Potential vehicle demand mining method, electronic equipment, storage medium and program product

Abstract

The embodiment of the invention provides a potential vehicle demand mining method, electronic equipment, a storage medium and a program product. The method comprises the steps of obtaining target comment data of a user for a vehicle; extracting product force evaluation vocabularies from the target comment data; performing hierarchical division on the product force evaluation vocabulary to obtain a hierarchical division result; performing hierarchical word frequency statistics on the hierarchical division result by using a mining model to obtain a hierarchical word frequency statistical result; based on the hierarchical word frequency statistical result, screening out a plurality of target vehicle model indexes; performing emotional tendency analysis on comment data related to the target vehicle model indexes in the target comment data by using the mining model, and determining an emotional tendency evaluation proportion corresponding to each target vehicle model index; based on the emotional tendency evaluation proportion, determining an emotional score corresponding to each target vehicle model index; and determining a potential vehicle demand of the user based on the emotion score. And the accuracy of demand mining is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a potential vehicle demand mining method, an electronic device, a storage medium and a program product. BACKGROUND

[0002] In the process of defining vehicle models, automobile manufacturers often create products based on their own technical reserves, production capacity and established market strategy planning, focusing on performance parameters, configuration combinations and cost control of vehicle models, and trying to create standardized products that meet their own concepts and market competition needs. However, with the continuous development of the automobile market and the increasing diversification of consumer demand, users' demand for vehicle models has far exceeded the traditional definition of manufacturers. This huge gap between vehicle model definition and user demand results in the inability of manufacturers to accurately meet the expectations of target user groups, thereby affecting the market performance and sales performance of products. Therefore, how to narrow the gap between vehicle model definition and user demand has become a problem to be solved. SUMMARY

[0003] The purpose of the embodiments of the present application is to provide a potential vehicle demand mining method, an electronic device, a storage medium and a program product, to realize the technical effect of accurately mining user vehicle demand.

[0004] The first aspect of the embodiments of the present application provides a potential vehicle demand mining method, the method comprising: obtaining target comment data of a user for a vehicle; extracting product power evaluation vocabulary from the target comment data; hierarchically dividing the product power evaluation vocabulary to obtain a hierarchical division result; performing hierarchical word frequency statistics on the hierarchical division result by using a mining model to obtain a hierarchical word frequency statistics result; based on the hierarchical word frequency statistics result, screening a plurality of target vehicle model indicators; performing sentiment tendency analysis on comment data related to the target vehicle model indicators in the target comment data by using the mining model to determine a sentiment tendency evaluation proportion corresponding to each target vehicle model indicator; based on the sentiment tendency evaluation proportion, determining a sentiment score corresponding to each target vehicle model indicator; based on the sentiment score, determining the potential vehicle demand of the user.

[0005] In the implementation process, the potential vehicle demand is finally determined by extracting product power vocabulary, hierarchical division, word frequency statistics, sentiment analysis and score calculation from user vehicle comments, realizing the whole process processing from user comment data to precise demand mining, providing decision basis for vehicle product optimization, and improving the efficiency and accuracy of demand mining.

[0006] Further, before the target comment data of the user for the vehicle is acquired, the method further comprises: acquiring initial comment data of the user for the vehicle; cleaning the initial comment data to obtain the target comment data; the cleaning process comprises a duplicate value removal process.

[0007] In the implementation process, the quality of the target comment data is ensured by cleaning (such as removing duplicates) the initial comment data, avoiding interference of repeated data in subsequent analysis.

[0008] Further, based on the sentiment tendency evaluation proportion, the emotion score corresponding to each target vehicle model index is determined, comprising: based on the hierarchical word frequency statistics result, determining the word frequency weight corresponding to each target vehicle model index; based on the word frequency weight of each target vehicle model index and the corresponding sentiment tendency evaluation proportion, obtaining the emotion score corresponding to each target vehicle model index.

[0009] In the implementation process, by introducing the word frequency weight, the user attention heat of the target vehicle model index and the sentiment tendency evaluation proportion are combined to calculate the emotion score, so that the score reflects the attention degree and the sentiment attitude, and the score is more in line with the real demand priority of the user.

[0010] Further, the hierarchical division result comprises an evaluation dimension and an evaluation index belonging to the evaluation dimension; the hierarchical word frequency statistics of the product power evaluation vocabulary by using the mining model comprises: the word frequency statistics of the product power evaluation vocabulary by using the mining model according to the evaluation dimension, and screening out a target evaluation dimension with a word frequency order meeting a first preset requirement; based on the hierarchical word frequency statistics result, screening out a plurality of target vehicle model indexes, comprising: performing word frequency statistics on the evaluation indexes belonging to the target evaluation dimension, screening out a target evaluation index with a word frequency order meeting a second preset requirement, and determining the target evaluation index as the target vehicle model index.

[0011] In the implementation process, the target evaluation dimension with high attention is first screened through the evaluation dimension-evaluation index hierarchical structure, and then the target evaluation index is screened therefrom, thereby improving the pertinence and efficiency of target vehicle model index screening.

[0012] Further, the hierarchical word frequency statistics result is obtained by inputting the hierarchical division result into the mining model, so that the mining model respectively counts the occurrence frequency of each product power evaluation word in the target review data according to the hierarchical structure of the hierarchical division result, and generates the hierarchical word frequency statistics result based on the occurrence frequency of each product power evaluation word and the corresponding hierarchical information. Further, the hierarchical word frequency statistics result is obtained by inputting the hierarchical division result into the mining model, so that the mining model respectively counts the occurrence frequency of each product power evaluation word in the target review data according to the hierarchical structure of the hierarchical division result, and generates the hierarchical word frequency statistics result based on the occurrence frequency of each product power evaluation word and the corresponding hierarchical information.

[0013] In the implementation process, the process of mining model counting product power word frequency according to hierarchical structure and generating structured result is determined, which ensures the association between word frequency and hierarchy, and provides structured and directly usable quantitative data for subsequent screening and analysis.

[0014] Further, the sentiment tendency includes positive, negative, and neutral; and the sentiment tendency analysis of the review data related to the target vehicle model index by using the mining model to determine the sentiment tendency evaluation proportion corresponding to each target vehicle model index, includes: The target word combination in the review data is identified by using the mining model to obtain the sentiment tendency judgment result of each review data; the target word combination includes sentiment words and degree adverbs, and the sentiment tendency judgment result includes positive result or negative result or neutral result; For each target vehicle model index, the sentiment tendency evaluation proportion is obtained based on the proportion of the number of review data of each sentiment tendency to the total number of review data corresponding to the target vehicle model index.

[0015] In the implementation process, the sentiment tendency of a single review is determined by identifying the combination of sentiment words and degree adverbs, and the sentiment tendency evaluation proportion is obtained by counting the proportion, thereby quantifying the sentiment distribution of users on the target vehicle model index.

[0016] Further, the potential vehicle demand of the user is determined based on the sentiment score, including: In the case where the sentiment score exceeds a first threshold value, the target vehicle model index corresponding to the sentiment score is determined as a superior demand point of the user; In the case where the sentiment score does not exceed a second threshold value, the target vehicle model index corresponding to the sentiment score is determined as an improved demand point of the user; the first threshold value is higher than the second threshold value; Based on the advantage demand point and the demand point to be improved, the potential vehicle demand is determined.

[0017] In the implementation process, by setting the first threshold and the second threshold, the emotional score is converted into the advantage demand point (user satisfaction) and the demand point to be improved (user dissatisfaction), and the advantages and disadvantages of the vehicle product are determined, thereby providing a clear decision direction for vehicle model optimization.

[0018] The second aspect of the embodiment of the application provides an electronic device, the electronic device comprising: a processor; a memory for storing processor-executable instructions; wherein the processor calls the executable instructions to implement the method of any one of the first aspect.

[0019] The third aspect of the embodiment of the application provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are executed by a processor to implement the steps of the method of any one of the first aspect.

[0020] The fourth aspect of the embodiment of the application provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the method of any one of the first aspect. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments of the application. It should be understood that the following drawings only show some embodiments of the application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0022] Figure 1 A flowchart of a potential vehicle demand mining method provided by the embodiment of the application; Figure 2 A user evaluation sentiment tendency analysis flowchart provided by the embodiment of the application; Figure 3 A mining principle block diagram provided by the embodiment of the application; Figure 4 A two-level classification word frequency statistical column chart provided by the embodiment of the application; Figure 5 A three-level classification word frequency statistical column chart provided by the embodiment of the application; Figure 6 An index sentiment analysis visualization chart provided by the embodiment of the application; Figure 7A structural block diagram of an electronic device is provided in the embodiments of the present application. DETAILED DESCRIPTION

[0023] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

[0024] It should be noted that similar reference numerals and letters refer to similar items in the following drawings, and therefore, once an item is defined in one drawing, it need not be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", and the like are merely used for distinguishing description, and cannot be understood as indicating or implying relative importance.

[0025] For any problem mentioned above, the embodiments of the present application provide a potential vehicle demand mining method, referring to Figure 1 , Figure 1 A flowchart of a potential vehicle demand mining method is provided in the embodiments of the present application.

[0026] In the embodiments, the method comprises: Step S10: obtaining target comment data of a user for a vehicle; In a specific implementation, the target comment data refers to an effective comment set focusing on a specific vehicle evaluation dimension (such as appearance) obtained from a car vertical medium (such as a car forum, a vertical car website, etc.) and after cleaning processing. For example, if the appearance category potential demand is mined, the cleaning process comprises: De-noising: deleting the content irrelevant to appearance, such as the comments of other categories (non-appearance) such as power, space, and interior of the car vertical medium comments, and only keeping the comments containing appearance evaluation (such as front face atmosphere, cool back shape); De-duplicate value: removing the same appearance evaluation comments published repeatedly; De-missing value: deleting the comments with incomplete appearance evaluation content, such as the comments mentioning only that the appearance is good without specific appearance detail description.

[0027] After these processes, the clean and effective target comment data focusing on the appearance category is obtained.

[0028] Step S20: extracting product power evaluation words from the target comment data; Exemplarily, the product force evaluation vocabulary refers to a vocabulary that can directly reflect the product force of the vehicle appearance, which is extracted based on the target review data focusing on the appearance. For example, from the review "The XX model's down-pressing front end creates a strong sense of movement, and the dragon-like design of the front face with sharp headlights is very domineering", the vocabulary "down-pressing front end", "dragon-like design" and "sharp headlights" can be extracted; from "The XX model's appearance is atmospheric, the waist line is handsome, and the sleek design is eye-catching", the vocabulary "appearance is atmospheric", "waist line is handsome" and "sleek design" can be extracted. These vocabulary points to the core evaluation points of the vehicle appearance, and therefore is determined as the product force evaluation vocabulary.

[0029] Step S30: performing hierarchical division on the product force evaluation vocabulary to obtain a hierarchical division result; Exemplarily, the hierarchical division result refers to a structured system in which the extracted product force evaluation vocabulary is disassembled into a "two-level classification-three-level index" structure according to the evaluation logic of the automobile product force (such as the user concern dimension), wherein the "two-level classification" is the core category of the vehicle product force, and the "three-level index" is the specific evaluation dimension belonging to the corresponding two-level classification.

[0030] The two-level classification is based on the characteristics of the automobile industry and the high-frequency concern direction of the user review, and can include but is not limited to driving, seating space, rear row, storage, air conditioning, main driver, operation, co-driver, quietness, battery, etc.

[0031] The three-level index is a refinement of the two-level classification, corresponding to the specific content of the user's actual evaluation, and can include but is not limited to driving smoothness, driving noise, main driver seating space, co-driver seating space, rear row seating space, etc. Each three-level index points to specific product force details under the two-level classification.

[0032] Step S40: performing hierarchical word frequency statistics on the hierarchical division result using a mining model to obtain a hierarchical word frequency statistics result; It should be noted that the mining model refers to a model with natural language understanding, text classification and statistical capabilities.

[0033] Exemplarily, the hierarchical word frequency statistics result refers to structured data containing "two-level classification, three-level index, frequency" generated by the mining model after counting the total frequency of the product force evaluation vocabulary corresponding to each three-level index in the target review data according to the "two-level classification-three-level index" hierarchy.

[0034] As an example, taking the two-level classification as seating space and the three-level index as rear row seating space and main driver seating space as an example: The hierarchical division rule is input to the mining model (such as the vocabulary "rear row space is large" and "rear row is not crowded" being classified as "seating space-rear row seating space"), and the model learns the rule through natural language understanding; The mining model traverses the target review data, and matches and classifies product power evaluation words in each comment: The comment "the rear space is large enough for three adults to sit without being crowded" is matched with "riding space-rear riding space", and the frequency is +1; The comment "the legroom of the driver is sufficient" is matched with "riding space-driver riding space", and the frequency is +1; The mining model outputs the hierarchical word frequency statistics result: Riding space-rear riding space-occurrence frequency 1113 times; Riding space-driver riding space-occurrence frequency 648 times.

[0035] Step S50: Based on the hierarchical word frequency statistics result, a plurality of target vehicle model indicators are screened out; It should be noted that the target vehicle model indicator refers to the three-level indicator with the highest user attention degree (i.e., the specific evaluation point mentioned most frequently in the comment) selected from the hierarchical word frequency statistics result. Alternatively, the top N (such as the top 10) indicators are selected as the target vehicle model indicators in descending order of the word frequency of the three-level indicators, so as to ensure that the core attention points of the user are covered.

[0036] As an example, all three-level indicator word frequencies under the secondary classification (such as rear riding space 1113 times, trunk space 747 times, and driving noise 457 times) are summarized; the top 10 three-level indicators (such as rear riding space, trunk space, and driver riding space) are selected in descending order of the word frequency; and these indicators are determined as the target vehicle model indicators as the object of subsequent sentiment analysis.

[0037] Step S60: The mining model is used to perform sentiment tendency analysis on the comment data related to the target vehicle model indicators in the target review data, and determine the sentiment tendency evaluation proportion corresponding to each target vehicle model indicator; It should be noted that the related comment data refers to the comment that explicitly mentions a target vehicle model indicator (such as the comment on driving noise: "the noise is too large when driving at high speed").

[0038] Sentiment tendency analysis refers to the mining model determining the sentiment tendency (positive, negative, or neutral) of each comment by identifying sentiment words (such as "good" and "bad"), degree adverbs (such as "very" and "too"), and negative words (such as "not") in the comment.

[0039] The sentiment tendency evaluation proportion refers to the percentage of the number of positive, negative, and neutral comments of each target vehicle model indicator in the total number of related comments of the indicator.

[0040] As an example, taking the target vehicle model indicators as "rear seat space" and "driving noise" as an example: the comments related to the above indicators are screened from the target comment data (such as comments "rear space is large and comfortable", "high-speed noise is too obvious and affects experience", etc.); the comments and sentiment judgment rules (such as sentiment judgment rules that the words "large" and "comfortable" are positive, and the words "too obvious" and "affect" are negative, etc.) are input into the mining model, and the mining model outputs the sentiment tendency of each comment through natural language understanding; the number of comments of each sentiment tendency is counted and the proportion is calculated, such as (target vehicle model indicator) rear seat space- (total number of related comments) 822- (positive proportion) 86%- (negative proportion) 3.3%- (neutral proportion) 10.7%; (target vehicle model indicator) driving noise- (total number of related comments) 457- (positive proportion) 10.3%- (negative proportion) 65.6%- (neutral proportion) 24.1%.

[0041] Step S70: determining the sentiment score corresponding to each of the target vehicle model indicators based on the sentiment tendency evaluation proportion; It should be noted that the sentiment score refers to a numerical value quantitatively reflecting the user's satisfaction with the target vehicle model indicators, which is calculated by the sentiment tendency evaluation proportion.

[0042] Converting the qualitative sentiment tendency into a quantitative score facilitates horizontal comparison of user satisfaction with different indicators.

[0043] In a specific implementation, the sentiment tendency evaluation proportion refers to, for a certain target vehicle model index, the percentage of the number of "positive evaluation", "neutral evaluation" and "negative evaluation" in the relevant comment data of the index in the total number of comments of the index (for example, the positive proportion of "rear seat space" is 86%, the neutral proportion is 10.7%, and the negative proportion is 3.3%). The sentiment score corresponding to each target vehicle model index is calculated by weighted summation of "sentiment tendency evaluation proportion x corresponding weight", wherein the weight corresponding to each sentiment tendency is dynamically generated based on the proportion of each sentiment tendency in the total comments of the target vehicle model index. Specifically, the positive weight = the positive evaluation proportion (reflecting the influence of positive sentiment in the total evaluation, the higher the proportion, the greater the positive weight); the neutral weight = the neutral evaluation proportion (reflecting the influence of neutral sentiment, the higher the proportion, the greater the neutral weight); the negative weight = - the negative evaluation proportion (the influence of negative sentiment is positively correlated with the proportion, and the symbol is negative, indicating negative influence). Then the sentiment score = the positive evaluation proportion x the positive weight + the neutral evaluation proportion x the neutral weight + the negative evaluation proportion x the negative weight. As an example, the positive evaluation proportion of (target vehicle model index) rear seat space = 86% (0.86), the neutral evaluation proportion = 10.7% (0.107), and the negative evaluation proportion = 3.3% (0.033); dynamically determine the weight: positive weight = 0.86, neutral weight = 0.107, and negative weight = -0.033; then the sentiment score of the rear seat space = 0.86x0.86 + 0.107x0.107x0 + 0.033x(-0.033) = 0.7396-0.001089≈0.7385. The score of the rear seat space is close to 0.74, indicating that the user's satisfaction with "rear seat space" is high, which is a core advantage demand point.

[0044] Step S80: Based on the sentiment score, determine the potential vehicle demand of the user.

[0045] It should be noted that the sentiment score quantitatively identifies the user's satisfaction with each core vehicle model index of the vehicle, and then extracts the advantage demand point and the demand point to be improved that the user is not satisfied with, and finally forms the potential vehicle demand that can guide the vehicle design, optimization or market decision. For example, integrating the advantage demand point and the demand point to be improved, the final potential vehicle demand conclusion is formed: "the user's satisfaction with the rear seat space and the trunk space of the vehicle is high, and these two are core advantage demand points, which need to be retained or strengthened in subsequent vehicle models; the user's satisfaction with the driving noise is low, which is a demand point that needs to be improved urgently, and needs to be optimized first." In the embodiment, by extracting product power vocabulary, hierarchical division, word frequency statistics, sentiment analysis and score calculation from user vehicle comments, the potential vehicle demand is finally determined, the whole process from user comment data to precise demand mining is realized, the decision basis for vehicle product optimization is provided, and the efficiency and accuracy of demand mining are improved.

[0046] On the basis of any of the above embodiments, before step S10, further comprising: Obtaining initial comment data of a user on a vehicle; Cleaning the initial comment data to obtain the target comment data; the cleaning process includes duplicate value removal processing.

[0047] It should be noted that automobile vertical media (such as automobile forums, vertical automobile websites, etc.) will preliminarily classify the comments published by users according to evaluation dimensions, and the initial comment data refers to the original comment data directly crawled from the automobile vertical media without filtering, which may include not only basic information such as reputation title, user nickname, vehicle series name, and car dealer, but also various comment contents which have been preliminarily classified by the automobile vertical media, and the specific classification dimensions include but are not limited to space, driving experience, handling, fuel consumption, comfort, appearance, interior, cost performance, intelligence, price policy, etc.

[0048] The cleaning process is to ensure the accuracy of subsequent vocabulary extraction and analysis. The cleaning process refers to the process of removing invalid interference data in the initial comment data to improve data quality. In addition to duplicate value removal processing, the cleaning process can also include noise removal processing, missing value removal processing, etc.

[0049] In the embodiment, by cleaning (such as removing duplicates) the initial comment data, the quality of the target comment data is ensured, and the subsequent analysis is avoided from being disturbed by duplicate data.

[0050] On the basis of any of the above embodiments, step S70 comprises: Based on the hierarchical word frequency statistics result, determine the word frequency weight corresponding to each target vehicle model index; Based on the word frequency weight of each target vehicle model index and the corresponding sentiment tendency evaluation ratio, obtain the sentiment score corresponding to each target vehicle model index.

[0051] It should be noted that by introducing the word frequency weight to correct the sentiment tendency evaluation ratio, the final score can reflect not only the user's emotional attitude but also the attention degree of the index (the higher the word frequency, the greater the influence weight on the final demand judgment), solving the problem that the score calculated based on the sentiment ratio alone ignores the difference in index attention degree, and making the score more consistent with the user's real demand priority.

[0052] Word frequency weight refers to the proportion of the occurrence frequency of a single target vehicle model index in the total occurrence frequency of all target vehicle model indexes, which is used to quantify the user's attention to the index (the higher the word frequency, the greater the weight, and the stronger the influence on the sentiment score).

[0053] In this embodiment, by introducing the word frequency weight, the user's attention to the target vehicle model index is combined with the sentiment tendency evaluation ratio to calculate the sentiment score, so that the score reflects both the attention degree and the emotional attitude, and the score is more in line with the user's real demand priority.

[0054] On the basis of any of the above embodiments, the hierarchical division result includes an evaluation dimension and an evaluation index belonging to the evaluation dimension; The hierarchical word frequency statistics of the product force evaluation vocabulary by using the mining model includes: The mining model is used to count the word frequency of the product force evaluation vocabulary according to the evaluation dimension, and to filter out a target evaluation dimension with a word frequency order meeting a first preset requirement; It should be noted that the evaluation dimension refers to a two-level classification dimension of vehicle product force, which is a macro field division of vehicle evaluation, such as space, driving feeling, comfort, appearance, etc.

[0055] The evaluation index refers to a three-level subdivision index belonging to a certain evaluation dimension, which is a specific attention point of the user under this dimension, such as rear seat space and driver seat space under the space dimension.

[0056] The hierarchical division result is a two-level hierarchical structure of evaluation dimension-evaluation index, such as space-rear seat space and driving feeling-driving smoothness.

[0057] The word frequency statistics according to the evaluation dimension refers to that the mining model classifies the product force evaluation vocabulary according to the evaluation dimension, and counts the total occurrence frequency of all words under each evaluation dimension (reflecting the overall attention of the user to the macro field).

[0058] The first preset requirement refers to a preset dimension filtering rule, such as "selecting the top M positions in descending order of word frequency" (M can be 10), to ensure focusing on the macro field that the user pays most attention to.

[0059] As an example, assume that the hierarchical division result includes evaluation dimensions: space, driving feeling, comfort, appearance, and fuel consumption. The model counts the total word frequency according to the dimension as follows: Space: 2500 times (including the total frequency of words such as "rear seat space" and "driver seat space"); Driving feeling: 1800 times (including the total frequency of words such as "driving smoothness" and "acceleration performance"); Comfort: 1200 times; Appearance: 900 times; Fuel consumption: 600 times.

[0060] Sorted by word frequency in descending order: Space (2500) → Driving experience (1800) → Comfort (1200) → Appearance (900) → Fuel consumption (600).

[0061] If the first preset requirement is "select the first 2", then the target evaluation dimensions are filtered out as: space and driving experience.

[0062] Based on the hierarchical word frequency statistics results, multiple target vehicle indicators are selected, including: For the evaluation indicators belonging to the target evaluation dimension, word frequency statistics are performed, and target evaluation indicators whose word frequency ranking meets the second preset requirements are selected. The target evaluation indicators are then determined as the target vehicle model indicators.

[0063] It should be noted that by first filtering high-attention target evaluation dimensions through word frequency statistics, and then selecting high-attention target evaluation indicators from these dimensions as target vehicle indicators, a progressive analysis was achieved that first focuses on the core dimensions and then delves into the core indicators within those dimensions, thus accurately focusing on the core needs of users.

[0064] Evaluation indicators belonging to the target evaluation dimension refer to the three-level sub-indicators under the selected target evaluation dimension (such as "rear seat space, driver's seat space, and front passenger seat space" under the "space" dimension).

[0065] The second preset requirement refers to the preset indicator selection rules, such as "selecting the top N after sorting by word frequency in descending order," to ensure that the most important detailed indicators within the focused dimension are selected. The first and second preset requirements can be the same or different. For example, the first and second preset requirements can both be to select the top 10 after sorting by word frequency in descending order; or the first preset requirement can be to select the top 2 after sorting by word frequency in descending order, and the second preset requirement can be to select the top 5 after sorting by word frequency in descending order.

[0066] The target evaluation metric, which is the final determined target vehicle model metric, is the sub-metric that users pay the most attention to under the target evaluation dimension. Both the target evaluation metric and the target vehicle model metric must include at least one.

[0067] In this embodiment, by using a hierarchical structure of evaluation dimensions and evaluation indicators, high-attention target evaluation dimensions are first screened, and then target evaluation indicators are screened from them, which improves the targeting and efficiency of target vehicle indicator screening.

[0068] Based on any of the above embodiments, the step of using a mining model to perform hierarchical word frequency statistics on the hierarchical division results to obtain hierarchical word frequency statistics results includes: The hierarchical division result is input into the mining model so that the mining model can count the frequency of each product strength evaluation term in the target review data according to the hierarchical structure of the hierarchical division result, and generate the hierarchical word frequency statistics result based on the frequency of each product strength evaluation term and the corresponding hierarchical information.

[0069] It should be noted that the mining model is a general-purpose large language model with natural language understanding and rule enforcement capabilities, such as the Tongyi Qianwen model. Through the mining model's understanding and execution of hierarchical structures, the association between vocabulary, hierarchy, and frequency is achieved.

[0070] The hierarchical word frequency statistics result refers to structured data that includes "evaluation dimensions, evaluation indicators, product strength evaluation words, and corresponding frequencies", which can intuitively present the number of times words appear under each level.

[0071] Specifically, first, the hierarchical structure of "evaluation dimension - evaluation indicator" and the vocabulary attribution rules are input into the data mining model (e.g., under the "space" dimension, words like "large rear seat space" belong to the "rear seat seating space" indicator). The model learns and clarifies the correspondence between vocabulary and hierarchy. Next, the model loads the target review data, iterates through and extracts product strength evaluation vocabulary, identifies the evaluation dimension and evaluation indicator to which each word belongs based on the learned rules, and accumulates the frequency of each word (e.g., "large rear seat space" and "not crowded in the back" are both included in the frequency of "space → rear seat seating space"). Then, the model summarizes the statistical results and organizes them into structured data according to the hierarchy of "evaluation dimension → evaluation indicator → specific vocabulary," which includes the frequency of individual words and can also summarize the total frequency of each evaluation indicator (e.g., the total frequency of "rear seat seating space" is 1113 times).

[0072] In this embodiment, the process of mining the model to statistically analyze the frequency of product power terms according to the hierarchical structure and generate structured results is clearly defined, which ensures the correlation between term frequency and hierarchy and provides structured and directly usable quantitative data for subsequent screening and analysis.

[0073] Based on any of the above embodiments, sentiment tendencies include positive, negative, and neutral; the step of using the mining model to perform sentiment tendency analysis on the comment data related to the target vehicle model indicators in the target comment data, and determining the sentiment tendency evaluation ratio corresponding to each target vehicle model indicator, includes: The mining model is used to identify target word combinations in the comment data to obtain the sentiment tendency determination result for each comment data; the target word combinations include sentiment words and degree adverbs, and the sentiment tendency determination result includes positive, negative, or neutral results; For each target vehicle model indicator, the sentiment tendency evaluation ratio is obtained based on the proportion of the number of comment data for each sentiment tendency to the total number of comment data corresponding to the target vehicle model indicator.

[0074] It should be noted that sentiment refers to the classification of users' attitudes toward the target vehicle model indicators, which can include three categories: positive (satisfaction, approval, etc.), negative (dissatisfaction, criticism, etc.), and neutral (no clear attitude or objective description).

[0075] The mining model can identify parts of speech such as sentiment words and degree adverbs in text, as well as their combinations.

[0076] Target vocabulary combinations refer to word combinations used in comments to express emotions, such as the combination of the emotion word "very satisfied" and an adverb of degree. Target vocabulary combinations include, but are not limited to, emotion words and adverbs of degree, and may also include negative words, etc.

[0077] Emotional words refer to words that directly express emotional attitudes (positive such as "good, satisfied, spacious"; negative such as "poor, dissatisfied, narrow"; neutral such as "average, ordinary").

[0078] Adverbs of degree are words used to modify the intensity of emotional words (such as "very", "very", "slightly", "slightly", etc.).

[0079] The sentiment assessment result refers to the classification conclusion made by the mining model on the sentiment attitude of a single comment, namely, a positive result, a negative result, or a neutral result.

[0080] As an example, the model loads comment data related to the target vehicle's specifications, analyzes the text sentence by sentence using natural language understanding, identifies combinations of sentiment words and degree adverbs, and determines the sentiment tendency of a single comment by combining the intensity and direction of the emotions expressed in the combination.

[0081] Taking comments related to the target model's indicator "rear passenger space" as an example: Comment 1: "The back seat is very spacious, and it's comfortable for three people to sit in it" → Identify the target word combination "very spacious" and "very comfortable" ("spacious" and "comfortable" are positive sentiment words, and "very" and "very" are adverbs of intensification) → Determine it as a positive result; Comment 2: "The rear seat space is a bit cramped, and it's tiring to sit for a long time" → Identify the target word combination "a bit cramped" and "tiring" ("cramped" and "tired" are negative sentiment words, and "a bit" and "somewhat" are adverbs of weakening degree) → Determine it as a negative result; Comment 3: "Rear seat space is moderate, neither large nor small" → Identify the target word combination "moderate" and "neither large nor small" (no explicit emotional words, only objective description) → Determine as a neutral result.

[0082] Next, the total number of relevant comments for a single target vehicle model indicator (i.e., the total number of all comments mentioning that indicator) is counted; the number of positive, negative, and neutral comments under that indicator is counted separately; the proportion of each sentiment type to the total number is calculated, thus obtaining the sentiment rating ratio. Taking the "rear seat space" indicator as an example: Total number of relevant comments: 822; Number of positive comments: 707; Number of negative comments: 27; Number of neutral comments: 88. The percentage of sentiment assessment is: Positive proportion = 707 ÷ 822 ≈ 86% Negative proportion = 27 ÷ 822 ≈ 3.3% Neutral ratio = 88 ÷ 822 ≈ 10.7%.

[0083] In specific implementations, such as Figure 2 As shown, Figure 2 The flowchart for user review sentiment analysis provided in this application embodiment firstly calculates the word frequency of the secondary category of product strength evaluation from user review data. Then, it calculates the word frequency of the tertiary indicators belonging to the top 10 secondary categories, and selects the top 10 indicators that users care about most. Then, it uses a mining model to analyze the relevant comments of these 10 indicators, identifies the target word combinations (sentiment words and degree adverbs), and outputs the sentiment tendency judgment result (positive, negative, or neutral) of each comment based on the sentiment direction and intensity expressed by the combination. Finally, it calculates the proportion of positive, negative, and neutral evaluations.

[0084] In this embodiment, the sentiment tendency of a single comment is determined by identifying the combination of sentiment words and degree adverbs, and then the sentiment tendency evaluation ratio is obtained by statistical analysis, thus quantifying the sentiment distribution of users towards the target car model indicators.

[0085] Based on any of the above embodiments, determining the user's potential vehicle needs based on the emotion score includes: If the emotional score exceeds the first threshold, the target vehicle model indicator corresponding to the emotional score will be identified as the key demand point that satisfies the user. If the emotional score does not exceed the second threshold, the target vehicle indicator corresponding to the emotional score is identified as the user's unsatisfactory improvement requirement; the first threshold is higher than the second threshold. Based on the aforementioned advantageous needs and the needs to be improved, the potential vehicle needs are determined.

[0086] It should be noted that the first threshold refers to a preset positive judgment threshold (such as 0.5), which is used to define the standard for high user satisfaction. Indicators with emotional scores exceeding this value are considered to be the core advantages recognized by the user.

[0087] The second threshold refers to a preset negative judgment threshold (e.g., -0.5), used to define the standard for high user dissatisfaction. Indicators with an emotional score not exceeding this value are considered defects that need to be prioritized for improvement. The positive judgment standard (e.g., 0.5) is numerically greater than the negative judgment standard (e.g., -0.5). Optionally, in addition to advantageous and unimproved needs, neutral needs can also be included. Neutral needs refer to target model indicators for which users have unclear or neutral emotional attitudes, neither achieving significant satisfaction nor significant dissatisfaction. Target model indicators located between the first and second thresholds (i.e., emotional scores greater than the second threshold and less than the first threshold) are considered neutral needs.

[0088] Advantage demand points refer to target vehicle indicators whose emotional scores exceed the first threshold, reflecting high user satisfaction with the indicator and representing the core competitiveness of the vehicle, which should be retained or strengthened in subsequent models.

[0089] The areas for improvement refer to the target vehicle indicators whose emotional scores do not exceed the second threshold. This reflects high user dissatisfaction with these indicators and represents a weakness in the vehicle, requiring targeted optimization in future models.

[0090] As an example, the emotional score of a target vehicle indicator is compared with a first threshold (e.g., 0.5). If the score is greater than the first threshold, it is considered a strong demand. For instance, the emotional score for the target vehicle indicator "luggage space" is 0.62 (>0.5), indicating high user satisfaction and the need to retain the current design in future models. The emotional score of a target vehicle indicator is then compared with a second threshold (e.g., -0.5). If the score is less than or equal to the second threshold, it is considered a demand requiring improvement. For instance, the emotional score for the target vehicle indicator "driving noise" is -0.553 (≤-0.5), indicating significant user dissatisfaction and the need to address it. If the emotional score of a target vehicle indicator satisfies the condition that the second threshold < the score < the first threshold (e.g., -0.5 < the score < 0.5), it is considered a neutral demand. For instance, the emotional score for the target vehicle indicator "passenger seat space" is 0.3 (-0.5 < 0.3 < 0.5), indicating no strong user demand for it. By integrating strengths, areas for improvement, and neutral needs, a list of potential needs is formed. For example, potential vehicle needs include: strengths: trunk space (retain the design); areas for improvement: driving noise (optimize sound insulation); neutral needs: passenger seat space (maintain the status quo).

[0091] In this embodiment, by setting first and second thresholds, the emotional score is transformed into advantageous needs (user satisfaction) and needs for improvement (user dissatisfaction), which clarifies the strengths and weaknesses of the vehicle product and provides a clear decision-making direction for vehicle model optimization.

[0092] Furthermore, addressing issues such as the disconnect between automakers' model definitions and user needs, inaccurate and incomplete tag mining, and a lack of dynamic market responsiveness, this application, based on the same inventive concept as the methods described in any of the above embodiments, also provides a method for mining popular car model tags based on multi-level word frequency statistics and sentiment analysis. This method utilizes multi-level word frequency statistics and sentiment analysis techniques to collect and analyze text data from multiple channels, accurately grasp user needs, comprehensively discover popular car model tags, and capture market dynamics in real time. Its effects are significant, helping automakers create models that are more aligned with the market and enhance product market competitiveness.

[0093] Reference Figure 3 , Figure 3 A mining principle block diagram provided in this application embodiment, the method includes: 1. Data Acquisition Layer: Crawling user review data of popular car models from automotive vertical media, cleaning the raw reviews to remove noise, duplicates, missing values ​​and other interfering factors, improving data quality, and providing a reliable data foundation for subsequent sentiment analysis and tag mining; 2. Multi-level word frequency statistics: Based on the characteristics of the automotive industry and common content in user reviews, product performance evaluation terms are hierarchically categorized. Then, based on a large language model, the frequency of secondary category terms in product performance evaluation is statistically analyzed, and the top 10 secondary categories are selected. Word frequency statistics are then performed on the tertiary categories belonging to the top 10 secondary categories, and the top 10 indicators most concerned by users are selected. These indicators represent the content that users care about most. Through multi-level word frequency statistics, while reducing computational power consumption, it is possible to deeply explore the specific indicators that users focus on in each product performance area. 3. Sentiment Analysis Layer: Based on a large language model, sentiment analysis is performed on the top 10 indicators that users care about most. Sentiment judgment results are extracted from sentences involving relevant indicators, the sentiment judgment results of all relevant sentences are aggregated, and the positive, negative and neutral evaluation ratios of each indicator are calculated. 4. Weighting and Statistics Layer: Based on the three-level word frequency statistics, word frequency weights are calculated to provide precise weight assignments for product performance evaluation. The sentiment weights of all sentences under each product performance indicator are accumulated to obtain the sentiment score for that indicator, thus more comprehensively reflecting the actual evaluation of the product in the minds of users. By comparing the scores of different competitors on product performance indicators and user sentiment tendencies, the gaps and advantages of the product compared to competitors can be identified, providing decision support for market competition.

[0094] In a specific implementation, let's take a certain car model as an example: During the data collection phase, some user reviews about a certain car model were crawled from channels such as car forums and vertical automotive websites. Here are some sample reviews: "So far, I'm quite satisfied and hope the system can be upgraded and optimized."; "The space in both the front and rear seats and the trunk is very large, and the rear seats can be folded down to be used as a bed."; "The steering wheel is very sensitive, and the power response is excellent. It's responsive and easy to overtake on the highway."; "Acceleration is quite fast, especially when starting, but I don't know how it performs on the highway yet."; "Considering the weather these past few days, the range is quite good. During the 618 shopping festival, the range was a little over 500 kilometers."

[0095] In the multi-level word frequency statistics phase, the secondary categories related to riding experience in the 1857 crawled reviews were statistically analyzed. These secondary categories included driver's seat, passenger seat, rear seats, air conditioning, quietness, driving, passenger and alighting space, passenger space, storage, operation, cleaning and maintenance, visibility, health, environmental protection, hazards, collision, battery, information, and functional safety. The frequency of each category was also analyzed. (Refer to...) Figure 4 , Figure 4 This application provides a secondary classification word frequency statistical histogram, based on... Figure 4 The word frequency statistics shown are based on the top 10 secondary categories: driving, passenger space, rear seats, storage, air conditioning, driver's seat, operation, passenger seat, quietness, and battery. Word frequency statistics were then performed on the tertiary indicators (driving smoothness, driving noise, driver's seat passenger space, passenger seat passenger space, rear seat passenger space, etc.) belonging to these categories.

[0096] In the sentiment analysis phase, refer to Figure 5 , Figure 5 This application provides a three-level classification word frequency statistical histogram, based on... Figure 5 The word frequency ranking shown selects the top 10 indicators most concerned by users: rear seat space, trunk space, driver's seat space, rear seat comfort, driver's seat comfort, quietness and comfort performance, driving noise, driving smoothness, passenger seat space, and seat operation. Sentiment analysis is performed on the comments under these indicators to determine whether they are positive, negative, or neutral, and the proportion of each sentiment tendency is calculated.

[0097] In the weighting and statistical phase, an analysis of the frequency of keywords related to the riding experience of a certain vehicle model revealed that rear seat space was mentioned most frequently (1113 times), followed by driver's seat space and rear seat comfort (648 and 631 times respectively), reflecting that these are key considerations for users' riding experience. Regarding storage, the trunk was mentioned frequently (747 times), indicating some interest in driver and passenger seat storage. In air conditioning-related mentions, front air conditioning and air conditioning functions were mentioned slightly more often. Regarding seat comfort and functions, driver's seat comfort was mentioned more frequently than passenger seat comfort. Additionally, quietness and comfort were mentioned frequently, indicating that users have certain requirements for cabin quietness. Battery protection and battery heat were mentioned relatively infrequently, suggesting that users pay relatively less attention to these two aspects. Overall, users' focus on the riding experience of this particular vehicle model is concentrated on space, storage, air conditioning, seat comfort, and quietness.

[0098] Reference Figure 6 , Figure 6 This application provides a visualization chart for sentiment analysis based on indicators. Figure 6 According to the sentiment analysis statistics of a certain car model, its rear passenger space, trunk space, and driver's seat space all received a large number of positive user feedback, showing a clear advantage. Although the overall rear passenger comfort and driver's seat comfort are positive, the proportion of negative reviews also indicates that there is room for improvement. User reviews on quiet comfort performance and seat operation are mixed, with no absolute trend. However, driving noise is a prominent issue, with a large number of negative reviews. Driving smoothness is acceptable, but further optimization is needed to reduce user dissatisfaction. Although the passenger seat space is recognized, the number of positive reviews is slightly lower than other advantages.

[0099] In this embodiment, by leveraging multi-level word frequency statistics and sentiment analysis technology, user review data is crawled and cleaned from automotive vertical media through the data acquisition layer. The product strength evaluation words are hierarchically divided and word frequency is statistically analyzed to accurately identify user attention indicators. By using a large language model to analyze the sentiment tendency of these indicators and statistically analyze sentiment scores, user needs are comprehensively grasped, popular model tags are accurately discovered, and market dynamics are captured in real time.

[0100] Based on the methods described in any of the above embodiments, this application also provides, as follows: Figure 7 The diagram shows the structure of an electronic device. Figure 7 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to implement the methods described in any of the above embodiments.

[0101] Based on the methods described in any of the above embodiments, this application also provides a computer storage medium storing a computer program, which, when executed by a processor, can be used to perform the methods described in any of the above embodiments.

[0102] Based on the methods described in any of the above embodiments, this application also provides a computer program product, which includes one or more computer programs or instructions. The computer program or instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. When executed by a processor, the computer program implements the methods described in any of the above embodiments.

[0103] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0104] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0105] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0106] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0108] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A method for mining potential vehicle demand, characterized in that, The method includes: Obtain user-defined review data for vehicles; Extract product performance evaluation terms from the target review data; The product strength evaluation terms are hierarchically divided to obtain the hierarchical division results; The hierarchical word frequency statistics are obtained by using a mining model to analyze the hierarchical division results. Based on the hierarchical word frequency statistics, multiple target vehicle indicators were selected; The mining model is used to perform sentiment analysis on the comment data related to the target vehicle model indicators in the target comment data to determine the sentiment evaluation ratio corresponding to each target vehicle model indicator; Based on the aforementioned sentiment tendency evaluation ratio, determine the sentiment score corresponding to each of the target vehicle indicators; Based on the emotional score, the user's potential vehicle needs are determined.

2. The method according to claim 1, characterized in that, Before obtaining the user's target review data for the vehicle, the following is also included: Obtain initial user review data for the vehicle; The initial comment data is cleaned to obtain the target comment data; the cleaning process includes deduplication.

3. The method according to claim 1, characterized in that, The step of determining the emotional score corresponding to each target vehicle model indicator based on the emotional tendency evaluation ratio includes: Based on the hierarchical word frequency statistics results, determine the word frequency weight corresponding to each of the target vehicle indicators; Based on the word frequency weight of each target vehicle indicator and the corresponding sentiment tendency evaluation ratio, the sentiment score corresponding to each target vehicle indicator is obtained.

4. The method according to claim 1, characterized in that, The hierarchical division results include evaluation dimensions and evaluation indicators belonging to the evaluation dimensions; The method of using a mining model to perform hierarchical word frequency statistics on the product strength evaluation vocabulary yields hierarchical word frequency statistics results, including: The product strength evaluation vocabulary is statistically analyzed by evaluation dimension using a mining model, and the target evaluation dimension whose word frequency ranking meets the first preset requirement is selected. Based on the hierarchical word frequency statistics results, multiple target vehicle indicators are selected, including: For the evaluation indicators belonging to the target evaluation dimension, word frequency statistics are performed, and target evaluation indicators whose word frequency ranking meets the second preset requirements are selected. The target evaluation indicators are then determined as the target vehicle model indicators.

5. The method according to claim 1, characterized in that, The step of using a mining model to perform hierarchical word frequency statistics on the hierarchical division results yields hierarchical word frequency statistics results, including: The hierarchical division result is input into the mining model so that the mining model can count the frequency of each product strength evaluation term in the target review data according to the hierarchical structure of the hierarchical division result, and generate the hierarchical word frequency statistics result based on the frequency of each product strength evaluation term and the corresponding hierarchical information.

6. The method according to claim 1, characterized in that, Sentiment bias includes positive, negative, and neutral; the process of using the mining model to perform sentiment bias analysis on the comment data related to the target vehicle model indicators in the target comment data, and determining the sentiment bias evaluation ratio corresponding to each target vehicle model indicator, includes: The mining model is used to identify target word combinations in the comment data to obtain the sentiment tendency determination result for each comment data; the target word combinations include sentiment words and degree adverbs, and the sentiment tendency determination result includes positive, negative, or neutral results; For each target vehicle model indicator, the sentiment tendency evaluation ratio is obtained based on the proportion of the number of comment data for each sentiment tendency to the total number of comment data corresponding to the target vehicle model indicator.

7. The method according to claim 1, characterized in that, The process of determining a user's potential vehicle needs based on the sentiment score includes: If the emotional score exceeds the first threshold, the target vehicle model indicator corresponding to the emotional score will be identified as the key demand point that satisfies the user. If the emotional score does not exceed the second threshold, the target vehicle indicator corresponding to the emotional score is identified as the user's unsatisfactory improvement requirement; the first threshold is higher than the second threshold. Based on the aforementioned advantageous needs and the needs to be improved, the potential vehicle needs are determined.

8. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; Wherein, when the processor invokes the executable instructions, it implements the method described in any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, It stores computer instructions that, when executed by a processor, implement the steps of any of the methods described in claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images