An Analysis Method of Sensory Demand Trends and Mutation Detection for Intelligent Connected Vehicles

Abstract

The present invention discloses a method for analyzing the trend of emotional needs and detecting mutations in automotive intelligent networking, including: obtaining user comment data, constructing a comment data set, training a Bi-LSTM neural network, and calculating the sentiment value of each comment within the time period to be analyzed; defining NSF to represent the normalized sentiment frequency value within a certain time interval, and calculating the NSF within the time period to be analyzed; verifying the trend stability of the NSF value. If the result is "stable", at this time, it is considered that the NSF value remains stable on the time series formed by all the comment release times to be analyzed, and no trend analysis is performed; while for the sequence with the result of "unstable", further trend tests are carried out to obtain the test result of an upward or downward trend; mutation analysis is carried out within the time period to be analyzed through various methods to determine the credible mutation time.

Description

Technical Field

[0001] The present invention relates to the fields of text mining and time series analysis, and particularly to a method for analyzing the trend of emotional needs and detecting mutations in intelligent connected vehicles. Background Art

[0002] With the rapid development of the automotive industry and intelligent connected technology, the competition in the intelligent connected vehicle industry is becoming increasingly fierce. For automobile enterprises, how to quickly identify the functions that need to be upgraded and iterated, and formulate product design and upgrade strategies with market commonality and enterprise characteristics is of great significance for occupying market share.

[0003] Kansei engineering is a user-centered product development method that combines sensibility and engineering, and its main task is to formulate product design and improvement plans based on user preferences. At present, there are relatively few patents related to extracting requirements using big data of user reviews, especially for monitoring the preferences of intelligent connected functions and assisting design decisions. In order to improve the user-centered product experience, automobile enterprises have gradually established an internal monitoring section based on the automotive ecosystem to mine users' emotional needs through big data.

[0004] The patent application with the publication number CN103605658A provides a search engine system based on text sensibility analysis, which includes four parts: sensibility analysis, emotion value and keyword statistics, index establishment, and hot topic extraction. This solution only analyzes the keywords with high corresponding word frequencies over time, rather than observing the sensibility changes in the field of this keyword globally.

[0005] The patent application with the publication number CN108763210A provides a sensibility analysis and prediction system based on automated data collection, which can automatically collect data and continuously give sensibility prediction results based on historical data during this process; however, this solution focuses on predicting the sensibility tendency and does not give the feature mining and review of historical time series information. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for analyzing the trend of emotional needs and detecting mutations in intelligent connected vehicles, so as to overcome the problems existing in the prior art.

[0007] To achieve the above tasks, the present invention adopts the following technical solutions:

[0008] A method for analyzing the trend of emotional needs and detecting mutations in intelligent connected vehicles includes the following steps:

[0009] Obtain user comment data and construct a comment dataset; divide the comment dataset into a training set and a validation set, clean the user comments in the comment dataset, and then perform word segmentation and stop word removal; extract keywords from the comments after word segmentation and stop word removal, and classify the extracted keywords according to different topics to construct a user keyword library; based on the keywords in the user keyword library, manually label the sentiment tendency of each comment in the training set with positive and negative labels.

[0010] Perform word embedding operations on the word segmentation results in the training set for each comment as a unit to generate corresponding word vectors, and import the word vectors of the comments and the corresponding positive / negative label results into a Bi-LSTM neural network for training. The input of the Bi-LSTM model is the word vector of each comment in the comment dataset, and the output is the sentiment value corresponding to each comment; use the trained Bi-LSTM model to calculate the sentiment values of each comment within the time period to be analyzed.

[0011] Define NSF to represent the normalized sentiment value frequency within a certain time interval, and calculate the NSF within the time period to be analyzed through the following formula:

[0012]

[0013]

[0014]

[0015]

[0016] where represents the normalized sentiment values of the self-owned brand and competing brands; represents the sentiment value of the i-th comment of the self-owned brand and competing brands that includes a certain design / configuration, represents the number of comments of the self-owned brand and competing brands that include a certain design / configuration; represents the sentiment value of the i-th comment among all comments of the self-owned brand and competing brands, represents the number of all comments of the self-owned brand and competing brands; ISF represents the inverse frequency of comment sentiment values;

[0017] Verify the trend stability of the NSF value. If the result is "stable", at this time, it is considered that the NSF value remains stable on the time series composed of the release times of all comments to be analyzed, and no trend analysis is performed; for the sequences with the result of "unstable", further trend tests are performed to obtain the test results of rising or falling trends.

[0018] Perform mutation analysis within the time period to be analyzed through various methods to determine the credible mutation time.

[0019] Furthermore, the mutation analysis within the time period to be analyzed is performed by multiple methods, including:

[0020] The mutation time point position is determined by comprehensive comparison of the three algorithms: Mann-Kendall, Pettitt, and SNHT;

[0021] Let a, b, and c represent three algorithms respectively, T a 、T b 、T c is the set of possible mutation time points solved by the corresponding method, and t i , t j , t k , ... indicates; t α The number of occurrences of (α=i,k,k…) is recorded as β α , considering the differences in algorithms, we allow t α ±t γ The error, t γ is the threshold time;

[0022] For possible mutation time points, frequency statistics are performed and finally T is specified as the credible mutation time based on the voting mechanism:

[0023] T={t α |β α ≥2}

[0024] Furthermore, the method for verifying the trend stability of the NSF value is an ADF test method.

[0025] Furthermore, the trend test adopts the Mann-Kendall method.

[0026] Furthermore, the test results and the credible mutation time are used to provide development suggestions to the enterprise.

[0027] Furthermore, the development proposals include:

[0028] For a certain design / configuration, when the trend stability analysis result of the NSF value is stable, it is recommended that the enterprise pay attention to the development trend of related designs / configurations in the industry in order to maintain the competitiveness of the design / configuration in the market;

[0029] For a certain design / configuration, when the trend analysis result is an upward trend, the mutation time point in the upward trend can be further detected. This time point indicates that users’ satisfaction with the design / configuration of their own brand is significantly higher than that of brands in the same industry. It is recommended that manufacturers focus on optimizing and improving the design / configuration.

[0030] For a certain design / configuration, when the result of trend analysis shows a decline, by further detecting the mutation time points during the decline process and combining with the comment information, study the reasons for lagging behind the industry peers near that time point.

[0031] Compared with the prior art, the present invention has the following technical features:

[0032] 1. Compared with traditional Kansei engineering analysis, it reduces the cost of research. This solution can quickly monitor the preferences for intelligent networked functions and assist in design decisions.

[0033] 2. Using deep learning methods to automatically extract Kansei requirements, reducing the workload of manual analysis in existing enterprises.

[0034] 3. Analyze the data in the time series dimension. Compared with the static analysis method of observing a single time slice, it has the advantage of mining data trends and can extract richer information.

[0035] 4. Comprehensively apply a variety of mutation detection methods, which not only shows the time of mutation but also can verify the accuracy of the mutation time.

[0036] 5. Innovatively propose the NSF concept and establish a new online comment Kansei requirement analysis system. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a schematic flow chart of the method in an embodiment of the present invention;

[0038] Figure 2 is a schematic diagram showing the change of NSF with the comment release time on the abscissa in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Referring to the accompanying drawings, a method for analyzing the trend and detecting mutations of Kansei requirements for intelligent networked vehicles provided by the present invention includes the following steps:

[0040] Step 1, obtaining demand intention words

[0041] Step 1.1, data acquisition and data cleaning.

[0042] Use computer programming or tools such as Octopus to crawl network data, crawl user comments on Internet platforms such as e-commerce platforms, forums, and social media, obtain user IDs, speech times, comment contents, etc., and form a comment data set. Here, taking Chinese comment texts as an example, to reduce the interference of bad data and ensure data validity, the filtering rules of this solution are set as follows:

[0043] System default comments. This means that the user has only posted non-text comments, such as ratings, pictures, videos, etc. In this case, it is automatically determined as "not filled in by this user". In this case, this type of information that does not contain user opinions should be removed.

[0044] Incomplete comments. Comments that only contain a lot of modal particles and daily expressions, such as "OK", "Thank you", "Congratulations", etc. Information that does not match the topic should be removed.

[0045] Duplicate comments. Refers to online comments with the same text content repeated multiple times in a short period of time. Such comments may be related to the user's own behavior, such as to earn comment points and check-in day rewards, but they may also be malicious comments from competitors. For this reason, this solution removes a large number of duplicate comments and only retains the first comment.

[0046] Step 1.2: Segment the comments in the comment dataset and remove stop words.

[0047] Tokenizing is to reasonably segment the comment content. The document obtained after tokenization is filtered through a stop word list to obtain a reasonable tokenization result.

[0048] Original text: The new model has a very cool appearance, the seats are very high-end and comfortable, and the driving experience is first-class

[0049] After word segmentation: new model / product / of / appearance / very / cool / seat / very / high-end / comfortable / driving / experience / quite / first-class

[0050] After removing stop words: new style / product / appearance / cool / seat / high-end / comfortable / driving / experience / first-class

[0051] Since the comment information obtained above is not labeled, and the text volume is relatively large and the field is relatively concentrated, the TF-IDF algorithm is sensitive to document information and is used to extract keywords from the comments after word segmentation and stop word removal to obtain keywords such as ACC, USB, battery, steering wheel, instrument, etc. According to the common key designs of automobiles, the extracted keywords are classified according to different themes to build a user keyword library.

[0052] User keyword table example

[0053]

[0054] Step 1.3, manually annotate the training set comments in the comment dataset.

[0055] By means of manual annotation, tags are assigned to each comment in the training set; for comments containing keywords mentioned in the keyword library (such as seat, steering wheel, automatic start-stop, etc.) and with a positive overall meaning, a label of 1 is assigned, while for those with a negative meaning, a label of 0 is assigned.

[0056] Step 2, Establishment of the perceptual evaluation system

[0057] Regarding the evaluation methods of text sentiment polarity, there are existing methods such as sentiment polarity analysis, sentiment intensity analysis, subjective-objective analysis, etc. Due to the diverse expressions of perceptual features, subjectively, words describing emotions such as happy, sad, distressed, etc. can be used for expression, but this can only rely on perceptual knowledge to evaluate the positive / negative direction and intensity of perception. In particular, in order to conveniently evaluate the change of perception from the time series perspective, the present invention establishes a new perceptual evaluation system, introducing the sentiment value and the number of comments of the self-owned brand and competing brands as parameters, which can effectively measure the relative sentiment value of the self-owned brand in the market within different time intervals.

[0058] Definition 1: Sentiment Value. Define a real number in the range of [0,1], which is a floating-point number and can take continuous values within the defined range. Tending to 0 indicates negative polarity, and tending to 1 indicates positive polarity.

[0059] The calculation method of the sentiment value is as follows:

[0060] Step 2.1, Through word embedding operations on the words after word segmentation and stop word removal, corresponding word vectors are generated. The semantic similarity of word vectors is reflected as spatial similarity, that is, words with closer spatial distances have higher semantic similarity. Here, the word vector is the mathematical representation of the word, which is a kind of structured data for facilitating subsequent text data modeling operations.

[0061] Step 2.2, Divide the comment data set into a training set and a validation set. Use the keras interface in the deep learning framework TensorFlow to implement a Bi-LSTM (Bidirectional Long Short-Term Memory) neural network. Using the labeled training set, the input of the Bi-LSTM model is the word vectors in the training set, and the output is the sentiment value. After training, verify through the validation set to obtain the trained model. Note: The sentiment value is only for one comment. The NSF mentioned later represents the normalized sentiment frequency value obtained from all comments within a certain time interval.

[0062] Step 2.3, Use the trained Bi-LSTM model to calculate the sentiment value (Sentiment Value) of each comment within the time period to be analyzed (word segmentation, stop word removal, and word embedding processing are required before inputting into the network), and obtain the following example results.

[0063] Emotional Tendency Score Table

[0064]

[0065] Define 2NSF (Normalized Sentiment Frequency) as the normalized sentiment frequency value within a certain time interval (such as one day, one week, one month, etc.).

[0066] Calculate the NSF within the time period to be analyzed through the following formula:

[0067]

[0068]

[0069]

[0070]

[0071] Wherein, represents the normalized sentiment value of the self-owned brand and the competing brand; represents the sentiment value of the i-th comment containing a certain design / configuration of the self-owned brand and the competing brand, represents the number of comments containing a certain design / configuration of the self-owned brand and the competing brand; represents the sentiment value of the i-th comment among all comments of the self-owned brand and the competing brand, represents the number of all comments of the self-owned brand and the competing brand; ISF represents the inverse frequency of the comment sentiment value.

[0072] Step 3, Implementation of the Emotional Time Series Monitoring Method

[0073] The present invention combines the NSF value with the time series analysis method to determine the mutation time point of the NSF value. In order to describe the change of the desired NSF in the time series, the present invention adopts trend analysis and mutation analysis as means to mine data characteristics.

[0074] Step 3.1, Use the existing ADF test method (Augmented Dickey-Fuller test) to verify the trend stability of the NSF value. If the result is "stable", at this time, it is considered that the NSF value remains stable in the time series composed of the comment release times to be analyzed, without a significant upward or downward trend, then no trend analysis is performed; for the sequence with the result of "unstable", further perform the Mann-Kendall trend test (abbreviated as M-K trend test). The principle of the M-K trend test is as follows:

[0075] Hypothesis:

[0076] H0: The data in the sequence is randomly arranged, i.e., there is no significant trend.

[0077] H1: There is an upward or downward trend in the sequence, i.e., a monotonic trend.

[0078] For the sequence X = {x 1 , x 2 ,......, x n}, x j is the data point at the previous moment, and x i is the data point at the current moment. First, determine the size relationship between x i and x j in all paired values (x i , x j , i > j). Let n be the total number of data points in the time series. The test statistic is denoted as S as shown in the formula:

[0079]

[0080] Among them, the definition of the sign function is shown in the formula:

[0081]

[0082] Here, the S statistic is standardized to obtain the Z statistic. The calculation method of Z is shown in the formula:

[0083]

[0084] Among them, Var is the variance of the S value. The calculation method of Var(S) is shown in the formula:

[0085]

[0086] The meaning of the Z statistic is as follows: Looking up the normal distribution table, it is known that the critical statistic values of Z at the significance levels α = 0.01, 0.05, and 0.1 are 2.58, 1.96, and 1.65 respectively. At this time, if the test statistic rejects the null hypothesis at this time, it is considered that there is a significant change trend at this level; if the test statistic accepts the null hypothesis, indicating that the trend is not significant. In addition, the Z statistic has another meaning: If the null hypothesis is rejected, that is, when there is an obvious upward or downward trend, when Z > 0, it is an upward trend, and when Z < 0, it is a downward trend.

[0087] Since the NSF value represents the relative perceptual frequency value of brand A compared to competitor brand B in the market. If the NSF value continues to increase, it indicates that the customer satisfaction of the relevant design / configuration of this enterprise is in a growing state. At this time, from the perspective of product update and iteration, it shows that this design / configuration of this brand has certain development potential, and the function can be focused on for optimization and improvement, and then recommended to customers as an "ace" function.

[0088] When the NSF value has been in a downturn or shows a downward trend for a period of time, it indicates that the satisfaction of this enterprise's products is not high compared with other products in the market, and is even worse than the industry average. Therefore, this enterprise can take this as a warning of poor customer satisfaction, customer loss or lack of market competitiveness of its products.

[0089] Step 3.2, mutation analysis.

[0090] The purpose of mutation analysis is to examine whether there are significant mutation points in the period to be analyzed with a trend. There are four forms of mutation: namely, mean mutation (change from one mean to another), variance mutation (the mean remains unchanged while the variance changes sharply), seesaw mutation (the mean repeatedly shows jump changes), and turning point mutation (the mean of a certain period changes monotonically continuously and suddenly changes monotonically in the opposite direction at a certain point).

[0091] This invention focuses on the situation of mean mutation. By comprehensively comparing the three algorithms of Mann-Kendall, Pettitt, and SNHT (Standard Normal Homogeneity Test), the position of the mutation time point is judged. Now, let a, b, and c represent the three algorithms respectively, and T a , T b , T c are the sets of possible mutation time points solved by the corresponding methods, and the elements of the set are represented by t i , t j , t k ,... The occurrence times of t α (α = i, k, k...) are recorded as β α . Considering the differences in algorithms, here it is allowed that t α has an error of ±t γ , and the threshold time t γ is determined according to actual needs.

[0092] T a = {t i , t j ,...}

[0093] T b = {t k , t l ,...}

[0094] T c = {t m , t n ,...}

[0095] For the possible mutation time points, frequency statistics are carried out and finally according to the voting mechanism, T is defined as the credible mutation time:

[0096] T = {t α | β α ≥ 2}

[0097] Supplementary note: All possible mutation time points t calculated by the three algorithms α are regarded as elements in a large set and frequency statistics are carried out. Since the lengths of the time series are different in the analysis process, the results of different algorithms generally have deviations, and it is difficult to accurately estimate the results to the same day. Therefore, a threshold time t γ is introduced as an adjustable allowable error, and t γ is determined according to actual needs. The specific calculation process is as follows:

[0098] Analyze the frequency statistics results in different cases. If a possible mutation time point t α appears with a frequency β α greater than or equal to 2, it is regarded that the results of at least two algorithms point to this possible mutation time. At this time, t α becomes an element of the credible mutation time T; if the value of a possible mutation time point t α appears with a frequency β α less than 2, it is considered that the possible mutation time calculated this time is accidental and has low credibility, and it is not regarded as a credible mutation time point. At this time, this t α does not meet the condition of becoming an element of the credible mutation time T and is discarded.

[0099] For easy understanding, an example is given as follows:

[0100] Suppose t γ is 5 days. Then when the three algorithms obtain the following results:

[0101] T a = {"2017 / 1 / 1", "2017 / 2 / 1"}

[0102] T b = {"2017 / 1 / 5"}

[0103] T c = {"2017 / 3 / 1"}

[0104] Among them, when t i = "2017 / 1 / 1" and t j = "2017 / 1 / 5", at this time t j meets the range of t i ± t γ That is, it is regarded that t i appears 2 times, β i = 2. Finally, from T = {t α | β α≥2} gives T = {"2017 / 1 / 1"}

[0105] After completing the mutation analysis, the following suggestions can be given:

[0106] Case 1: For a certain design / configuration, when the trend stability analysis result of the NSF value is stable, it indicates that the satisfaction of users with this design / configuration of this brand is on par with that of other brands in the market. At this time, it is recommended that enterprises pay attention to the development trends of relevant designs / configurations in the industry, grasp the development direction in a timely manner, so as to maintain the competitiveness of this design / configuration in the market.

[0107] Case 2: For a certain design / configuration, when the trend analysis result of the NSF value is on the rise, the mutation time point during the rising process can be further detected. This time point indicates that the satisfaction of users with this design / configuration of the own brand is significantly higher than that of brands in the same industry, reflecting that this design / configuration has characteristics leading the market. For this reason, it is recommended that production enterprises focus on optimizing and improving this design / configuration, and then recommend it to customers as an "ace" function to increase user stickiness and seize market share by leveraging their own advantages.

[0108] Case 3: For a certain design / configuration, when the trend analysis result of the NSF value is on the decline, it indicates that the satisfaction of users with this design / configuration of this brand is lower than the market average level during the same period. By further detecting the mutation time point during the decline process, the reasons for lagging behind competitors near this time point can be studied by combining comment information. At this time, it is recommended that enterprises pay more attention to customer complaints, strengthen after-sales management, promptly discover the deficiencies in services or products, and make up for the gaps in a timely manner to avoid customer loss caused by too low customer satisfaction.

[0109] Embodiment

[0110] Taking TJA (Traffic Jam Assist) as an example, sentiment analysis is performed on the comment texts under this keyword within a certain time interval to obtain sentiment values.

[0111]

[0112] Sampling is carried out with a one-week time interval to calculate the NSF value, as shown in the following table:

[0113]

[0114]

[0115] For the NSF value, trend analysis is carried out to obtain "extremely significant increase".

[0116]

[0117] For the NSF value, mutation time point analysis is carried out to obtain the following results.

[0118]

[0119] According to the above solution, first, it is obtained from trend analysis that the NSF value of the "TJA" design / configuration of Brand A cars shows an "extremely significant increase" trend. At this time, mutation analysis is further carried out. If the threshold time t γ is set to 10 days, according to the solution process, the credible mutation time T = {"2017 / 12 / 31"} is calculated, which means that the credibility of the mean mutation of the NSF value on December 31, 2017 is the highest. Through this analysis, it can be found that from September 2017 to June 2018, the NSF value of the "TJA" design / configuration of Brand A cars was in a state of extremely significant increase, and December 31, 2017 was the credible mutation time of the mean mutation.

[0120] As Figure 2 shown in the figure, in the figure, first, the trend analysis obtained an upward result. At this time, mutation analysis is further carried out, that is, to study at which time point the NSF mean value has a step change before and after, and this time point is determined as the mutation time point; the vertical trend lines in the figure show the possible mutation time point results obtained by various methods. For traditional kansei engineering analysis, manual research costs are high. The present invention uses deep learning methods to automatically extract kansei requirements, which can effectively reduce the workload of manual analysis in existing enterprises. For the problem of converting user comments into kansei information, the present invention proposes a kansei analysis model and innovatively puts forward the concept of NSF (Normalized Sensory Frequency), which fully considers the information of the current analyzed brand and brands of the same industry in the automotive industry, and is more objective. For the existing static analysis method of observing a single time slice, the present invention analyzes the NSF data in a time series, which has the advantage of mining data trends and can extract richer information. For the detection of mutation points by a single method in time series analysis, the present invention uses a variety of mutation detection methods, which can not only show the time of mutation but also verify the accuracy of the mutation time.

Claims

1. An analysis method for the emotional demand trend and mutation detection of automotive intelligent networking, characterized in that, it includes the following steps: Obtain user comment data and construct a comment data set; Divide the comment data set into a training set and a validation set, clean the user comments in the comment data set, and then perform word segmentation and stop word removal processing; Extract keywords from the comments after word segmentation and stop word removal, classify the extracted keywords according to different themes, so as to construct a user keyword library; based on the keywords in the user keyword library, manually label the data in the training set to assign a positive or negative emotional label to each comment; Perform word embedding operations on the vocabulary in the comment data set to generate corresponding word vectors; Use the comment data set to train a Bi-LSTM neural network. The input of the Bi-LSTM model is the word vectors in the comment data set, and the output is the emotional value; use the trained Bi-LSTM model to calculate the emotional value of each comment within the time period to be analyzed; Define NSF to represent the normalized emotional frequency value within a certain time interval, and calculate the NSF within the time period to be analyzed through the following formula: Among them, represents the standardized sentiment value of the self-owned brand and competing brands; represents the sentiment value of the i-th comment on the self-owned brand and competing brands that includes a certain design / configuration, represents the number of comments on the self-owned brand and competing brands that include a certain design / configuration; represents the sentiment value of the i-th comment among all comments on the self-owned brand and competing brands, represents the number of all comments on the self-owned brand and competing brands; ISF represents the inverse frequency of comment sentiment values; Verify the trend stability of the NSF value. If the result is "stable", at this time, it is considered that the NSF value remains stable on the time series composed of all the comment release times to be analyzed, and no trend analysis is performed; for the sequences with the result of "unstable", further trend tests are performed to obtain the test results of upward or downward trends; Perform mutation analysis within the time period to be analyzed through multiple methods to determine the credible mutation time; The mutation analysis within the time period to be analyzed through multiple methods includes: Comprehensively compare through the Mann-Kendall, Pettitt, and SNHT algorithms to judge the position of the mutation time point; Now, let a, b, and c represent three algorithms respectively, and T a , T b , T c are the sets of possible mutation time points solved by the corresponding methods. The elements of the set are represented by t i , t j , t k ,...; The number of occurrences of t α is denoted as β α , where α = i, j, k..., considering the differences in algorithms, here t α is allowed to have an error of ±t γ . The threshold time t γ is determined by yourself according to actual needs; T a = {t i , t j , …} T b = {t k , t l , …} T c = {t m , t n , …} For the possible mutation time points, perform frequency statistics and finally, according to the voting mechanism, stipulate that T is the credible mutation time: T = {t α | β α ≥ 2}.

2. The analysis method for the emotional demand trend and mutation detection of automotive intelligent networking according to claim 1, characterized in that, the method for verifying the trend stability of the NSF value is the ADF test method.

3. The analysis method for the emotional demand trend and mutation detection of automotive intelligent networking according to claim 1, characterized in that, the trend test adopts the Mann-Kendall method.

4. The analysis method for the emotional demand trend and mutation detection of automotive intelligent networking according to claim 1, characterized in that, the test results and the credible mutation time are used to provide development suggestions to enterprises.

5. The analysis method for the emotional demand trend and mutation detection of automotive intelligent networking according to claim 4, characterized in that, the development suggestions include: For a certain design / configuration, when the trend stability analysis result of the NSF value is stable, it is recommended that enterprises pay attention to the development trends of relevant designs / configurations in the industry in order to maintain the competitiveness of this design / configuration in the market; For a certain design / configuration, when the trend analysis result shows an upward trend, the mutation time point during the upward process can be further detected. This time point indicates that the user's satisfaction with this design / configuration of the private brand has increased significantly compared to that of the industry brands. It is recommended that the manufacturing enterprise focus on optimizing and improving this design / configuration. For a certain design / configuration, when the trend analysis result shows a downward trend, by further detecting the mutation time point during the downward process, the reasons for lagging behind the industry level near this time point can be studied in combination with the review information.

Images