E-commerce demand analysis method based on large language model

Abstract

The invention provides an e-commerce demand analysis method and system based on a large language model. Product characteristics and market strategies are optimized by analyzing user generated content (UGC). The method comprises the following steps: firstly, collecting UGC data such as commodity comments and user questions and answers from an e-commerce platform; then, performing deep analysis on the data by utilizing a large language model, extracting product attributes and emotion information, and converting the product attributes and the emotion information into structured data; then, generating a market-level dual importance degree matrix, and evaluating the importance of the product attribute on the influence of the purchase decision and the satisfaction degree after purchase; further, for merchant products, actual performance values of attributes are calculated, and a product level dual importance-performance matrix is generated to identify optimization requirements. And finally, the analysis result is visually displayed, a report is generated, and market insight and decision support are provided for merchants. According to the method, merchants are helped to accurately position the product optimization direction, and the market competitiveness is improved.

Description

Technical Field

[0001] The present invention relates to e-commerce big data technology and natural language processing (NLP) technology, and in particular to an e-commerce demand analysis method based on a large language model. Background Art

[0002] In today's fierce market competition environment, insight into user needs has become the core of enterprises to enhance the competitiveness of products and services. Especially in the field of e-commerce, user-generated content (UGC) as an important channel for demand feedback is increasingly valued by enterprises. Compared with traditional offline surveys, user-generated content has the advantages of wide coverage, less time constraints and strong real-time performance, and can more intuitively reflect consumers' actual needs, improvement suggestions and future expectations. The comments and feedback information generated by users on e-commerce platforms can provide valuable insights into products and services, especially in product design and optimization, customer experience improvement and other aspects.

[0003] With the continuous development of big data technology and natural language processing (NLP) technology, enterprises can extract valuable demand information from massive user-generated content, thereby providing data support for product innovation and service optimization. However, how to effectively explore user needs, evaluate their priorities, and accurately interpret consumer attitudes remains a difficult problem that needs to be solved. In response to this problem, how to help enterprises conduct demand analysis efficiently and accurately through innovative methodologies and technical means has become an important topic in theoretical research and practical exploration.

[0004] In recent years, natural language processing technology based on large language models has developed rapidly and has become a powerful tool for processing various language tasks. With its powerful context understanding ability, large language models are widely used in many fields such as text classification, sentiment analysis, and information extraction, showing strong language processing capabilities. In the field of e-commerce, text analysis technology based on large language models can accurately extract key information from complex user-generated content. Through deep learning technology, large language models can capture subtle language features in large-scale data, identify potential patterns in text, and help companies automatically extract user needs from user-generated content such as comments and questions and answers. Based on this technology, companies can quickly obtain user demand dynamics from data and optimize products and services accordingly.

[0005] In terms of demand mining based on user-generated content, existing technical solutions mainly revolve around two core steps: comment data processing and demand mining analysis, that is, first processing the comment data into structured data, and then presenting the demand analysis results through statistical analysis methods.

[0006] In terms of comment data processing, currently, research on product demand mining based on online reviews mainly relies on keyword extraction techniques and sentiment analysis methods. Common keyword extraction methods include TF-IDF and LDA topic models, which extract key information in reviews by counting word frequencies or mining latent topics. In addition, some researchers also use deep learning models such as BERT, leveraging pre-trained language models for sentiment analysis and information extraction, and then extracting user sentiment and concerns.

[0007] In terms of demand mining and analysis, current research mainly relies on the Kano model and IPA (Importance-Performance Analysis) method to evaluate the impact of product attributes on user satisfaction and their importance. The Kano model divides product features into basic, expected, and exciting types through user feedback, thereby analyzing the role of these features in satisfaction. The IPA method, on the other hand, helps enterprises identify the most valuable features for consumers by comparing the performance of product attributes with user expectations, in order to optimize resource allocation.

[0008] Traditional views usually regard customer satisfaction as a linear function of product or service quality, that is, the higher the perceived quality, the higher the satisfaction. However, simply improving product attributes may not necessarily effectively enhance customer satisfaction, because the essence of customer needs may not be a linear or symmetric relationship. Academic research shows that there may be non-linear or asymmetric characteristics between attribute performance and satisfaction. Kano et al. proposed the Kano model in 1984, which is specifically used to analyze this relationship and classify demands. Although the Kano model itself does not directly measure the importance of attributes, it can provide a basis for prioritization according to the degree of influence of attributes on satisfaction or dissatisfaction.

[0009] The Kano model classifies product attributes into five categories (see Figure 1 ): exciting demands (A), expected demands (O), basic demands (M), indifferent demands (I), and reverse demands (R), corresponding to different satisfaction influence mechanisms respectively. Exciting demands refer to features that exceed user expectations. Their presence can greatly enhance satisfaction, while their absence will not cause dissatisfaction. Expected demands are positively correlated with satisfaction. Users have clear expectations for them, and the better the performance, the higher the satisfaction. Basic demands are the basic expectations of users. Although their good performance will not significantly enhance satisfaction, their absence will lead to strong dissatisfaction. Indifferent demands have little impact on satisfaction, and users usually have a neutral attitude towards their presence or absence. Reverse demands refer to the negative attitude of users towards the presence of certain features, and such demands may reduce satisfaction.

[0010] The IPA method (Importance-Performance Analysis) was proposed by Martilla and James in 1977 as a tool for classifying product or service attributes and determining optimization priorities. This method constructs a two-dimensional matrix with the horizontal axis representing product performance and the vertical axis representing the importance perceived by users, and divides each attribute into different quadrants (see Figure 2 ). In the Q1 (Continue to maintain) quadrant, attributes with high importance and high performance are the core advantages that enterprises should maintain, and their stability should be ensured to avoid performance decline. In the Q2 (Priority improvement) quadrant, attributes with high importance but low performance need to be optimized first. These factors are crucial to users, but the current performance fails to meet expectations. Therefore, resources need to be invested for improvement to enhance customer satisfaction. In the Q3 (Insignificant) quadrant, attributes with low importance and low performance have limited impact on customers. Resources can be preferentially allocated to more critical parts rather than focusing on improving these features. Finally, the Q4 (Appropriately reduce) quadrant contains attributes with low importance but good performance. Although these features perform well, they receive low user attention. Therefore, investment can be reduced to avoid waste of resources.

[0011] Current methods for mining product requirements based on online reviews mainly include two core parts: review data processing and requirement mining analysis. However, this process has the following significant defects, which limit merchants' accurate understanding of user requirements:

[0012] 1. Incomplete processing of structured review data

[0013] Existing methods mostly use keyword extraction techniques (such as TF-IDF, LDA topic models, etc.) for review analysis. Although these methods have low computational costs and can quickly process large-scale review data, they have obvious limitations. Especially in a market environment with increasing product homogenization, relying solely on keyword frequencies for analysis is likely to overlook low-frequency but key product features, and these low-frequency features are often important bases for product differential competition. Therefore, traditional methods may lead to incomplete demand identification, which in turn affects the product optimization direction.

[0014] 2. Error accumulation in sentiment analysis

[0015] Existing sentiment analysis methods usually adopt the pipeline method, that is, first extract aspect words and then perform sentiment classification separately. However, this step-by-step processing method may generate errors in each link and gradually amplify in subsequent processing, ultimately affecting the accuracy of overall demand analysis. For example, if the aspect word extraction stage fails to accurately identify the key attributes in user reviews, the subsequent sentiment analysis results may also be biased, leading to misinterpretation of user requirements by merchants.

[0016] 3. Single perspective, ignoring pre-purchase demand analysis

[0017] Current demand analysis methods mainly rely on user evaluations after purchase, while ignoring consumers' concerns and decision-making factors before purchase. However, in actual online shopping scenarios, consumers' purchase decisions are influenced by the expectancy theory, that is, expectations are formed based on product descriptions, user evaluations, etc. before purchase, and post-purchase evaluations more reflect the gap between expectations and actual experiences. Therefore, conducting demand analysis solely based on post-purchase reviews may not fully depict consumers' true needs and may even lead merchants to focus on the wrong direction during product optimization. For example, some key attributes may be core concerns of consumers before purchase, but if the performance of this attribute meets expectations, consumers may not mention it in the reviews, resulting in post-purchase reviews being unable to comprehensively reflect the true needs. Therefore, the lack of analysis of pre-purchase concerns limits merchants' in-depth understanding of consumers' needs.

[0018] It should be noted that the information disclosed in the above background art section is only for understanding the background of the present application, and thus may include information that does not constitute prior art known to those of ordinary skill in the art. Summary of the Invention

[0019] The main objective of the present invention is to overcome the defects existing in the above background art, and provide an e-commerce demand analysis method based on a large language model, which optimizes the accuracy and comprehensiveness of demand analysis by introducing dual perspectives before and after purchase.

[0020] To achieve the above objective, the present invention adopts the following technical solutions:

[0021] An e-commerce demand analysis method based on a large language model, comprising the following steps:

[0022] S1. Data collection: Collect user-generated content (UGC) of the merchant's own products and target competitors from e-commerce platforms, including product review data and user Q&A data;

[0023] S2. Data processing and analysis: Use a large language model to process and analyze the collected user-generated content, extract product attributes and their sentiment information therefrom, and convert the extracted information into structured data;

[0024] S3. Generation of a dual importance matrix at the market level: Based on the extracted structured data, calculate the pre-purchase attention and post-purchase importance of each product attribute respectively, and generate a dual importance matrix at the market level by combining the attention and importance, for evaluating the relative importance of different product attributes in consumers' purchase decisions and satisfaction;

[0025] S4. Generation of the dual importance - performance matrix at the product level: Combining the dual importance matrix at the market level, for the merchant's own products, calculate the actual performance values of each product attribute, and generate the dual importance - performance matrix at the product level for identifying product features that need to be optimized;

[0026] S5. Result presentation and report generation: Present the generated dual importance matrix and dual importance - performance matrix in a visual way, and generate an analysis report to provide market insights and decision - making support for product optimization for the merchant.

[0027] Further, in the step S2, when using the large - language model to process and analyze user - generated content, it specifically includes the following steps:

[0028] Pre - define the attribute list of product categories;

[0029] For each comment, use a preset prompt template to guide the large - language model to identify the attributes involved in the comment and their corresponding sentiment polarities;

[0030] Convert the identified attributes and their sentiment information into structured data.

[0031] Further, in the step S3, when calculating the pre - purchase attention of each product attribute, it specifically includes the following steps:

[0032] Analyze the user question data in the user Q&A module, and count the mention frequency of each attribute;

[0033] Calculate the pre - purchase attention of each attribute based on the mention frequency.

[0034] Further, in the step S3, when calculating the post - purchase importance of each product attribute, it specifically includes the following steps:

[0035] Based on the comment data, adopt the conjoint analysis method, and through multiple regression analysis modeling, calculate the influence degree of each attribute on customer satisfaction;

[0036] Determine the post - purchase importance of each attribute according to the difference of regression coefficients.

[0037] Further, in the step S3, when generating the dual importance matrix at the market level, it specifically includes the following steps:

[0038] Normalize the pre - purchase attention and post - purchase importance;

[0039] Taking the post - purchase importance as the horizontal axis and the pre - purchase attention as the vertical axis, draw the dual importance matrix;

[0040] Divide the four quadrants according to the mean value in the matrix, which respectively correspond to core features, focus features, low-value features, and potential optimization features.

[0041] Further, in the step S4, when calculating the actual performance value of each product attribute, it specifically includes the following steps:

[0042] Based on the review data, calculate the proportion of positive sentiment of each attribute as the actual performance value of this attribute.

[0043] Further, in the step S4, generating the dual importance - performance matrix at the product level specifically includes the following steps:

[0044] Determine the mean value of the importance before purchase and the importance after purchase of each attribute as the benchmark for evaluating the relative importance of the attribute; use the conditional judgment mechanism to determine the weight adjustment factor of each attribute, and the weight adjustment factor is dynamically adjusted based on the comparison results of the importance before purchase and the importance after purchase of the attribute with their respective mean values; based on the weighted calculation of the weight adjustment factor, the importance before purchase, and the importance after purchase of each attribute, obtain the dual importance of each attribute;

[0045] Taking the dual importance of the attribute as the vertical axis and the actual performance value as the horizontal axis, draw the dual importance - performance matrix at the product level; among them, divide the dual importance - performance matrix into five modules, and the characteristics of each module are as follows:

[0046] The attributes included in the first module have high dual importance and high actual performance value, indicating that these attributes have a great impact on consumer decisions and perform well, and should be continued to be maintained;

[0047] The attributes included in the second module have high dual importance but low actual performance value, indicating that these attributes have a great impact on consumer decisions but perform poorly, and need to be improved first;

[0048] The attributes included in the third module have relatively high dual importance but low actual performance value, indicating that these attributes have a certain impact on consumers but the performance does not meet expectations, and can be maintained for low-priority optimization;

[0049] The attributes included in the fourth module have low dual importance and low actual performance value, indicating that these attributes have little impact on consumer decisions and the performance is average, and can be maintained as unimportant;

[0050] The attributes included in the fifth module have low dual importance but high actual performance value, indicating that although these attributes have little impact on consumer decisions, they perform excellently, and the investment can be appropriately reduced.

[0051] Further, in step S4, the dual importance of each attribute is the dual importance calculated by weighted calculation based on the pre-purchase attention, the post-purchase importance, and the weight adjustment factor adjusted based on condition judgment; the specific condition judgment mechanism is as follows: when both the pre-purchase importance and the post-purchase importance of the attribute are higher than their means, a higher weight adjustment factor is given; when either importance is higher than its mean, a medium weight adjustment factor is given; when both are lower than their means, a lower weight adjustment factor is given.

[0052] Further, in step S5, when presenting the results and generating the report, the following steps are specifically included:

[0053] Display the dual importance matrix at the market level and the dual importance - performance matrix at the product level through a visualization tool;

[0054] Generate an analysis report to provide market insights and decision-making support for product optimization.

[0055] An e-commerce demand analysis system based on a large language model includes the following steps:

[0056] Data collection module: Collect user-generated content (UGC) of merchants' own products and target competitors from e-commerce platforms, including product review data and user Q&A data;

[0057] Data processing and analysis module: Use a large language model to process and analyze the collected user-generated content, extract product attributes and their sentiment information therein, and convert the extracted information into structured data;

[0058] Dual importance matrix generation module at the market level: Based on the extracted structured data, calculate the pre-purchase attention and post-purchase importance of each product attribute respectively, and generate a dual importance matrix at the market level in combination with the attention and importance, which is used to evaluate the relative importance of different product attributes in consumers' purchase decisions and satisfaction;

[0059] Dual importance - performance matrix generation module at the product level: Combine the dual importance matrix at the market level, calculate the actual performance values of each product attribute for the merchants' own products, and generate a dual importance - performance matrix at the product level, which is used to identify product features that need to be optimized;

[0060] Result presentation and report generation module: Present the generated dual importance matrix and dual importance - performance matrix in a visual way, and generate an analysis report to provide market insights and decision-making support for product optimization for merchants.

[0061] The present invention has the following beneficial effects:

[0062] The present invention provides an e-commerce demand analysis method and system based on large language models, breaking through the limitations of traditional methods that only rely on post-purchase feedback. Starting from the dual perspectives before and after consumers' purchases, it comprehensively identifies and analyzes key demands, significantly improving the accuracy of decision-making support. The system collects product review data of the merchant's own products and target competitors, as well as user Q&A data from the "Ask Everyone" module, and performs text analysis based on large language models to extract product attributes and their sentiment information, and converts them into structured data. Then, the system calculates the attention of users to each attribute before purchase, and evaluates the impact of each attribute on customer satisfaction, generating a dual importance matrix at the market level to reveal the role of each attribute in consumers' purchase decisions and post-purchase satisfaction. At the same time, based on the review data, the system calculates the actual performance of product attributes, combines the dual importance indicators, and generates a dual importance-performance matrix at the product level to accurately locate the product features that need to be optimized.

[0063] The e-commerce demand analysis method and system based on large language models proposed by the present invention optimize the accuracy and comprehensiveness of demand analysis by introducing dual perspectives before and after purchase. Specifically:

[0064] 1. Improve the accuracy of structured processing

[0065] The present invention uses large language models (such as GPT, BERT) to perform in-depth semantic understanding and processing of user-generated content (UGC), avoiding the problem of error accumulation in traditional pipeline methods. Large language models can more accurately extract product attributes and their sentiment information in reviews, and can effectively identify low-frequency but important features, thereby improving the comprehensiveness and accuracy of demand analysis and providing higher-quality data support for merchants.

[0066] 2. Construct a demand analysis framework from dual perspectives

[0067] The present invention innovatively proposes a demand analysis method that combines perspectives before and after purchase, comprehensively revealing the behavioral differences of consumers in the purchase decision-making stage and the use feedback stage. By calculating the pre-purchase attention (based on "Ask Everyone" data) and the impact of post-purchase attributes on user satisfaction (based on review sentiment analysis), a dual importance matrix at the market level is constructed to quantify the impact of different product attributes on consumers' decisions and satisfaction. At the same time, further combined with the actual performance of the product, a dual importance-performance matrix at the product level is generated to help merchants accurately identify which product attributes need to be optimized and provide a scientific optimization direction.

[0068] 3. Provide accurate market insights and optimization strategy support

[0069] The analysis framework of the present invention not only supports demand comparison at the market level, helps enterprises identify key demand characteristics within the industry, provides a clear direction for new product design and market promotion, but also can focus on the relationship between the importance and performance of the attributes of a single product, providing a scientific basis for merchants in terms of resource allocation, product optimization, and formulation of market competition strategies. Through this method, merchants can more accurately understand the dynamic needs of consumers, improve market response capabilities, and enhance the market adaptability and competitiveness of products.

[0070] The present invention overcomes the deficiencies of traditional single-dimensional analysis methods by comprehensively analyzing demands at the market level and the product level and proposing an innovative visualization solution, providing merchants with more comprehensive and accurate market insights, supporting scientific decision-making in product optimization, resource allocation, and market promotion, and enhancing the market adaptability and competitiveness of products.

[0071] Other beneficial effects in the embodiments of the present invention will be further described below. Brief Description of the Drawings

[0072] Figure 1 Schematic diagram of the existing Kano model.

[0073] Figure 2 Schematic diagram of the existing IPA method.

[0074] Figure 3 Block diagram of the e-commerce demand analysis system based on the large language model in the embodiments of the present invention.

[0075] Figure 4 Dual importance matrix at the market level in the embodiments of the present invention.

[0076] Figure 5 Dual importance - performance matrix at the product level in the intermediate process in the embodiments of the present invention.

[0077] Figure 6 Final version of the dual importance - performance matrix at the product level in the embodiments of the present invention. Detailed Description of the Preferred Embodiments

[0078] The following provides a detailed description of the embodiments of the present invention. It should be emphasized that the following description is merely exemplary and not intended to limit the scope of the present invention and its applications.

[0079] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more such features. In the description of the embodiments of the present invention, "a plurality" means two or more unless otherwise specifically defined.

[0080] Existing e-commerce demand analysis methods usually focus on feedback after users' purchases, ignoring consumers' concerns and decision-making processes before purchases. This single-dimensional analysis method often fails to comprehensively reflect consumers' true needs, resulting in merchants having an inaccurate understanding of market trends, thus affecting the effectiveness of product optimization and market promotion. The present invention proposes an innovative method and system for helping merchants conduct demand analysis based on e-commerce user-generated content (UGC). Different from the prior art, the present invention comprehensively considers user needs in two stages, before and after purchase, and combines the powerful semantic understanding ability of large language models to accurately analyze consumers' concerns before purchase and factors affecting post-purchase satisfaction, overcoming the limitation of traditional methods that only rely on post-purchase comments. By constructing a dual importance matrix at the market level and product level, the present invention provides merchants with a more comprehensive and accurate demand analysis, helping merchants not only identify the true feedback after consumers' purchases but also gain insights into potential concerns before purchase.

[0081] Referring to Figures 3 to 6 , an embodiment of the present invention provides an e-commerce demand analysis method based on a large language model, including the following steps:

[0082] Step S1, data collection: Collect user-generated content (UGC) of the merchant's own products and target competitors from the e-commerce platform, including product review data and user Q&A data;

[0083] Step S2, data processing and analysis: Use a large language model to process and analyze the collected user-generated content, extract product attributes and their sentiment information therein, and convert the extracted information into structured data;

[0084] Step S3, generation of the dual importance matrix at the market level: Based on the extracted structured data, calculate the pre-purchase attention (used as the pre-purchase importance) and post-purchase importance of each product attribute respectively, and generate a dual importance matrix at the market level in combination with the attention and importance, for evaluating the relative importance of different product attributes in consumers' purchase decisions and satisfaction;

[0085] Step S4, generation of the dual importance - performance matrix at the product level: Combining the dual importance matrix at the market level, calculate the actual performance values of each product attribute for the merchant's own products, and generate a dual importance - performance matrix at the product level for identifying product features that need to be optimized;

[0086] Step S5, result presentation and report generation: Present the generated dual importance matrix and dual importance - performance matrix in a visual manner, and generate an analysis report to provide merchants with market insights and decision support for product optimization.

[0087] In a preferred embodiment, in step S2, when using a large language model to process and analyze user-generated content, the following steps are specifically included: predefined a list of attributes for product categories; for each comment, use a preset prompt template to guide the large language model to identify the attributes involved in the comment and their corresponding sentiment polarities; convert the identified attributes and their sentiment information into structured data.

[0088] In a preferred embodiment, in step S3, when calculating the pre-purchase attention of each product attribute, the following steps are specifically included: analyze the user question data in the user Q&A module and count the mention frequency of each attribute; calculate the pre-purchase attention of each attribute based on the mention frequency. When calculating the post-purchase importance of each product attribute, the following steps are specifically included: based on the comment data, adopt a conjoint analysis method, and through multiple regression analysis modeling, calculate the influence degree of each attribute on customer satisfaction; determine the post-purchase importance of each attribute according to the difference in regression coefficients. When generating a dual importance matrix at the market level, the following steps are specifically included: normalize the pre-purchase attention and post-purchase importance; use the post-purchase importance as the horizontal axis and the pre-purchase attention as the vertical axis to draw a dual importance matrix; divide the four quadrants according to the mean value in the matrix, corresponding to core features, concern features, low-value features, and potential optimization features respectively.

[0089] In some embodiments, in step S4, when generating a dual importance - performance matrix at the product level, the following steps are specifically included: combine the dual importance matrix at the market level and calculate the weighted value of the dual importance of each attribute; use the weighted value of the dual importance as the vertical axis and the actual performance value as the horizontal axis to draw a dual importance - performance matrix; multiple quadrants can be divided according to the mean value in the matrix to correspond to different product attribute categories or strategy guidance. As an example, such as four quadrants, corresponding to continue to maintain demand, enhance demand, observe demand, and potential demand respectively.

[0090] In a preferred embodiment, in step S4, when calculating the actual performance values of each product attribute, based on the review data, calculate the proportion of positive sentiment for each attribute as the actual performance value of the attribute. Generating the dual importance - performance matrix at the product level specifically includes the following steps: Determine the mean of the pre - purchase importance and the post - purchase importance of each attribute as the benchmark for evaluating the relative importance of the attribute; Use a conditional judgment mechanism to determine the weight adjustment factor for each attribute, and the weight adjustment factor is dynamically adjusted based on the comparison results of the pre - purchase importance and the post - purchase importance of the attribute with their respective means; Based on the weighted calculation of the weight adjustment factor, pre - purchase importance, and post - purchase importance of each attribute, obtain the dual importance of each attribute; Taking the dual importance of the attribute as the vertical axis and the actual performance value as the horizontal axis, draw the dual importance - performance matrix at the product level; Among them, the dual importance - performance matrix is divided into five modules, and the characteristics of each module are as follows: The first module contains attributes with high dual importance and high actual performance values, indicating that these attributes have a great impact on consumer decisions and perform well, and should be continued to be maintained; The second module contains attributes with high dual importance but low actual performance values, indicating that these attributes have a great impact on consumer decisions but perform poorly and need to be improved first; The third module contains attributes with relatively high dual importance but low actual performance values, indicating that these attributes have a certain impact on consumers but do not meet expectations and can be optimized with low priority; The fourth module contains attributes with low dual importance and low actual performance values, indicating that these attributes have little impact on consumer decisions and perform generally and can be considered insignificant; The fifth module contains attributes with low dual importance but high actual performance values, indicating that although these attributes have little impact on consumer decisions, they perform excellently and the investment can be appropriately reduced.

[0091] In a preferred embodiment, the dual importance of each attribute is obtained by weighted calculation based on the pre - purchase attention (pre_importance), post - purchase importance, and the weight adjustment factor (θ) adjusted based on conditional judgment; The specific conditional judgment mechanism is as follows: When both the pre - purchase importance and the post - purchase importance of the attribute are higher than their means, a higher weight adjustment factor (θ) is given to highlight the importance of the attribute; When either importance is higher than its mean, a medium weight adjustment factor (θ) is given to indicate that the attribute is relatively important at a certain stage; When both are lower than their means, a lower weight adjustment factor (θ) is given to indicate that the attribute has a small impact on the overall demand.

[0092] In step S5, when presenting the results and generating the report, it specifically includes the following steps: Display the dual importance matrix at the market level and the dual importance - performance matrix at the product level through a visualization tool; Generate an analysis report to provide market insights and support for product optimization decisions.

[0093] An embodiment of the present invention also provides an e-commerce demand analysis system based on a large language model, including the following steps:

[0094] Data collection module: Collect user-generated content (UGC) of merchants' own products and target competitors from e-commerce platforms, including product review data and user Q&A data;

[0095] Data processing and analysis module: Use a large language model to process and analyze the collected user-generated content, extract product attributes and their sentiment information therefrom, and convert the extracted information into structured data;

[0096] Market-level dual importance matrix generation module: Based on the extracted structured data, calculate the pre-purchase attention and post-purchase importance of each product attribute respectively, and generate a market-level dual importance matrix in combination with the attention and importance, for evaluating the relative importance of different product attributes in consumers' purchase decisions and satisfaction;

[0097] Product-level dual importance-performance matrix generation module: Combine the market-level dual importance matrix, calculate the actual performance values of each product attribute for the merchants' own products, and generate a product-level dual importance-performance matrix for identifying product features that need to be optimized;

[0098] Result presentation and report generation module: Present the generated dual importance matrix and dual importance-performance matrix in a visual manner, and generate an analysis report to provide market insights and product optimization decision support for merchants.

[0099] The following further describes specific embodiments of the present invention.

[0100] System framework

[0101] The technical solution of the present invention includes the following main modules:

[0102] 1. Data collection module

[0103] This module is responsible for obtaining product review data and review stars of merchants' products and target competitors, as well as user Q&A data in the "Ask Everyone" module. By scraping product reviews and user Q&A content from e-commerce platforms, it provides rich data support for subsequent demand analysis.

[0104] 2. Data processing and analysis module

[0105] This module utilizes large language models to deeply process the collected user-generated content (UGC), extracts the product attributes and sentiment information involved in each comment, and simultaneously extracts the attributes involved in the questions in the "Ask Everyone" section, and organizes this information into structured data. This module is the core of the requirements analysis, ensuring the extraction of high-quality and deeply valuable insights from unstructured data.

[0106] 3. Market-level Dual Importance Matrix Generation Module

[0107] This module combines the attention and satisfaction levels of users before and after purchase, and based on the data of the merchant's own products and similar competing products, draws a market-level dual importance matrix. The specific steps are as follows:

[0108] Calculation of pre-purchase attention: By analyzing the user questions in the "Ask Everyone" section, measure the attention of users to each attribute before purchase (taking Taobao and JD.com platforms as examples).

[0109] Calculation of post-purchase importance: Based on the review data and review stars, use conjoint analysis methods to identify the impact degree of each attribute on overall customer satisfaction.

[0110] Drawing of the market-level dual importance matrix: The horizontal axis of this matrix represents the impact degree of each attribute on overall customer satisfaction, and the vertical axis represents the pre-purchase user attention, thus forming a dual importance matrix in the market dimension to help merchants comprehensively understand market demands.

[0111] 4. Product-level Dual Importance - Performance Matrix Generation Module

[0112] Based on the market-level dual importance matrix, for the merchant's own products, calculate the actual performance of each attribute, and combine it with the dual importance matrix to generate a performance analysis matrix in the product dimension. The specific steps are as follows:

[0113] Calculation of attribute performance: Based on the review data, calculate the performance value of each attribute, such as the actual effect and quality of the product.

[0114] Calculation of dual importance weighted values: Combine the data in the market-level dual importance matrix to calculate the weighted values for each attribute.

[0115] Drawing of the product-level dual importance - performance matrix: The horizontal axis of this matrix represents the actual performance value of each attribute, and the vertical axis represents the weighted value of dual importance, finally forming a dual importance - performance matrix in the product dimension to help merchants identify the priorities for product optimization.

[0116] 5. Result Presentation and Report Generation Module

[0117] This module presents the analysis results to merchants through visualization means, including the double importance matrix at the market level and the double importance - performance matrix at the product level. Through these matrices, merchants can clearly identify key requirements and provide accurate support for product optimization decisions based on the analysis results.

[0118] Technical implementation example

[0119] 1. Data collection and preprocessing:

[0120] The data collection module obtains review data and data from the "Ask Everyone" module on e - commerce platforms through API or web crawler technology. The number of crawls by the web crawler is determined by the time range or the number of data entries preset by the merchant to ensure that the collected data covers a sufficient sample size. When processing review data, the system first performs data cleaning, removing duplicate reviews, reviews over 100 words (usually template - style reviews), and other irrelevant data (such as garbled characters, repeated outputs, etc.) to ensure the quality of the data and the accuracy of the analysis.

[0121] 2. Information extraction from text data:

[0122] The data processing module uses generative large language models (such as DeepSeek or GPT) to deeply analyze the text data. First, the system pre - defines a list of attributes related to different product categories. For example, for plant milk, it may include attributes such as "nutritional components", "product taste", "shelf life", etc.; for toner, it may include attributes such as "product texture", "functional effects", etc. Then, for each review, the system uses a preset prompt template to guide the large language model to identify the relevant attributes mentioned in the review and their corresponding sentiment polarities (such as positive, neutral, negative). In each prompt template, the system provides few - shot learning examples to enhance the accuracy of attribute recognition. Each review may contain multiple attribute and sentiment information. For example, the review "This product has a long shelf life and a good taste" will be parsed as: {[shelf life, positive], [product taste, positive]}. Similarly, for the "Ask Everyone" data, the system will focus on extracting the involved attribute information and finally generate structured attribute and sentiment analysis results.

[0123] 3. Generation of the double importance matrix:

[0124] This part measures the market situation based on the review data of the merchant's own products and competing products of the same category, and draws the double importance matrix at the market level. It mainly includes the following three steps: calculation of pre - purchase importance, calculation of post - purchase importance, and finally generation of the double importance matrix at the market level.

[0125] Calculation of pre - purchase importance:

[0126] The importance before purchase is calculated based on the data from "Ask Everyone" on the e-commerce platform. This data reflects the product attributes that users actively focus on before purchase. We measure the attention of users to each attribute by analyzing the mention frequency of each attribute in the question data. Specifically, two measurement methods can be adopted:

[0127] Proportion of extensive frequency: Calculate the proportion of the number of mentions of an attribute to the total number of mentions of all attributes.

[0128] Proportion of relative frequency: Measure the overall mention frequency of an attribute in all products, and further examine whether this attribute is mainly concentrated in a few products or evenly distributed among multiple products.

[0129] According to the proportion of mention frequency, the attention before purchase of each attribute is obtained, and thus the importance pre_importance of this attribute before purchase is calculated.

[0130] Calculation of importance after purchase:

[0131] The importance after purchase is measured through user comment data, and conjoint analysis is used to determine the relative contribution of each attribute to customer satisfaction. Through multiple regression analysis for modeling, with the satisfaction and dissatisfaction of attributes in the comments and the star ratings as data inputs, the impact of each attribute on customer satisfaction is calculated. The basic form of the model is as follows:

[0132]

[0133] Among them, y j represents the star rating of comment j (i.e., customer satisfaction), and are the regression coefficients of attribute i, representing the impact of attribute i on customer satisfaction in the case of satisfaction (positive) and dissatisfaction (negative), respectively. and respectively record the satisfaction and dissatisfaction of attribute i in comment j. a is the intercept term of the model, indicating the baseline value of the user rating when the sentiment values of all attributes are 0. Its value is estimated through ordinary least squares regression (OLS) and automatically calculated by the regression model.

[0134] The impact degree of each attribute on satisfaction is measured by analyzing the difference in regression coefficients. If the difference of a certain attribute i is large, it indicates that the performance of this attribute in the comments significantly affects the sentiment of users, and thus it is considered a key attribute. Finally, by calculating the difference between the positive and negative impacts of each attribute, the importance post_importance of each attribute after purchase is obtained.

[0135] Drawing of the dual importance matrix at the market level:

[0136] Based on the previous calculation steps, the pre_importance and post_importance of each attribute at the market level are obtained. The horizontal axis represents the impact of post-purchase attributes on customer satisfaction, and the vertical axis represents the pre-purchase user attention. Through normalization, two values corresponding to each attribute are obtained, and the mean values in the matrix are marked with two dashed lines, representing the average level of the market. According to these two dashed lines, the matrix is divided into four quadrants, and finally, a dual importance matrix at the market level as shown in Figure 4 is formed, providing guidance on demand priorities for merchants.

[0137] Demand analysis of the four quadrants:

[0138] (1) Quadrant Q1: Core features

[0139] The features in this quadrant have important impacts on both consumers' purchase decisions and post-purchase satisfaction. They are the characteristics that consumers pay the most attention to when choosing products and are also the key factors ultimately affecting customer satisfaction. Enterprises should prioritize optimizing these features to meet consumers' core needs.

[0140] For example: In a plant milk product, "nutritional components" may be frequently mentioned before purchase by consumers and also show strong emotional feedback in post-purchase reviews. By improving the performance of this feature, such as adding more nutritional components or highlighting its health benefits, consumers' purchase decisions and satisfaction can be effectively enhanced.

[0141] (2) Quadrant Q2: Attention features

[0142] These features have a relatively high level of attention before purchase but have a relatively small impact on customer satisfaction. Such attributes are usually of interest to consumers before purchase, but their actual effects may not be as influential on overall satisfaction as expected. Merchants can improve the performance of these features while maintaining a reasonable cost investment.

[0143] For example: The "packaging design" of plant milk may attract relatively high attention from consumers before purchase, but if the packaging design has no significant impact on the product's function or taste during actual use, its contribution to customer satisfaction is relatively small.

[0144] (3) Quadrant Q3: Low-value features

[0145] The features in this quadrant have a relatively low level of attention before purchase and also have a relatively small impact on customer satisfaction. Generally, these features can be moderately considered during product design and optimization, but excessive resources should not be invested. For example: In plant milk, the "product appearance" may have a relatively low frequency of mention before purchase and has a relatively small impact on consumers' actual experience, so it can be considered as a secondary optimization feature.

[0146] (4) Quadrant Q4: Potential Optimization Features

[0147] These features receive less attention before purchase, but they have a greater impact on customer satisfaction. Merchants can enhance the overall satisfaction of the product by optimizing these features, especially in areas that users do not anticipate.

[0148] For example, the "shelf life" of plant-based milk may receive less attention before purchase, but if consumers find that the product has a short shelf life during use, this may have a greater negative impact on their satisfaction. Merchants can improve the performance of this feature and enhance the consumer experience by extending the shelf life or improving the storage method.

[0149] Comparative Analysis

[0150] Different from the previous method of identifying important features based only on post-purchase metrics (post_importance), the present invention classifies features more comprehensively from a dual perspective before and after purchase (post_importance and pre_importance). In the traditional method, features are usually classified into core features and low-value features, only considering the needs shown by consumers after purchase. However, by combining the demand perspectives before and after purchase, the present invention can not only identify core features but also further subdivide them into attention features, low-value features, and potential optimization features, providing a more accurate demand analysis. This dual-perspective analysis method enables merchants to understand consumers' needs more comprehensively, optimize product features, and improve the accuracy of decision-making and market adaptability.

[0151] Table 1 Comparison between the Dual Importance Matrix and Single Importance Features

[0152]

[0153] 4. Construction of the Dual Importance - Performance Matrix at the Product Level:

[0154] Based on the dual importance matrix at the market level, the system calculates the actual performance of each attribute based on the review data of the merchant's own products and combines the dual importance indicators to draw the dual importance - performance matrix at the product level. This process focuses on the merchant's product itself, and the specific steps are as follows:

[0155] Calculation of Dual Importance:

[0156] First, define several variables:

[0157] pre_importance i : Importance of attribute (i) before purchase

[0158] post_importancei : Importance (i) after purchase

[0159] pre_importancce_mean: Mean of importance before purchase for all attributes

[0160] post_importance_mean: Mean of importance after purchase for all attributes

[0161] θ: Weight adjusted based on conditional judgment, taking values of 0, 1, or 2. The calculation formula is as follows:

[0162]

[0163] w 1 +w 2 = 1

[0164] The calculation method of θ is as follows:

[0165]

[0166] Explanation:

[0167] Weighted calculation: First, combine the importance before and after purchase through the weighting factors w 1 and w 2 , with both defaulting to one-half and can be specifically selected by the user.

[0168] θ value adjustment: Adjust the weight according to the performance level of the attribute before and after purchase through the conditional judgment θ. Specifically, the value of θ is obtained by comparing the pre-purchase attention of attribute i with the pre-purchase mean of all attributes and the post-purchase importance with the post-purchase mean:

[0169] When the importance of attribute i before and after purchase is higher than the mean, θ is 2, indicating that this attribute has a greater impact on the merchant;

[0170] When one of the dimensions is higher than the mean, θ is 1, indicating that this attribute is relatively important;

[0171] When both dimensions are lower than the mean, θ is 0, indicating that this attribute has a smaller impact on the demand.

[0172] The calculated dual_importance synthesizes the influence of pre- and post-purchase characteristics and weights according to its importance.

[0173] Product performance calculation

[0174] The performance of the product is measured by the proportion of positive sentiment in the review data, and this indicator is called performance. This sentiment proportion represents the proportion of positive feedback for each attribute in the user reviews out of the total feedback, serving as an indicator to measure user satisfaction. A higher proportion of positive sentiment indicates better performance of the attribute, and vice versa.

[0175] Draw a dual importance - performance matrix

[0176] Based on the calculated dual_importance and the performance of each attribute, these metrics are normalized, and a dual importance - performance matrix at the product level is drawn. As Figure 5 shown, the horizontal axis of the matrix represents the actual performance value of each attribute, and the vertical axis represents the weighted dual importance value.

[0177] In Figure 5 the horizontal axis of the dashed line represents the mean of performance, and there are two dashed lines on the vertical axis, one is one - third and the other is two - thirds. These two dashed lines correspond to the calculation criteria in the dual importance calculation formula. These dashed lines divide the vertical axis into three regions, reflecting the importance of the attribute before and after purchase.

[0178] Specifically, the vertical axis measures the dual importance and is divided into three regions:

[0179] The first region: For the attributes in this region, both the importance before purchase and the importance after purchase are less than the mean.

[0180] The second region: For the attributes in this region, at least one of the importance before purchase and the importance after purchase is greater than the mean.

[0181] The third region: For the attributes in this region, both the importance before purchase and the importance after purchase are greater than the mean.

[0182] To ensure that each attribute accurately belongs to its corresponding region and the attributes within the same region can be reasonably compared, the dual importance calculation formula weights the two importances and multiplies by one - third to prevent any attribute from deviating from its category.

[0183] Next, each quadrant in Figure 5 and each quadrant described in Table 2 ( Figure 5 and Table 2 are intermediate processes of the preferred embodiment) are described in detail. Each quadrant is labeled in counter - clockwise order, and the quadrants are divided based on whether the attribute exceeds the mean. Specifically:

[0184] Table 2 Dual importance - performance method matrix division

[0185]

[0186] Figure 5 Shows the primary version of the dual importance - performance matrix at the product level, which can be used as an intermediate process for the preferred embodiment (described in further detail below).

[0187] Q1 quadrant: The attributes in this quadrant perform excellently, and both the importance before and after purchase are high. The attributes are above the two - thirds line, indicating that both the importance before and after purchase of this attribute exceed the average value.

[0188] Q2 quadrant: The attributes in this quadrant perform poorly, but both the importance before and after purchase are high. That is, although consumers pay great attention to this attribute before purchase and its impact on satisfaction is relatively large, this attribute fails to meet the expected performance.

[0189] Q3 quadrant: The attributes in this quadrant perform poorly, and only one of the importance before or after purchase exceeds the average value. This indicates that although this attribute has a relatively large impact in one aspect, the overall performance fails to meet the user's needs.

[0190] Q4 quadrant: The attributes in this quadrant perform poorly, and both the importance before and after purchase are low.

[0191] Q5 quadrant: The attributes in this quadrant perform well, but both the importance before and after purchase are low. Although these attributes perform well in actual performance, their impact on consumers' purchase decisions and satisfaction is small, and they may not be fully explored or promoted.

[0192] Q6 quadrant: The attributes in this quadrant perform well, but only one of the importance before and after purchase is high.

[0193] Under the condition of effective resources, merchants need to clarify the importance of requirements to determine priorities, so as to help evaluate and optimize resource allocation. By calculating the performance of each attribute at the product level and combining the dual importance calculation formula, the final dual importance - performance matrix is obtained. On this basis, the preferred embodiment combines the original Q1 and Q6 into one module, corresponding to the "continue to maintain" strategy, in order to better optimize resource allocation. Therefore, the preferred embodiment finally divides the features into five modules. Figure 6 Shows the final version of the dual importance - performance matrix at the product level of this preferred embodiment of the present invention.

[0194] Characteristic analysis of the five modules:

[0195] Q1 area: Continue to maintain (excellent performance and high dual importance)

[0196] The characteristics of this area have an important impact on consumers' purchase decisions and satisfaction, both before and after purchase. These characteristics are either highly concerned before purchase, or have a significant impact on satisfaction after purchase, or both are high. Regardless of which stage, the degree of attention paid to them indicates their importance to the overall experience. Even if the performance in a certain stage fails to meet expectations, businesses should still regard these characteristics as a priority for "continuing to maintain" because they affect consumers' core needs and long-term loyalty. Businesses should ensure that the advantages of these characteristics are not lost, otherwise it may seriously affect consumers' satisfaction and even lead to customer churn.

[0197] For example: The "nutritional components" in plant milk are core characteristics that consumers highly concern about and have also received significant positive evaluations after purchase. Businesses should ensure to continue maintaining the advantages of this characteristic, avoid its quality decline, and maintain consumers' loyalty.

[0198] Area Q2: Priority for improvement (poor performance and high dual importance)

[0199] The attributes in this area are crucial for consumers' purchase decisions and satisfaction, indicating that their importance is relatively high both before and after purchase. However, the performance of these characteristics fails to meet consumers' expectations. Businesses should take these attributes as the priority for improvement, enhance their performance as soon as possible to meet consumers' needs, so as not to affect the overall satisfaction.

[0200] For example: The "taste" of plant milk may be a characteristic that consumers highly concern about before purchase, but if the taste fails to meet expectations, businesses should prioritize improvement to enhance consumers' overall experience.

[0201] Area Q3: Low priority (poor performance and relatively high dual importance)

[0202] The attributes in this area, although having a certain impact on consumers, only one of their importance before or after purchase is high, and the performance is poor. The importance of such attributes is relatively high, but their performance fails to meet expectations. Therefore, compared with Q2, businesses can optimize them as a secondary priority. In the case of limited resources, the optimization of these attributes can be slightly postponed, and the key issues in Area Q2 should be resolved first.

[0203] For example: The "packaging design" of plant milk may attract relatively high attention before purchase, but its actual effect fails to meet consumers' expectations. Businesses can make improvements when resources permit.

[0204] Area Q4: Irrelevant (poor performance and low dual importance)

[0205] The characteristics of this area have little impact on consumers' purchase decisions and satisfaction, and the performance is also poor. Merchants should consider these characteristics as low-priority attributes and can ignore them when resources are limited. Since these characteristics are neither highly concerned by consumers nor bring obvious benefits to consumers, merchants do not need to invest too many resources.

[0206] For example, the "product appearance" of plant milk may not receive extensive attention from consumers and has no significant impact on the product usage experience. Therefore, merchants can choose not to optimize it temporarily.

[0207] Area Q5: Appropriately reduce (good performance but low dual importance)

[0208] The attributes in this area perform well and have received positive feedback from consumers in actual use. However, consumers pay less attention to these characteristics, indicating that they have little impact on purchase decisions. Merchants can consider reducing further investment in these attributes, especially when resources are limited, and concentrating more resources on key characteristics with greater influence.

[0209] For example, the "shelf life" of plant milk may perform excellently after purchase, but consumers pay less attention to this attribute before purchase. Merchants can appropriately reduce the investment in this characteristic and instead optimize those characteristics that have a greater impact on purchase decisions.

[0210] Comparative analysis

[0211] Different from the traditional importance-performance (IPA) analysis method, the present invention proposes a dual importance-performance analysis by introducing a comprehensive perspective before and after purchase. Compared with existing research, the method of the present invention is more comprehensive and accurate, and can simultaneously consider consumers' concerns before purchase and satisfaction feedback after purchase, so as to more accurately identify key needs and optimization directions. Through the dual importance matrix, we can not only evaluate the influence of each product characteristic on consumers, but also cross the limitations of a single dimension and comprehensively understand the performance of the product at different stages.

[0212] Furthermore, the preferred embodiment of the present invention divides the requirements at the product level into five areas, helping merchants more scientifically identify which characteristics should be optimized first, which should maintain the existing state, and which can reduce resource investment. This detailed division method provides clear guidance for merchants' decision-making, especially when resources are limited, and can help merchants make more efficient optimizations and resource allocations.

[0213] The following table compares the differences between the method of the present invention and existing methods. The existing IPA method mainly relies on post - purchase metrics (post_importance&performance) to measure the priority of requirements, while the present invention starts from the dual dimensions before and after purchase and the specific performance of the product (pre_importance&post_importance&performance), providing a more detailed and comprehensive requirements analysis. In addition, the preferred embodiment of the present invention innovatively proposes Figure 6 the visualization scheme in, which displays the three dimensions in a two - dimensional graph, enabling more intuitive display of more information. Compared with a three - dimensional scatter plot, this two - dimensional graph is not only convenient for observation but also helps merchants quickly understand and make decisions.

[0214] Table 3 Differences between the method of the present invention and existing methods

[0215]

[0216] 5. Result presentation and report generation:

[0217] The system visually presents the analysis results to merchants through visualization tools, including the dual - importance matrix at the market level and the dual - importance - performance matrix at the product level. These matrices not only help merchants accurately identify the key requirements that consumers are concerned about but also provide clear guidance for product optimization and decision - making, assisting merchants in making more informed strategic choices in the fierce market competition.

[0218] The technical solution of the present invention structurally processes the user - generated content (UGC) on the e - commerce platform through a large - language model, combines the dual - importance matrix and the dual - importance - performance matrix at the market level and the product level, and comprehensively reveals the behavioral differences of consumers in the purchase decision - making and usage feedback. This innovative requirements analysis method can provide more accurate market insights and product optimization suggestions for merchants, having significant theoretical value and practical significance.

[0219] The important innovative work and technical key points of the present invention include:

[0220] 1. Requirements analysis method based on a large - language model

[0221] The present invention proposes an innovative e - commerce requirements analysis method based on a large - language model. Through sentiment analysis and attribute extraction of the user - generated content (UGC) on the e - commerce platform, a comprehensive requirements analysis is carried out from two dimensions: the pre - purchase user attention and the post - purchase customer satisfaction, comprehensively capturing the changes in consumer requirements.

[0222] 2. Generation of the dual - importance matrix

[0223] The present invention proposes a new method to generate a dual importance matrix. By calculating the pre - purchase user attention and post - purchase customer satisfaction, this method helps merchants identify the relative importance of product attributes at the market level, providing data - based demand priority ranking for merchants.

[0224] 3. Construction of the Dual Importance - Performance Matrix

[0225] The present invention further combines the actual performance of products to propose a dual importance - performance matrix. This matrix helps merchants analyze the relationship between the actual performance of each product attribute and its importance in the market, thereby providing scientific decision - making support for product optimization for merchants.

[0226] 4. Demand Analysis Framework from a Brand - New Perspective

[0227] The present invention conducts demand analysis from the dual perspectives before and after purchase, solving the problem that traditional methods only focus on post - purchase satisfaction. This framework not only pays attention to consumers' concerns before purchase decisions but also analyzes the impact of the product's performance after actual use on consumer satisfaction, thus providing a complete demand analysis process to help merchants achieve more accurate product optimization.

[0228] Compared with existing e - commerce demand analysis technologies based on traditional data analysis methods, the present invention has significant advantages. Traditional technologies mostly rely on simple keyword extraction or sentiment analysis, usually only focusing on users' feedback after purchase and ignoring consumers' concerns and decision - making processes before purchase. The present invention, through the application of large - language models, combines the demand perspectives of the two stages before and after purchase, comprehensively analyzes users' attention and satisfaction, and can more accurately capture the real needs and emotional changes of consumers in the purchase decision.

[0229] In addition, the present invention innovatively proposes methods for constructing the dual importance matrix and the dual importance - performance matrix, which combine users' behavioral characteristics before and after purchase. It not only helps merchants identify the most critical demand characteristics in the market but also analyzes the relationship between the actual performance of product attributes and their importance. This method provides a more detailed demand priority ranking and can provide more accurate decision - making support for merchants in aspects such as product optimization, resource allocation, and market promotion. Therefore, the present invention has higher accuracy and comprehensiveness than traditional methods and can bring greater market competitiveness to merchants.

[0230] The embodiment of the present invention also provides a storage medium for storing a computer program, which when executed, at least executes the method described above.

[0231] An embodiment of the present invention also provides a control device, including a processor and a storage medium for storing a computer program; wherein, the processor is configured to execute at least the method described above when executing the computer program.

[0232] An embodiment of the present invention also provides a processor, which executes a computer program and at least executes the method described above.

[0233] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM, Read Only Memory), a programmable read-only memory (PROM, Programmable Read-Only Memory), an erasable programmable read-only memory (EPROM, Erasable Programmable Read-Only Memory), an electrically erasable programmable read-only memory (EEPROM, Electrically Erasable Programmable Read-Only Memory), a ferromagnetic random access memory (FRAM, Ferromagnetic Random Access Memory), a flash memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM, Compact Disc Read-Only Memory); the magnetic surface memory can be a disk memory or a tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memories.

[0234] In several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed with each other can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical, or other forms.

[0235] The units described as separate components above may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units; some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0236] In addition, in each embodiment of the present invention, each functional unit can be entirely integrated into one processing unit, or each unit can be separately regarded as one unit, or two or more units can be integrated into one unit; the above-mentioned integrated unit can be implemented in the form of hardware, or in the form of a combination of hardware and software functional units.

[0237] Those of ordinary skill in the art can understand that all or part of the steps for implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including the above method embodiments; and the foregoing storage medium includes: removable storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.

[0238] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present invention, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. And the foregoing storage medium includes: removable storage devices, ROM, RAM, magnetic disks, or optical discs and other various media that can store program codes.

[0239] The methods disclosed in several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0240] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0241] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0242] The above content is a further detailed description of the present invention in combination with specific preferred implementation manners, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those skilled in the technical field to which the present invention pertains, without departing from the concept of the present invention, several equivalent substitutions or obvious variations can be made, and as long as the performance or use is the same, they should all be regarded as falling within the protection scope of the present invention.

Claims

1. An e-commerce demand analysis method based on a large language model, characterized in that: The following steps are involved: S1. Data collection: Collect user-generated content (UGC) of merchants’ own products and target competitors’ products from e-commerce platforms, including product review data and user question and answer data; S2. Data processing and analysis: Use the large language model to process and analyze the collected user-generated content, extract product attributes and their sentiment information, and convert the extracted information into structured data; S3. Generation of dual importance matrix at market level: Based on the extracted structured data, the pre-purchase attention and post-purchase importance of each product attribute are calculated respectively, and the dual importance matrix at market level is generated by combining the attention and importance, which is used to evaluate the relative importance of different product attributes in consumer purchase decisions and satisfaction; S4. Product-level dual importance-performance matrix generation: Combined with the market-level dual importance matrix, for the merchant's own products, calculate the actual performance value of each product attribute, and generate a product-level dual importance-performance matrix to identify product features that need to be optimized; S5. Result presentation and report generation: The generated dual importance matrix and dual importance-performance matrix are presented in a visual way, and an analysis report is generated to provide merchants with market insights and product optimization decision support.

2. The e-commerce demand analysis method based on a large language model according to claim 1 is characterized in that: In step S2, when the large language model is used to process and analyze the user-generated content, the following steps are specifically included: List of attributes for predefined product categories; For each comment, a preset prompt template is used to guide the large language model to identify the attributes involved in the comment and their corresponding sentiment polarity; Convert the identified attributes and their sentiment information into structured data.

3. The e-commerce demand analysis method based on a large language model according to claim 1 is characterized in that: In step S3, when calculating the pre-purchase attention of each product attribute, the following steps are specifically included: Analyze the user question data in the user question and answer module and count the mention frequency of each attribute; Calculate the pre-purchase attention of each attribute based on the frequency of mentions.

4. The e-commerce demand analysis method based on a large language model according to claim 1 is characterized in that: In step S3, when calculating the post-purchase importance of each product attribute, the following steps are specifically included: Based on the review data, the joint analysis method is used to calculate the impact of each attribute on customer satisfaction through multiple regression analysis modeling; Based on the difference in regression coefficients, the post-purchase importance of each attribute is determined.

5. The e-commerce demand analysis method based on a large language model according to claim 1 is characterized in that: In step S3, when generating the dual importance matrix at the market level, the following steps are specifically included: Normalize pre-purchase attention and post-purchase importance; Draw a dual importance matrix with post-purchase importance as the horizontal axis and pre-purchase attention as the vertical axis; According to the mean value in the matrix, four quadrants are divided, corresponding to core features, focus features, low-value features and potential optimization features.

6. The e-commerce demand analysis method based on a large language model according to claim 1 is characterized in that: In step S4, when calculating the actual performance value of each product attribute, the following steps are specifically included: Based on the review data, the positive sentiment ratio of each attribute is calculated as the actual performance value of the attribute.

7. The e-commerce demand analysis method based on a large language model according to claim 1 is characterized in that: In step S4, generating the dual importance-performance matrix at the product level specifically includes the following steps: Determine the mean of the pre-purchase importance and the post-purchase importance of each attribute as a benchmark for evaluating the relative importance of the attribute; determine the weight adjustment factor of each attribute using a conditional judgment mechanism, and dynamically adjust the weight adjustment factor based on the comparison result of the pre-purchase importance and the post-purchase importance of the attribute with their respective means; and obtain the dual importance of each attribute based on the weighted calculation of the weight adjustment factor, the pre-purchase importance, and the post-purchase importance of each attribute; With the dual importance of the attribute as the vertical axis and the actual performance value as the horizontal axis, a dual importance-performance matrix at the product level is drawn; the dual importance-performance matrix is ​​divided into five modules, and the characteristics of each module are as follows: The first module contains attributes with high dual importance and high actual performance values, indicating that these attributes have a great impact on consumer decisions and perform well, and should continue to be maintained; The second module contains attributes with high dual importance but low actual performance values, indicating that these attributes have a great impact on consumer decisions but perform poorly and need to be improved as a priority; The third module contains attributes with high dual importance but low actual performance values, indicating that these attributes have some impact on consumers but the performance is not as expected, and can be kept at low priority for optimization; The fourth module contains attributes with low dual importance and low actual performance values, indicating that these attributes have little impact on consumer decisions and average performance and can remain insignificant; The fifth module contains attributes with low dual importance but high actual performance values, indicating that although these attributes have little impact on consumer decisions, they perform well and can be appropriately reduced in investment.

8. The e-commerce demand analysis method based on a large language model according to claim 7 is characterized in that: In step S4, the dual importance of each attribute is a dual importance obtained by weighted calculation based on the pre-purchase attention, the post-purchase importance and the weight adjustment factor adjusted based on conditional judgment; the conditional judgment mechanism is specifically as follows: when the pre-purchase importance and the post-purchase importance of the attribute are both higher than their mean, a higher weight adjustment factor is assigned; when any importance is higher than its mean, a medium weight adjustment factor is assigned; when both are lower than their mean, a lower weight adjustment factor is assigned.

9. The e-commerce demand analysis method based on a large language model according to claim 1, characterized in that: In step S5, when presenting the results and generating the report, the following steps are specifically included: Visualization tools are used to display the dual importance matrix at the market level and the dual importance-performance matrix at the product level; Generate analytical reports to provide market insights and product optimization decision support.

10. An e-commerce demand analysis system based on a large language model, characterized in that: The following steps are involved: Data collection module: collects user-generated content (UGC) of merchants’ own products and target competitors’ products from e-commerce platforms, including product review data and user question and answer data; Data processing and analysis module: Use the large language model to process and analyze the collected user-generated content, extract product attributes and their sentiment information, and convert the extracted information into structured data; Market-level dual importance matrix generation module: Based on the extracted structured data, the pre-purchase attention and post-purchase importance of each product attribute are calculated respectively, and the market-level dual importance matrix is ​​generated by combining the attention and importance, which is used to evaluate the relative importance of different product attributes in consumer purchase decisions and satisfaction; Product-level dual importance-performance matrix generation module: Combined with the market-level dual importance matrix, for the merchant's own products, the actual performance value of each product attribute is calculated, and the product-level dual importance-performance matrix is ​​generated to identify product features that need to be optimized; Result presentation and report generation module: The generated dual importance matrix and dual importance-performance matrix are presented in a visual way, and an analysis report is generated to provide merchants with market insights and product optimization decision support.

Images