission reduction brought by the replacement of NEVs on the consumer side, while ignoring the environmental damage caused by the extraction of technical equipment and primary processing in the supply chain of the NEV industry chain.In future research, we will try to study the full life cycle of new energy vehicles to explore the dynamic impact between environmental damage and environmental benefits.Second, this study did not further analyze the carbon emission reduction effect of the replacement of NEVs generated by consumption on the consumer side.In future research, we will also carry out more in-depth discussion and study the environmental benefits.Finally, in future studies, we also plan to consider the factors affecting the energy loss rate of charging stations and their impact on consumers' purchasing behavior, so as to enrich the research results of NEVs.

Figure 6 .
Figure 6.Trends of NEV sales in four economic development regions.

Figure 8 .
Figure 8. Spatial distribution of consumer sentiment indices.(a) Consumer sentiment index in 2017; (b) Consumer sentiment index in 2019; (c) Consumer sentiment index in 2021; (d) Consumer sentiment index in 2022.

Figure 8 .
Figure 8. Spatial distribution of consumer sentiment indices.(a) Consumer sentiment index in 2017; (b) Consumer sentiment index in 2019; (c) Consumer sentiment index in 2021; (d) Consumer sentiment index in 2022.

Figure 9 .
Figure 9. Nonlinear estimation results of the partial linear functional coefficient dynamic panel data model.

Figure 9 .
Figure 9. Nonlinear estimation results of the partial linear functional coefficient dynamic panel data model.

Table 1 .
Performance of machine learning models.

Table 3 .
Benchmark model regression results.

Table 4 .
Regression results of the robustness test.

Table 5 .
Regression results of heterogeneity analysis.

Table 6 .
Linear estimation results of the partial linear functional coefficient dynamic panel data model.

Table 6 .
Linear estimation results of the partial linear functional coefficient dynamic panel data model.Standard errors in parentheses, *** p < 0.01.