Emotion-based vehicle sales prediction method

Abstract

The invention relates to the technical field of sales volume prediction, and discloses an emotion-based vehicle sales volume prediction method, which comprises the steps of collecting vehicle forum comments and vehicle official website news monthly, performing data cleaning, word segmentation and feature conversion, and calculating average emotion scores (a first average emotion score and a second average emotion score) by using an emotion dictionary. And sales volume data acquisition: acquiring historical sales volume data, calculating a monthly year-on-year growth rate, a monthly year-on-year growth rate and a sales volume growth trend, and performing first-order and second-order derivation on the growth trend. And feature construction: taking months as features, including a first average emotion score, a second average emotion score, a year-on-year growth rate, a ring-on-year growth rate, a first derivative, a second derivative and a sales volume. And model training: training the model by using sales volume data in past four years, aggregating features by year and month, and taking a root mean square error as an evaluation standard. Model selection: predicting the sales volume of the fourth year according to the sales volume data of the first three years, calculating the RMSE of the three models, and selecting the model with the minimum RMSE per month as a prediction model.

Description

Technical Field

[0001] The present invention relates to the technical field of sales forecasting, and in particular to a vehicle sales forecasting method based on emotion. Background Art

[0002] With the popularization of the Internet and the rise of social media, consumers are increasingly inclined to refer to other users' comments and opinions before making purchase decisions. The automotive industry is no exception. Potential car buyers often learn about the performance, reputation and experience of other users of vehicles through forums and official websites. These online comments and discussions contain a lot of emotional information, which can reflect consumers' preferences and attitudes towards specific car series, and this emotional information may be closely related to the sales of vehicles.

[0003] Traditional sales forecasting methods usually rely on historical sales data, market trends, macroeconomic indicators, etc., but these methods often ignore the possible impact of consumer sentiment on sales. With the development of natural language processing technology, sentiment analysis has become an effective tool for extracting sentiment information from unstructured text data. Applying sentiment analysis to vehicle sales forecasting can more accurately capture changes in market dynamics and consumer sentiment, thereby improving the accuracy of sales forecasts.

[0004] Moreover, how car companies predict sales is a pain point. Different sales can formulate different marketing strategies. Usually, car companies will make linear predictions based on data from previous years, but the results are inaccurate and ignore the sales trend problems that may be caused by user emotions. Summary of the invention

[0005] In view of this, the present invention proposes a vehicle sales prediction method based on emotion, aiming to solve the problem in the current technology that the traditional method ignores the possible impact of consumer emotions on sales.

[0006] The present invention proposes a vehicle sales prediction method based on emotion, comprising:

[0007] Collect historical comments in automobile forums and historical comments on news releases on official websites of car series by month, perform data cleaning, word segmentation and feature conversion on the historical comments in the forums and historical comments on news releases, calculate the average sentiment score through the sentiment dictionary, and obtain the first average sentiment score and the second average sentiment score;

[0008] Collect historical sales data, obtain the year-on-year growth rate, month-on-month growth rate and growth trend of each month, perform first-order and second-order derivatives on the growth trend to obtain the first-order derivative and the second-order derivative; wherein the growth trend is the difference between the sales volume of this month and the sales volume of last month;

[0009] Form features for each month, the features including: a first average sentiment score, a second average sentiment score, a year-on-year growth rate, a month-on-month growth rate, a first-order derivative, a second-order derivative, and sales volume;

[0010] The sales data of the past four years are used for training. All the features are aggregated by year and month, and the root mean square error is used as the evaluation criterion to establish a sales forecasting model.

[0011] Predict the sales data of the fourth year based on the sales data of the same month in the previous three years, calculate the root mean square error of the three models, and select the sales forecast model with the smallest root mean square error in each month as the forecast model for that month;

[0012] Substitute the sales data of any month in the previous three years to obtain the predicted sales data of the same month in the current year.

[0013] Furthermore, the method for collecting forum historical comments and news release historical comments includes:

[0014] Determine the URLs of the automobile forum and the official website of the automobile series;

[0015] Determine whether the automobile forum and the official website of the car series can obtain API. If the judgment result is yes, obtain the API and collect comment data. If the judgment result is no, use a web crawler tool to crawl the page content and extract comment data.

[0016] Furthermore, the data cleaning includes:

[0017] Remove special characters, punctuation marks and non-text content from the comment data; remove stop words according to a stop word list; remove duplicate records and text content from each of the comment data; and use a spell checker to correct spelling errors in the text.

[0018] Furthermore, the word segmentation processing includes:

[0019] A natural language processing library is used to perform Chinese word segmentation and a six-category sentiment dictionary is established. The six-category sentiment dictionary includes sentiment categories and corresponding sentiment words. The sentiment categories include joy, sadness, anger, fear, surprise and disgust.

[0020] Furthermore, the feature conversion includes:

[0021] The frequency of occurrence of each sentiment category and the corresponding sentiment words in all comments is counted to obtain word frequency data.

[0022] Furthermore, the average sentiment score is calculated through the sentiment dictionary, including:

[0023] Initialize the score of each sentiment category to 0;

[0024] Divide the review data by month, and calculate the average sentiment score of the review data for each month, including the first average sentiment score and the second average sentiment score, where:

[0025] The first average sentiment score is:

[0026]

[0027] The second average sentiment score is:

[0028]

[0029] Among them, si is the corresponding six categories of sentiment scores, ni represents the corresponding number of entries, the first average sentiment score corresponds to the historical comments in the automobile forum, and the second average sentiment score corresponds to the historical comments on the news release of the official website of the car series.

[0030] Furthermore, the collecting of historical sales data specifically includes:

[0031] Count the historical sales data t and calculate the year-on-year growth rate r1 and month-on-month growth rate r2 for each month, where:

[0032] r1 = (this month's sales volume - last year's same month's sales volume) / last year's same month's sales volume × 100%;

[0033] r2 = (sales volume this month - sales volume in the same month last year) / sales volume last month × 100%.

[0034] Furthermore, the growth trend of sales volume is calculated, growth trend f = this month's sales volume - last month's sales volume, and the first-order derivative f1 and the second-order derivative f2 of f are calculated respectively.

[0035] Furthermore, the establishment of the sales forecasting model includes respectively using polynomial regression, decision tree regression and XGBoost regression methods.

[0036] Furthermore, the features include: a first average sentiment score, a second average sentiment score, a year-on-year growth rate, a month-on-month growth rate, a first-order derivative, a second-order derivative, and sales volume.

[0037] Compared with the prior art, the beneficial effect of the present invention is that the present invention discloses a vehicle sales forecasting method based on emotion, which collects and analyzes historical comments in automobile forums and historical comments on news releases on the official website of the vehicle series, calculates the average emotion score in combination with the emotion dictionary, takes these emotion scores as one of the features, and combines historical sales data to establish a sales forecasting model. Traditional sales forecasting methods may not be able to capture subtle changes in consumer emotions, but the present invention can more comprehensively consider the factors affecting sales by introducing emotion analysis, thereby improving the accuracy of the forecast. Forum comments and news releases can reflect the market's views and emotions on vehicles in real time. The present invention can capture changes in market emotions in a timely manner by collecting these data in real time, and provide the latest information for sales forecasting. By combining multiple features, such as emotion scores, year-on-year growth rates, month-on-month growth rates, etc., the model of the present invention can better adapt to changes in different market environments and sales trends. The present invention uses the root mean square error as an evaluation criterion, selects the model with the smallest error for prediction, thereby reducing the prediction error and improving the reliability of the prediction. By analyzing consumer emotions, automobile manufacturers and sellers can better understand market demand and consumer preferences, thereby formulating more effective marketing strategies and product improvement plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0039] Figure 1 This is a flow chart of a vehicle sales prediction method based on emotion according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features described in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0041] like Figure 1 As shown, an embodiment of the present invention provides a vehicle sales prediction method based on emotion, which includes:

[0042] Collect historical comments in automobile forums and historical comments on news releases on official websites of car series by month, perform data cleaning, word segmentation and feature conversion on the historical comments in the forums and historical comments on news releases, calculate the average sentiment score through the sentiment dictionary, and obtain the first average sentiment score and the second average sentiment score;

[0043] Collect historical sales data, obtain the year-on-year growth rate, month-on-month growth rate and growth trend of each month, perform first-order and second-order derivatives on the growth trend to obtain the first-order derivative and the second-order derivative; wherein the growth trend is the difference between the sales volume of this month and the sales volume of last month;

[0044] Form features for each month, the features including: a first average sentiment score, a second average sentiment score, a year-on-year growth rate, a month-on-month growth rate, a first-order derivative, a second-order derivative, and sales volume;

[0045] The sales data of the past four years are used for training. All the features are aggregated by year and month, and the root mean square error is used as the evaluation criterion to establish a sales forecasting model.

[0046] Predict the sales data of the fourth year based on the sales data of the same month in the previous three years, calculate the root mean square error of the three models, and select the sales forecast model with the smallest root mean square error in each month as the forecast model for that month;

[0047] Substitute the sales data of any month in the previous three years to obtain the predicted sales data of the same month in the current year.

[0048] It is understandable that the method involves a car sales prediction model, which uses data analysis and machine learning technology to predict future sales by combining historical forum comments, historical news release comments on the official website of the car series, and historical sales data.

[0049] This includes collecting historical comments in the car forum and historical comments on news releases on the official website of the car series by month, and collecting historical sales data, including the year-on-year growth rate and month-on-month growth rate for each month. Data cleaning, word segmentation and feature conversion are performed on historical comments on the forum and news releases. These steps help to extract useful information from unstructured text data. The average sentiment score is calculated using the sentiment dictionary to obtain the first average sentiment score and the second average sentiment score. Sentiment analysis helps to understand consumers' overall sentiment tendencies towards car brands. The difference between monthly sales and previous month sales is calculated to obtain the growth trend. The first-order and second-order derivatives of the growth trend are taken to obtain the first-order and second-order derivatives, which reflect the rate of change in sales. The data obtained in the above steps are integrated into the features of each month, including sentiment score, growth rate, first-order derivative, second-order derivative and sales. The sales data of the past four years are used to train the model, and all features are aggregated by year and month to establish a sales prediction model. The root mean square error is used as an evaluation criterion to assess the prediction accuracy of the model. Based on the sales data of the same month in the previous three years, the sales data of the fourth year are predicted, and the root mean square error of the three models is calculated. The model with the smallest root mean square error in each month is selected as the prediction model for that month. Substitute the sales data of any month in the previous three years to obtain the predicted sales data of the same month in the current year.

[0050] In this way, the method is able to leverage historical data and text analytics to predict future car sales, providing valuable information to automakers and market analysts to help them make more informed decisions.

[0051] In this preferred embodiment, the method for collecting forum historical comments and news release historical comments includes:

[0052] Determine the URLs of the automobile forum and the official website of the automobile series;

[0053] Determine whether the automobile forum and the official website of the car series can obtain API. If the judgment result is yes, obtain the API and collect comment data. If the judgment result is no, use a web crawler tool to crawl the page content and extract comment data.

[0054] It is understandable that the method for collecting the forum historical comments and news release historical comments includes the following steps: First, it is necessary to determine the URLs of the automotive forums and the official websites of vehicle series, which is the basis for data collection. Check whether these websites provide API interfaces that allow external systems to obtain data programmatically. If the website provides an API interface, use this interface to obtain comment data. This method is usually more efficient and complies with the website's data usage regulations. If the website does not provide an API interface, or the use of the API interface is restricted, then web crawler tools need to be used to scrape the page content. These tools can simulate the behavior of a browser, automatically browse the website, and extract the required comment data.

[0055] Through this method, the collection method can select a suitable data collection method according to the specific conditions of the automotive forums and the official websites of vehicle series. If the website provides an API interface, give priority to using the API for data collection; if there is no API interface or the use of the API interface is restricted, use web crawler tools for data collection. This flexible data collection method helps to ensure that the required data can be obtained from various sources.

[0056] In this preferred embodiment, the data cleaning includes:

[0057] Removing special characters, punctuation marks, and non-text content from the comment data; removing stop words according to the stop word list; removing duplicate records and text content in each comment data; using a spelling check tool to correct spelling mistakes in the text.

[0058] It is understandable that the data cleaning process is an important part of data preprocessing. Its purpose is to improve the accuracy and efficiency of subsequent analysis, remove noise and useless information, and make the data more suitable for further processing and analysis. In the comment data, there may be special characters, punctuation marks, and non-text content (such as pictures, videos, etc.), which are useless for text analysis and therefore need to be removed. Stop words refer to words that often appear in the text but are meaningless for analysis, such as "de", "he", "shi", etc. Using a stop word list can automatically identify and remove these words to reduce interference with text analysis. In the collected comment data, there may be duplicate records or text content, and these duplicate data are of no value for analysis and therefore need to be deleted to avoid bias in the analysis results. Spelling mistakes may affect the results of text analysis, so a spelling check tool is used to correct these mistakes to ensure the accuracy of the text data.

[0059] Through these data cleaning steps, the method can improve the data quality and provide a more accurate data basis for subsequent text analysis, sentiment analysis, etc. This data cleaning method is a key step to ensure the accuracy and effectiveness of data analysis results.

[0060] In this preferred embodiment, the word segmentation process includes:

[0061] A natural language processing library is used to perform Chinese word segmentation and a six-category sentiment dictionary is established. The six-category sentiment dictionary includes sentiment categories and corresponding sentiment words. The sentiment categories include joy, sadness, anger, fear, surprise and disgust.

[0062] It is understandable that the method involves using natural language processing (NLP) technology to segment Chinese text and establish a six-category sentiment dictionary. Segmentation is a basic step in Chinese text processing, which divides a continuous sequence of Chinese characters into vocabulary units with independent meanings. The sentiment dictionary is a database containing sentiment words and their sentiment categories (such as joy, sadness, anger, etc.) for analyzing the sentiment tendency of the text. Select an NLP library suitable for Chinese processing, such as jieba, HanLP, etc., to perform segmentation on the collected Chinese comment data. The result of the segmentation splits the text into words or phrases, providing a basis for sentiment analysis. Create a sentiment dictionary containing six sentiment categories, namely joy, sadness, anger, fear, surprise and disgust. Each sentiment category contains words related to the emotion, which can be extracted from an existing sentiment dictionary, or obtained by manual annotation or training based on an existing data set. Use the segmented text and the sentiment dictionary for sentiment analysis. The analysis process may include calculating the sentiment value of each word and inferring the sentiment tendency of the entire text based on the frequency of occurrence and sentiment value of the word in the text.

[0063] Through this method, the method can use NLP technology to segment Chinese review data and combine it with a sentiment dictionary to determine the sentiment tendency of the text. This sentiment analysis helps understand consumers' emotional reactions to car brands or models, thereby providing valuable market insights for automakers.

[0064] In this preferred embodiment, the feature conversion includes:

[0065] The frequency of occurrence of each sentiment category and the corresponding sentiment words in all comments is counted to obtain word frequency data.

[0066] It is understandable that the feature conversion involves statistical analysis of the words in the sentiment dictionary to obtain the frequency of occurrence of each sentiment category and its corresponding sentiment words in all comments. These frequency data can be used as features of sentiment analysis for subsequent model training and prediction. First, all comment data are segmented using natural language processing technology to split the text into words or phrases. The segmented words are matched with the words in the sentiment dictionary to determine the sentiment category corresponding to each word. The number of times each sentiment category and its corresponding sentiment words appear in all comments is counted to obtain the word frequency data of each sentiment category and its words. The word frequency data reflects the importance of each sentiment category and its words in the comment data, so it can be used as a feature for subsequent model training and prediction.

[0067] Through this method, the feature conversion step can extract key features in the review data from the perspective of the sentiment dictionary, which helps the model better understand the sentiment information in the text, thereby improving the accuracy of sentiment analysis and the reliability of prediction.

[0068] In this preferred embodiment, the average sentiment score is calculated by the sentiment dictionary, including:

[0069] Initialize the score of each sentiment category to 0;

[0070] Divide the review data by month, and calculate the average sentiment score of the review data for each month, including the first average sentiment score and the second average sentiment score, where:

[0071] The first average sentiment score is:

[0072]

[0073] The second average sentiment score is:

[0074]

[0075] Among them, si is the corresponding six categories of sentiment scores, ni represents the corresponding number of entries, the first average sentiment score corresponds to the historical comments in the automobile forum, and the second average sentiment score corresponds to the historical comments on the news release of the official website of the car series.

[0076] In this preferred embodiment, the collecting of historical sales data specifically includes:

[0077] Count the historical sales data t and calculate the year-on-year growth rate r1 and month-on-month growth rate r2 for each month, where:

[0078] r1 = (this month's sales volume - last year's same month's sales volume) / last year's same month's sales volume × 100%;

[0079] r2 = (sales volume this month - sales volume in the same month last year) / sales volume last month × 100%.

[0080] Calculate the growth trend of sales, growth trend f = this month's sales - last month's sales, and calculate the first-order derivative f1 and the second-order derivative f2 of f respectively.

[0081] It is understood that the method involves statistical analysis of historical sales data to calculate the year-on-year growth rate and month-on-month growth rate for each month, as well as the growth trend of sales. These statistics are important inputs for building a sales forecasting model and help understand the trend and speed of sales changes. Statistics of historical sales data t: Collect sales data for each month to form a time series containing sales values ​​for each month. Calculate the year-on-year growth rate r1: The year-on-year growth rate is an important indicator to measure the sales growth rate, which indicates the percentage increase in sales compared with the same period last year. The calculation formula is: r1 = (sales this month - sales in the same month last year) / sales in the same month last year × 100%. Calculate the month-on-month growth rate r2: The month-on-month growth rate is another important indicator to measure the sales growth rate, which indicates the percentage increase in sales compared with the previous month. The calculation formula is: r2 = (sales this month - sales in the same month last year) / sales last month × 100%. Calculate the growth trend f of sales: The growth trend is another important indicator to measure the change in sales, which indicates the change in sales from one month to the next. The calculation formula is: f = sales this month - sales last month. Find the first-order derivative f1 and the second-order derivative f2: The first-order derivative and the second-order derivative are concepts in mathematics, which are used to measure the instantaneous rate of change of a function at a certain point and the instantaneous rate of change of the instantaneous rate of change. Here, they are used to measure the speed of change of the sales growth trend and the speed of change of the speed of change. The calculation formula is: f1 = df / dt, f2 = d 2f / dt 2 .

[0082] Through these steps, the method can extract useful statistical information from historical sales data, including year-on-year growth rate, month-on-month growth rate, growth trend and its derivative. These data are essential for building and training sales forecasting models, which are helpful for predicting future sales trends.

[0083] In this preferred embodiment, the establishment of the sales forecasting model includes using polynomial regression, decision tree regression and XGBoost regression methods respectively.

[0084] It is understandable that the method for establishing the sales forecast model includes three different regression analysis methods, namely polynomial regression, decision tree regression and XGBoost regression. These methods are commonly used prediction analysis techniques in statistics and machine learning, and they each have different characteristics and applicable scenarios.

[0085] Specifically:

[0086] Polynomial regression is an extension of linear regression that builds a nonlinear model by introducing additional independent variables (i.e., the power of the independent variable). This method is suitable for data with nonlinear relationships and can capture the curve trend in the data. Decision tree is a simple nonparametric regression method that builds a tree structure through a series of decision rules. Each node represents a test of a feature and each branch represents a certain value range of a feature. This method is easy to interpret and can capture nonlinear relationships in the data, but it may be too dependent on the order of features. XGBoost is an ensemble learning method based on decision trees that improves the predictive ability of the model by optimizing the objective function. XGBoost can automatically select the best feature combination and can handle high-dimensional data and missing values. It is a very powerful prediction model.

[0087] By using these three different regression methods, the method can select the most appropriate model to predict future sales based on the characteristics and needs of historical sales data. This method improves the prediction accuracy and generalization ability of the model, making the prediction results more reliable and accurate.

[0088] In this preferred embodiment, the features include: a first average sentiment score, a second average sentiment score, a year-on-year growth rate, a month-on-month growth rate, a first-order derivative, a second-order derivative, and sales volume.

[0089] It is understood that the features refer to various variables or data points used to build a sales forecasting model. These features are extracted from multiple data sources to describe and quantify different aspects related to sales. Specifically, these features include: First average sentiment score: This is extracted from the historical comments of the forum to indicate the overall sentiment of consumers towards the product or service. Second average sentiment score: This is extracted from the historical comments of the news release of the official website of the car series to indicate another sentiment of consumers towards the product or service. Year-on-year growth rate: This is extracted from the historical sales data to indicate the percentage increase in sales compared with the same period last year. Month-on-month growth rate: This is extracted from the historical sales data to indicate the percentage increase in volume compared with the previous month. First-order derivative: This is extracted from the growth trend of sales to indicate the speed of sales growth. Second-order derivative: This is extracted from the growth trend of sales to indicate the rate of change of the sales growth rate. Sales volume: This is extracted from the historical sales data to indicate the size of actual sales.

[0090] By combining these features, the method can describe and quantify factors related to sales from multiple perspectives, thereby building a more comprehensive and accurate sales forecasting model. These features help the model better understand the complexity of sales changes and improve the accuracy of forecasts.

[0091] In summary, the present invention discloses a vehicle sales forecasting method based on emotion. By collecting and analyzing historical comments in automobile forums and historical comments on news releases on the official website of the vehicle series, the average emotion score is calculated in combination with the emotion dictionary, and these emotion scores are used as one of the features. Combined with historical sales data, a sales forecasting model is established. Traditional sales forecasting methods may not be able to capture subtle changes in consumer emotions, but the present invention can more comprehensively consider the factors affecting sales by introducing emotion analysis, thereby improving the accuracy of the forecast. Forum comments and news releases can reflect the market's views and emotions on vehicles in real time. The present invention can capture changes in market emotions in a timely manner by collecting these data in real time, and provide the latest information for sales forecasting. By combining multiple features, such as emotion scores, year-on-year growth rates, month-on-month growth rates, etc., the model of the present invention can better adapt to changes in different market environments and sales trends. The present invention uses the root mean square error as an evaluation criterion, selects the model with the smallest error for prediction, thereby reducing the prediction error and improving the reliability of the prediction. By analyzing consumer emotions, automobile manufacturers and sellers can better understand market demand and consumer preferences, thereby formulating more effective marketing strategies and product improvement plans.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A vehicle sales prediction method based on emotion, characterized in that: include: Collect historical comments in automobile forums and historical comments on news releases on official websites of car series by month, perform data cleaning, word segmentation and feature conversion on the historical comments in the forums and historical comments on news releases, calculate the average sentiment score through the sentiment dictionary, and obtain the first average sentiment score and the second average sentiment score; Collect historical sales data, obtain the year-on-year growth rate, month-on-month growth rate and growth trend of each month, perform first-order and second-order derivatives on the growth trend to obtain the first-order derivative and the second-order derivative; wherein the growth trend is the difference between the sales volume of this month and the sales volume of last month; The features are formed for each month and trained based on the sales data of the past four years. All the features are aggregated by year and month, and the root mean square error is used as the evaluation standard to establish a sales forecasting model. Predict the sales data of the fourth year based on the sales data of the same month in the previous three years, calculate the root mean square error of the three models, and select the sales forecast model with the smallest root mean square error in each month as the forecast model for that month; Substitute the sales data of any month in the previous three years to obtain the predicted sales data of the same month in the current year.

2. The emotion-based vehicle sales prediction method according to claim 1, characterized in that: The method for collecting forum historical comments and news release historical comments includes: Determine the URLs of the automobile forum and the official website of the automobile series; Determine whether the automobile forum and the official website of the car series can obtain API. If the judgment result is yes, obtain API and collect comment data. If the judgment result is no, use a web crawler tool to crawl page content and extract comment data.

3. The emotion-based vehicle sales prediction method according to claim 2, characterized in that: The data cleaning includes: Remove special characters, punctuation marks and non-text content from the comment data; remove stop words according to a stop word list; remove duplicate records and text content from each of the comment data; and use a spell checker to correct spelling errors in the text.

4. The emotion-based vehicle sales prediction method according to claim 3, characterized in that: The word segmentation processing includes: A natural language processing library is used to perform Chinese word segmentation and a six-category sentiment dictionary is established. The six-category sentiment dictionary includes sentiment categories and corresponding sentiment words. The sentiment categories include joy, sadness, anger, fear, surprise and disgust.

5. The emotion-based vehicle sales prediction method according to claim 4, characterized in that: The feature conversion includes: The frequency of occurrence of each sentiment category and the corresponding sentiment words in all comments is counted to obtain word frequency data.

6. The emotion-based vehicle sales prediction method according to claim 5, characterized in that: Calculate the average sentiment score through the sentiment lexicon, including: Initialize the score of each sentiment category to 0; Divide the review data by month, and calculate the average sentiment score of the review data for each month, including the first average sentiment score and the second average sentiment score, where: The first average sentiment score is: The second average sentiment score is: Among them, si is the corresponding six categories of sentiment scores, ni represents the corresponding number of entries, the first average sentiment score corresponds to the historical comments in the automobile forum, and the second average sentiment score corresponds to the historical comments on the news release of the official website of the car series.

7. The emotion-based vehicle sales prediction method according to claim 6, characterized in that: The collection of historical sales data includes: Count the historical sales data t and calculate the year-on-year growth rate r1 and month-on-month growth rate r2 for each month, where: r1 = (this month's sales volume - last year's same month's sales volume) / last year's same month's sales volume × 100%; r2 = (sales volume this month - sales volume in the same month last year) / sales volume last month × 100%.

8. The emotion-based vehicle sales prediction method according to claim 7, characterized in that: The collecting of historical sales data also includes: Calculate the growth trend of sales, growth trend f = this month's sales - last month's sales, and calculate the first-order derivative f1 and the second-order derivative f2 of f respectively.

9. The emotion-based vehicle sales prediction method according to claim 8, characterized in that: The sales forecasting model is established by using polynomial regression, decision tree regression and XGBoost regression methods respectively.

10. The emotion-based vehicle sales prediction method according to claim 9, characterized in that: The features include: a first average sentiment score, a second average sentiment score, a year-on-year growth rate, a month-on-month growth rate, a first-order derivative, a second-order derivative, and sales volume.

Images