A user demand data analysis method and system for product improvement strategy formulation

Abstract

The present invention discloses a user demand data analysis method and system for product improvement strategy formulation, including obtaining comment data of target product categories, constructing a product element dictionary, and obtaining the weight of elements based on satisfaction through joint analysis; obtaining user purchase records and product process parameters, using a two-stage utility model to obtain the influence coefficient, and using a product element dictionary to obtain the weight of elements based on purchase tendency; based on a dual-weight model, using two types of weight results to classify product elements to obtain targeted product improvement strategies. The present invention provides product improvement guidance for product manufacturers through user demand analysis based on comprehensive online comments and purchase records of users, and effectively saves labor costs and improves the efficiency and comprehensiveness of user demand analysis through comprehensive data use and systematic data analysis.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular, to a user demand analysis method based on user data, and more particularly, to a user demand data analysis method and system for formulating product improvement strategies. Background Art

[0002] With the development of e-commerce, the acceleration of product iteration and the intensification of competition have made it more immediate for enterprises to understand customer needs. Analysis methods based on user-generated big data have gradually replaced traditional survey methods. Using traditional data analysis models to analyze user-generated content has gradually become a necessary step before product improvement. However, the existing data analysis models are single-dimensional when processing customer needs, ignoring the differences in user preferences at different stages of purchase, resulting in incomplete data analysis results and insufficient support for the formulation of enterprise product improvement strategies. Summary of the invention

[0003] In order to solve the above defects in the prior art, the purpose of the present invention is to provide a data analysis framework for product improvement strategy to improve the comprehensiveness and effectiveness of user demand analysis.

[0004] The objectives of the present invention are achieved through the following technical solutions.

[0005] On the one hand, the present invention provides a user demand data analysis method for product improvement strategy, which integrates two kinds of data: user online comments and purchase records, and uses methods such as conjoint analysis and Markov Chain Monte Carlo technology to construct a dual-weight model to realize the function of systematically analyzing user needs to support the formulation of product improvement strategies.

[0006] The user demand data analysis method for formulating a product improvement strategy provided by the present invention includes:

[0007] Obtain review data for target product categories, construct a product factor dictionary, and obtain satisfaction-based weights of product factors through joint analysis;

[0008] Obtain user purchase records and product process parameters, use the two-stage utility model to obtain the influence coefficient, and use the constructed product element dictionary to obtain the weight of product elements based on user purchase tendency;

[0009] Based on the dual-weight model, the product elements are classified using the weights of product elements based on satisfaction and the weights of product elements based on purchase intention, and targeted product improvement strategies are obtained.

[0010] Preferably, the calculation of the weights of product factors based on satisfaction includes:

[0011] P11. Structuring the text of the target product category reviews;

[0012] P12. Score the text-structured reviews through a joint analysis model and calculate the weights of product elements based on satisfaction.

[0013] Preferably, in step P11, the text in the target product category review is structured by the following method:

[0014] P111. Construct a product factor dictionary through a loop method, group and segment all comments, extract product function related words from the first group, add high-frequency words to the word library, delete the redundant words that have been added to the word library from the remaining groups, and loop until all groups are processed to obtain a word library containing product function related words;

[0015] Based on word2vec, the similarity between every two words in the vocabulary is calculated, and clustering is performed based on the similarity to obtain a product element dictionary. The structure of the obtained dictionary is: product element - words obtained in the review;

[0016] P112, perform a reverse search on all reviews based on the product factor dictionary, and generate a review-factor relationship matrix based on whether the review contains the corresponding words in the factor dictionary;

[0017] P113. For the comment-element relationship matrix, a sentiment dictionary is constructed using part-of-speech tagging and implicit Dirichlet allocation, and the sentiment of the elements contained in the comment is converted into a 0-1 matrix indicating whether the comment contains positive or negative features, thereby obtaining a structured comment text.

[0018] Preferably, for the user purchase records and product process parameters of the target product, the weight based on the user purchase tendency is calculated in the following manner:

[0019] P21. Use the two-stage utility model to obtain the influence coefficient of product process parameters on user purchase tendency;

[0020] P22. Use the correlation matrix to convert the influence coefficient of process parameters on user purchasing tendency into the influence weight of product factors on user purchasing tendency.

[0021] Preferably, the step P22 comprises the following steps:

[0022] P221, conduct fuzzy evaluation of the correlation between factors and process parameters, the evaluation includes five-level scale variables: very weak, weak, medium, strong and very strong;

[0023] P222. Calculate the weight of the influence of product elements on user purchasing tendency.

[0024] Preferably, the product elements are classified using the obtained weights of product elements based on satisfaction and the weights of product elements based on purchasing tendency, and a graphical classification method is adopted: a dual-weight model is used to classify product elements. The horizontal and vertical axes of the model correspond to the influence weights based on user satisfaction and purchasing tendency, respectively. The model is divided into four categories according to the similarity and level of user satisfaction and purchasing tendency: Category I, Category II, Category III and Category IV.

[0025] Product elements are categorized into:

[0026] Category I: This type of factor has a great impact on user purchase intention and satisfaction. It is necessary to improve product functions and carefully consider innovation to improve user purchase utility and satisfaction after use.

[0027] Category II: This type of factor has a greater impact on user purchase intention than satisfaction, and unsuccessful product function innovation will not reduce user satisfaction;

[0028] Category III: This type of factor is not important for purchase intention and satisfaction. It is necessary to reduce investment in functions and seek cost-optimized alternatives.

[0029] Category IV: This type of factor has a greater impact on satisfaction than on purchase intention. The overall quality of product functions should be improved to maintain and improve user satisfaction.

[0030] Through the above analysis framework, the user's comment data is structured through text mining and sentiment analysis, and the impact weight of product factors based on user satisfaction is calculated by the joint analysis method; the user's purchase record and product process parameters are modeled by a two-stage utility model, and the Markov Monte Carlo method is used to calculate the impact weight of product factors based on user purchase tendency; the weights of product factors for satisfaction and purchase tendency are integrated through the double-weight model, and the product factors are divided into four categories, and corresponding product improvement strategies are formulated according to the corresponding category characteristics.

[0031] In another aspect, the present invention provides a user demand data analysis system for formulating product improvement strategies, comprising:

[0032] Online data collection device, used to collect product online review data and product process parameters;

[0033] A review text mining device is used to perform text analysis on the collected review data to obtain a product element database and structured reviews;

[0034] An element satisfaction analysis device is used to perform joint analysis on structured comments to obtain an element satisfaction measurement database;

[0035] User purchase behavior collection device, used to collect user purchase records;

[0036] The purchase behavior analysis device combines the product factor database with the two-stage utility model to obtain the factor-based purchase tendency measurement database;

[0037] The dual-factor analysis device integrates the factor satisfaction measurement database and the factor-based purchase tendency measurement database, classifies and analyzes the factors based on the dual-weight model, and obtains the product improvement strategy database, providing product manufacturers with targeted product improvement strategies.

[0038] To achieve the purpose of the invention, the present invention uses purchase records to measure the impact of product elements on purchasing tendencies. The present invention analyzes the impact of products on overall satisfaction through online comments after purchase. The present invention provides a dual-weight model, which integrates the two dimensions of user comments and purchases and divides product features into four categories. According to different categories, corresponding product design suggestions are provided to product designers. Therefore, we provide a comprehensive analysis framework to study the needs from multiple dimensions of user behavior and user speech, before and after use, purchase motivation and satisfaction factors. This project combines user comments and user purchase records for the first time, introduces them into the user demand analysis framework, and obtains more comprehensive analysis results. The method has good robustness and can be implemented in many product areas.

[0039] Other features and advantages of the present invention will be described in detail in the following detailed description.

[0040] The present invention adopts the above technical solution, which has the following beneficial effects:

[0041] 1) Comprehensiveness of user needs: This method expands the dimension of user needs analysis and takes into account the differences in user needs at different stages.

[0042] 2) Reduce the innovation risk in product improvement for product manufacturers. This method uses a dual-weight model to classify the needs of users at different stages into four categories, where the second category of elements is characterized by a high weight in purchase intention but a low weight in overall satisfaction. Therefore, implementing changes and innovations using these features is considered to be low risk.

[0043] 3) It has strong category expansibility. The data used in this method is highly available, and the analysis method already includes the processing of unstructured data. This method can be applied to enterprises and individuals with product improvement needs to conduct demand analysis on various product categories. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute an improper limitation of the present invention. In the drawings:

[0045] Figure 1 is a schematic diagram of a system structure for user needs for product improvement strategy formulation according to an embodiment of the present invention;

[0046] Figure 2 is a user demand analysis flow chart for formulating a product improvement strategy according to an embodiment of the present invention;

[0047] Figure 3 is an example diagram of a two-stage utility model according to an embodiment of the present invention;

[0048] Figure 4 This is an example diagram of dual-weight model classification according to an embodiment of the present invention;

[0049] Figure 5 It is the classification result of the product elements corresponding to the embodiment of the present invention based on the double weight model. DETAILED DESCRIPTION

[0050] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The exemplary embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0051] Figure 1 Schematic diagram of a system structure for user needs in formulating product improvement strategies according to an embodiment of the present invention. Figure 1 The user demand data analysis system for product improvement strategy formulation of the present invention shown in the figure includes: an online data collection device 10, which is used to collect product online review data and product process parameters; a review text mining device 11, which is used to perform text analysis on the collected review data to obtain a product factor database 12 and structured reviews; a factor satisfaction analysis device 13, which is used to perform joint analysis on the structured reviews to obtain a factor satisfaction measurement database 14; a user purchase behavior collection device 20, which is used to collect user purchase records; a purchase behavior analysis device 21, which combines the product factor database 12 with the two-stage utility model to perform analysis to obtain a factor-based purchase tendency measurement database 22; a dual-factor analysis device 30, which integrates the factor satisfaction measurement database 14 and the factor-based purchase tendency measurement database 22, and classifies and analyzes the factors based on a dual weight model to obtain a product improvement strategy database 31, and provides product manufacturers with targeted product improvement strategies through output 32.

[0052] like Figure 2As shown, according to a user demand analysis flow chart for product improvement strategy formulation according to an embodiment of the present invention, the present invention is a comprehensive analysis framework, which refers to mature data acquisition technology and calculation methods, considers the differences in user needs at different stages, and provides a two-dimensional user demand analysis method. The analysis framework of the method includes three parts: the first part is demand mining based on user satisfaction, which is used to analyze the weight of product elements for user satisfaction, including: obtaining comment data of the target product category, structured comments, and establishing a joint analysis model to calculate the influence weight of product elements on overall satisfaction; the second part is demand mining based on user purchase tendency, which is used to analyze the weight of product elements for user purchase tendency, including: obtaining user purchase records and product process parameters, using a two-stage utility model to obtain the influence coefficient, and using a product element dictionary to obtain the weight of the element based on the purchase tendency; the third part is product improvement strategy analysis, which is used to summarize the two types of weights of the elements into a dual-weight model, classify the elements based on the level and similarity of the two types of weights, and formulate corresponding product improvement strategies.

[0053] The user demand data analysis method for formulating a product improvement strategy of the present invention comprises the following steps:

[0054] P1. Obtain review data for the target product category, build a product factor dictionary, and obtain the weights of product factors based on satisfaction through joint analysis.

[0055] The factor weight analysis based on satisfaction mainly analyzes the review data of the target product category, which mainly includes user identification (such as user ID), review star rating and review text content. The main processing process includes obtaining the review data of the target product category on the public e-commerce platform, including user ID, review star rating and review text content, and calculating the weight of the factor based on satisfaction.

[0056] The weights of factors based on satisfaction are calculated, including:

[0057] P11. Structure the text obtained from the reviews of the target product category.

[0058] The structuring of the target product category review text is completed through the following methods:

[0059] P111. Construct a product element dictionary through a loop method. Specifically, first group and segment all comments, extract words related to product functions from the first group, add high-frequency words to the word library, and delete the words that have been added to the word library from all subsequent groups, loop until all groups are processed, and obtain a word library containing words related to product functions; calculate the similarity between every two words in the word library based on word2vec, cluster based on similarity, and adjust according to product-related knowledge to obtain a product element dictionary. The structure of the obtained dictionary is: product element-words obtained in the comments; further, traverse all comments, judge the elements mentioned in each comment based on the product element dictionary, and obtain the relationship matrix X between comments and elements. ij , an element value of 1 indicates that it is mentioned.

[0060] P112. Perform a reverse search on all reviews based on the above product factor dictionary, and generate a review-factor relationship matrix based on whether the reviews contain corresponding words in the factor dictionary;

[0061] P113. For the comment-element relationship matrix, a sentiment dictionary is constructed using part-of-speech tagging and latent Dirichlet allocation (LDA), sentiment analysis is performed based on the elements, and the sentiment of the elements contained in the comment is converted into a 0-1 matrix indicating whether the comment contains positive or negative features, thereby obtaining a structured comment text.

[0062] For the review-factor relationship matrix X ij Perform sentiment analysis. The goal is to transform the comment-element relationship matrix X ij Each column of is split into two sub-columns, and the sentiment of the corresponding comment when the element is mentioned is judged to be positive Still negative Specifically, 10% of the comments were extracted for part-of-speech tagging, stop words were removed and part-of-speech screening was performed, a dictionary containing positive and negative emotions was obtained using the Latent Dirichlet Allocation (LDA) method, and the comments were traversed for reverse search. Based on the dependency relationship between attribute words and sentiment words, the structured comments were obtained.

[0063] P12. Apply the conjoint analysis model to the structured reviews and the corresponding review scores, and calculate the weights of product factors based on satisfaction through the following steps:

[0064] P121. The joint analysis model is:

[0065] Among them, y i is the star rating given to review i, representing the overall satisfaction of review i; and are indicator variables for whether comment i evaluates factor j positively (pos) or negatively (neg), respectively; and They represent the impact of the two evaluations of factor j on the overall satisfaction; α is the intercept; n is the number of product factors in the constructed product factor dictionary.

[0066] P122. Substitute the coefficients calculated by the above joint analysis model into the formula:

[0067] Calculated That is, the weight of the impact of product elements on overall satisfaction.

[0068] P2. Obtain user purchase records and product process parameters, use the two-stage utility model to obtain the influence coefficient, and use the constructed product element dictionary to obtain the weight of product elements based on purchase tendency. The user's purchase record should include the user's basic personal information (age, gender, etc.) and the purchased products.

[0069] The factor weight analysis based on user purchase tendency mainly analyzes data including user purchase records and product process parameters. The main processing steps include obtaining user purchase records and product process parameters of target products, and calculating the weight of factors based on purchase tendency.

[0070] For the user purchase records and product process parameters of the target product, the weights of the factors in the second step based on the user purchase tendency are calculated in the following way.

[0071] P21. Use the two-stage utility model to obtain the influence coefficient of product process parameters on user purchasing tendency, and model the process of users deciding to replace products and purchase a new product.

[0072] In the first stage of the model, the user's potential utility for the old product is modeled by the following formula:

[0073]

[0074] in, represents the utility brought by the old product to user i; x ik is the kth process parameter of the old product of user i; b k is the influence coefficient of the kth process parameter on user utility; y iis the number of days that user i uses the old product until it is replaced; c is the coefficient of the number of days of use on user utility; α0 is the intercept term; ε i is an independent and identically distributed residual term, subject to N(0,σ ε 2 ).

[0075] When the above utility is less than zero, the user makes a decision to change the product, and the probability mass function expressed in vector form is:

[0076] Among them, Replace i =1 represents that user i makes the decision to replace the old product; B is the influence coefficient of different process parameters on user utility b k The vector form of Y is the time from the user using the old product to the replacement. i The vector form of; α0 is the intercept term of the first-stage utility model formula; is the variance of the normal distribution followed by the residual term of the first-stage utility model; X i and Y i They are respectively the first-stage utility model of the user x ik and i is the vector form of ; φ is the probability density function of the univariate normal distribution.

[0077] A multinomial probit model is used to model the user's second-stage choice. Users choose products based on the principle of utility maximization. Specifically, user i will choose products that make their utility u ij 2 The largest product J.

[0078] In the second stage of the model, the user's second-stage utility is defined and a utility benchmark ui randomly selected from the selection set is defined. J The difference is:

[0079]

[0080] In order to calculate u ij 2 , define the new The vector form is expressed as:

[0081]

[0082] Among them, ε i is the residual term vector, subject to N J-1 (0,Σ).

[0083] The user's choice is represented as:

[0084]

[0085] Among them, Choice i represents the purchase choice made by user i in the second stage; j represents the jth product; u iJ is the utility of the benchmark product J to user i in the second stage; is the vector form of the utility set of all products of user i in the second stage.

[0086] The corresponding probability mass function is:

[0087] in, is the utility difference vector of the product based on the benchmark product; X i is the attribute difference vector of the product based on the benchmark product; P i is the price difference vector of the product based on the benchmark product; B and Γ are the influence coefficient vectors of the attribute and price difference respectively; Σ is the covariance matrix of the J-1-element normal distribution obeyed by the intercept term; φ J-1 is the probability density function of the J-1-element normal distribution;

[0088]

[0089] Using the two-stage utility equation to jointly estimate Using MCMC method to analyze the posterior Calculate and get the target parameter B. B is the target parameter of this stage, which represents the influence of different process parameters of the product on the user's purchasing tendency.

[0090] P22. Use a correlation matrix to convert the influence coefficient of process parameters on user purchasing tendency into the influence weight of product factors on user purchasing tendency.

[0091] The specific steps involved are:

[0092] P221. Perform fuzzy evaluation of the correlation between factors and process parameters. The evaluation consists of five-level scale variables (very weak, weak, medium, strong, very strong). The five-level scale variables are defined by the following membership function:

[0093]

[0094] Among them, (l, m, u) correspond to (0, 0, 0.3), (0, 0.25, 0.5), (0.3, 0.5, 0.7), (0.5, 0.75, 1) and (0.7, 1, 1) respectively.

[0095] P222. Use the following formula to calculate the weight of product factors affecting user purchasing tendency:

[0096]

[0097] Among them, Weight j p represents the weight of product factor j on user purchase tendency; δ k is the mean value of the influence coefficient of the process parameters after the aging period in the MCMC process (i.e. after the Markov chain tends to a stable distribution); R jk Represents the correlation coefficient between factor j and process parameter k.

[0098] For R jk The deblurring is performed by calculating the average evaluation result through the following formula:

[0099]

[0100] Among them, R jk represents the correlation coefficient between factor j and process parameter k; μ R is the membership function; r jk Represents the fuzzy evaluation result of the correlation between factor j and process parameter k.

[0101] P3. Based on the dual-weight model, the product elements are classified using the weights of product elements based on satisfaction and the weights of product elements based on purchase intention, and targeted product improvement strategies are obtained.

[0102] The product improvement strategy analysis based on the dual-weight model specifically means aggregating the two types of weight results obtained in the first and second steps into the dual-weight model, classifying the product elements in two dimensions, and obtaining targeted product improvement strategies.

[0103] A graphical classification method is used: a dual-weight model is used to classify product factors. The horizontal and vertical axes of the model correspond to the influence weights of factors based on user satisfaction and purchase intention, respectively. They are divided into four categories according to the similarity and level of the two. The specific classification benchmark calculation formula is:

[0104]

[0105] Among them, X is the horizontal axis coordinate, corresponding to the influence weight of the factor based on user satisfaction; Y is the vertical axis coordinate, corresponding to the influence weight of the factor based on user purchase intention; α is the similarity level, which is an adjustable parameter. Product manufacturers can adjust this parameter according to the actual product improvement needs.

[0106] The specific classification characteristics and corresponding product improvement strategies are:

[0107] Category I: These elements are very important for both user purchase intention and satisfaction, and are the core functions of the product. Their performance affects the reputation of the product, and the degree of differentiation determines the market competitiveness of the product. By improving product functions, companies can increase user purchase utility and satisfaction after use. However, if the functions are not improved appropriately, it will significantly reduce user purchase intention and satisfaction. Therefore, there is a high risk in changing these functions, and improving the performance of these functions is safer than pursuing innovation.

[0108] Category II: This category of factors has a greater impact on purchase intention than satisfaction. Users want to buy products that are different from competing products in these factors. In the overall evaluation after using the product, these features are ignored to a certain extent. On the contrary, unsuccessful product feature innovations will not reduce user satisfaction. We believe that innovations based on these features are low-risk, which is very different from Category I. Therefore, companies can innovate on these factors to widen the gap with competing products and attract users to purchase. Innovation on these factors has a favorable risk-return ratio.

[0109] Category III: This category is completely opposite to Category I. The elements of Category III are not important for purchase intention and satisfaction. In order to better control overall costs and obtain better benefits, companies can reduce investment in these features or even seek cost-optimized alternatives.

[0110] Category IV: This category of factors has a greater impact on satisfaction than on purchase intention. Users cannot perceive or do not care about the differences in these factors when purchasing, but only care about the actual performance when using the product. If the performance does not meet their expectations, it will affect the reputation of the product in the long run and thus affect the future development of the brand. Companies should invest in improving the overall quality of these features to maintain and improve user satisfaction, rather than focusing on innovation.

[0111] The following provides an example of an analysis process for user needs in formulating product improvement strategies according to the present invention, with the mobile phone market as the analysis object.

[0112] The online reviews and user purchase records used are from flagship stores of mobile phone brands on a large e-commerce platform and a major telecommunications operator. The first part extracts the product element dictionary, which contains 22 categories of product elements, as shown in Table (1).

[0113] Table (1) Mobile phone product element dictionary

[0114]

[0115]

[0116] According to the product element dictionary, the 28,198 reviews in the embodiment are structured, and the structured online reviews obtained are shown in Table (2).

[0117] Table (2) Structured online reviews

[0118]

[0119] Substitute the structured comments and corresponding comment scores into the joint analysis model Further substitute the estimated coefficient results into the formula The influence weights of product factors on user satisfaction are calculated, and the results are shown in Table (3).

[0120] Table (3) Factors and their weights of influence on user satisfaction

[0121]

[0122] The data obtained from the examples in the second part are shown in the following table (4).

[0123] Table (4) Statistics of user and product attribute data

[0124]

[0125]

[0126] Figure 3 is an example diagram of a two-stage utility model according to an embodiment of the present invention. Figure 3 As shown in Figure 5, the decision of users to change their mobile phones is divided into two stages. The first stage is the utility judgment of the old mobile phone. If the current mobile phone can no longer provide a positive utility, the user will change the mobile phone. The second stage represents the choice made by the user in the market. Among the competitive products provided by the market, the purchase with the greatest utility is selected. According to the user's purchase record, the user's two-stage decision, the user's attributes and the various attributes of the mobile phone are known items. They are substituted into the two-stage utility model and the Markov Monte Carlo (MCMC) method is used to solve the influence coefficient of process parameters on purchase tendency, as shown in Table (5).

[0127] Table (5) Influence coefficient of process parameters on purchase tendency

[0128]

[0129] Further, using and The weights of the factors for user purchasing propensity are calculated, as shown in Table (6).

[0130] Table (6) Factors and their weights of influence on user purchase tendency

[0131]

[0132] Figure 4 2 is an example diagram of dual weight model classification according to an embodiment of the present invention. Figure 4 As shown, the dual-weight model divides all product elements into four categories, representing four types of needs with different degrees of influence on user satisfaction and purchase intention. Substituting the results of Table (3) and Table (6) calculated in the embodiment into the model, we get Figure 5 result.

[0133] Figure 5 The figure is the classification result of the product elements corresponding to the embodiment of the present invention based on the dual weight model. As shown in the figure, the 22 product elements corresponding to the mobile phone products in this embodiment are divided into four categories according to the weights of their influence on satisfaction and purchase intention. Enterprises can formulate corresponding product improvement strategies comprehensively and specifically according to this classification result.

[0134] The entire data analysis process constructed by the present invention is presented in the form of pseudo code, as shown in Table (7).

[0135] Table (7) Pseudo code of data analysis process

[0136]

[0137]

[0138] The method of the present invention solves the differences in user preferences at different stages of purchase. The data analysis results are comprehensive. It provides product improvement guidance for product manufacturers by analyzing user needs based on comprehensive online comments and purchase records. It also provides support for the formulation of enterprise product improvement strategies through comprehensive data use and systematic data analysis. It is a user demand data analysis method for a feasible product improvement strategy.

[0139] The present invention is not limited to the above-mentioned embodiments. On the basis of the technical solution disclosed in the present invention, technicians in this field can make some substitutions and deformations to some technical features therein according to the disclosed technical content without creative labor, and these substitutions and deformations are all within the protection scope of the present invention.

Claims

1. A user demand data analysis method for product improvement strategy formulation, characterized in that: include: Obtain review data for target product categories, construct a product factor dictionary, and obtain satisfaction-based weights of product factors through joint analysis; Obtain user purchase records and product process parameters, use the two-stage utility model to obtain the influence coefficient, and use the constructed product element dictionary to obtain the weight of product elements based on user purchase tendency; Based on the dual-weight model, the product elements are classified using the weights of product elements based on satisfaction and product elements based on purchase intention, and targeted product improvement strategies are obtained; The calculation of product factors based on satisfaction weights includes: P11. Structuring the text of the target product category reviews; P12. Score the comments after text structure through the joint analysis model and calculate the weight of product elements based on satisfaction; The calculation of the weight based on the user's purchasing tendency is completed in the following way: P21. Use the two-stage utility model to obtain the influence coefficient of product process parameters on user purchase tendency; In the first stage of the model, the potential utility of users for the old product is modeled; A multinomial probit model is used to model the user's second-stage choice; P22. Use the correlation matrix to convert the influence coefficient of process parameters on user purchase tendency into the influence weight of product factors on user purchase tendency; The step P22 comprises the following steps: P221, conduct fuzzy evaluation of the correlation between factors and process parameters, the evaluation includes five-level scale variables: very weak, weak, medium, strong and very strong; P222. Calculate the weight of the influence of product elements on user purchasing tendency.

2. The method according to claim 1, characterized in that In step P11, text structuring in the target product category reviews is completed by the following method: P111, construct a product factor dictionary through a loop method, group and segment all comments, extract product function related words from the first group, add high-frequency words to the word library, delete the words that have been added to the word library from the remaining groups, and loop until all groups are processed to obtain a word library containing product function related words; Based on word2vec, the similarity between every two words in the vocabulary is calculated, and clustering is performed based on the similarity to obtain a product element dictionary. The structure of the obtained dictionary is: product element - words obtained in the review; P112, perform a reverse search on all reviews based on the product factor dictionary, and generate a review-factor relationship matrix based on whether the review contains the corresponding words in the factor dictionary; P113. For the comment-element relationship matrix, a sentiment dictionary is constructed using part-of-speech tagging and implicit Dirichlet allocation, and the sentiment of the elements contained in the comment is converted into a 0-1 matrix indicating whether the comment contains positive or negative features, thereby obtaining a structured comment text.

3. The method according to claim 1, characterized in that In step P12, the weights of product elements based on satisfaction are calculated by the following steps: The joint analysis model is: Among them, y i is the star rating given to review i, representing the overall satisfaction of review i; and are indicator variables for whether comment i evaluates factor j positively (pos) or negatively (neg), respectively; and They represent the impact of the two evaluations of factor j on the overall satisfaction; α is the intercept; n is the number of product factors in the constructed product factor dictionary; Substitute the coefficient results calculated by the joint analysis model into the formula Calculated That is, the influence weight of product element j on overall satisfaction.

4. The method according to claim 1, characterized in that: The two-stage utility model is as follows: In the first stage utility model, the potential utility of users for old products is modeled by the following formula: Among them, u i 1 represents the utility brought by the old product to user i; x ik is the kth process parameter of the old product of user i; b k is the influence coefficient of the kth process parameter on user utility; y i is the number of days that user i uses the old product until it is replaced; c is the coefficient of the number of days of use on user utility; α0 is the intercept term; ε i is an independent and identically distributed residual term, subject to N(0,σ ε 2 ); Second-stage utility model: define the user's second-stage utility and a utility benchmark u randomly selected from the choice set iJ The difference is: definition for The vector form is expressed as: Among them, ν i is the residual term vector, subject to N J-1 (0,Σ); The user's choice is represented as: Among them, Choice i represents the purchase choice made by user i in the second stage; j represents the jth product; u iJ is the utility of the benchmark product J to user i in the second stage; is the vector form of the utility set of all products of user i in the second stage.

5. The method according to claim 1, characterized in that Step P22 uses the following formula to calculate the influence weight of product factors on user purchase tendency: Among them, R jk represents the correlation coefficient between factor j and process parameter k; μ R is the membership function; r jk represents the fuzzy evaluation result of the correlation between factor j and process parameter k; represents the weight of product factor j on user purchase tendency; δ k is the mean of the influence coefficients of the process parameters after the aging period in the MCMC process.

6. The method according to claim 1, characterized in that The product elements are classified by using the obtained weights of product elements based on satisfaction and the weights of product elements based on purchase tendency. A graphical classification method is used: a dual-weight model is used to classify product elements. The horizontal and vertical axes of the model correspond to the influence weights of the elements based on user satisfaction and purchase tendency, respectively. According to the similarity and high and low levels of the weights based on user satisfaction and purchase tendency, the product elements are divided into four categories: Category I, Category II, Category III and Category IV. Category I: This type of factor has a great impact on user purchase intention and satisfaction. It is necessary to improve product functions and carefully consider innovation to improve user purchase utility and satisfaction after use. Category II: This type of factor has a greater impact on user purchase intention than satisfaction, and unsuccessful product function innovation will not reduce user satisfaction; Category III: This type of factor is not important for purchase intention and satisfaction. It is necessary to reduce investment in functions and seek cost-optimized alternatives. Category IV: This type of factor has a greater impact on satisfaction than on purchase intention. The overall quality of product functions should be improved to maintain and improve user satisfaction.

7. The method according to claim 6, characterized in that The calculation formula for the classification benchmark in the dual-weight model is: Among them, X is the horizontal axis coordinate, corresponding to the influence weight of the factor based on user satisfaction; Y is the vertical axis coordinate, corresponding to the influence weight of the factor based on user purchase tendency; α is the similarity level.

8. A user demand data analysis system for product improvement strategy formulation according to any one of claims 1 to 7, characterized in that: include: Online data collection device, used to collect product online review data and product process parameters; A review text mining device is used to perform text analysis on the collected review data to obtain a product element database and structured reviews; An element satisfaction analysis device is used to perform joint analysis on structured comments to obtain an element satisfaction measurement database; User purchase behavior collection device, used to collect user purchase records; The purchase behavior analysis device combines the product factor database with the two-stage utility model to obtain the factor-based purchase tendency measurement database; The dual-factor analysis device integrates the factor satisfaction measurement database and the factor-based purchase tendency measurement database, classifies and analyzes the factors based on the dual-weight model, and obtains the product improvement strategy database, providing product manufacturers with targeted product improvement strategies.

Images