Brand marketing system based on multi-algorithm fusion

Abstract

The invention relates to the technical field of brand marketing, in particular to a brand marketing system based on multi-algorithm fusion, which comprises a data collection module, a data processing module, a market segmentation module, a user portrait generation and evaluation module and a marketing strategy generation module, and is characterized in that the data collection module collects historical user behavior data and real-time user behavior data; the data processing module performs feature engineering on the historical user behavior data to generate initial user behavior features; the market segmentation module carries out clustering analysis on the users according to the initial user behavior characteristics, generates market clustering results and evaluates the market clustering results; a user portrait generation and evaluation module generates a user portrait of each market according to the market clustering result, and calculates a mean square error; wherein the evaluation result and the mean square error determine whether to adjust the data collection module; and the marketing strategy generation module generates a personalized marketing strategy based on collaborative filtering of the user according to the user portrait.

Description

Technical Field

[0001] The present invention relates to the field of brand marketing technology, and in particular to a brand marketing system based on multi-algorithm fusion. Background Art

[0002] Traditional brand marketing systems usually rely on static data analysis and human experience, lacking dynamic interaction and real-time feedback mechanisms, resulting in low accuracy and personalization of marketing strategies.

[0003] First, traditional systems rely more on historical data in data collection and lack the capture and analysis of real-time user behavior data, resulting in insufficient timeliness and diversity of data.

[0004] Secondly, data processing is limited to simple data cleaning and preliminary feature selection, lacking in-depth feature engineering processing, making it difficult to ensure the quality and consistency of data, which in turn affects the accuracy of subsequent analysis. In terms of market segmentation, traditional systems may rely on simple classification methods rather than refined processing based on cluster analysis and silhouette coefficient evaluation, resulting in inaccurate and ineffective user group segmentation. In the process of generating user portraits, traditional systems often rely on fixed feature settings, lack weighted processing of feature importance, and cannot dynamically adjust feature weights, resulting in low accuracy of user portraits and difficulty in truly reflecting user behavior and preferences.

[0005] In addition, in terms of marketing strategy generation, traditional systems lack personalized recommendation mechanisms based on collaborative filtering and user similarity calculations, and often adopt a broad-based marketing approach, resulting in waste of resources and poor user experience.

[0006] In general, traditional brand marketing systems are unable to achieve high-precision and high-reliability user analysis and market segmentation due to the lack of multi-algorithm fusion and dynamic feedback mechanisms. The personalization and optimization levels of marketing strategies are also obviously insufficient, making it difficult to effectively improve the brand's market competitiveness and user satisfaction. Summary of the invention

[0007] The purpose of the present invention is to provide a brand marketing system based on multi-algorithm fusion to solve the problems raised in the above background technology.

[0008] To achieve the above purpose, the present invention provides the following technical solution: a brand marketing system based on multi-algorithm fusion, including a data collection module, a data processing module, a market segmentation module, a user portrait generation and evaluation module and a marketing strategy generation module, wherein:

[0009] The data collection module collects historical user behavior data and real-time user behavior data;

[0010] The data processing module performs feature engineering on the historical user behavior data, wherein the feature engineering includes data cleaning, feature selection, feature transformation, and constructing composite features to generate initial user behavior features;

[0011] The market segmentation module performs cluster analysis on users according to the initial user behavior characteristics, generates market clustering results, and uses the silhouette coefficient to evaluate the market clustering results, wherein:

[0012] If the silhouette coefficient evaluation fails, adjusting the feature engineering in the data processing module;

[0013] If the silhouette coefficient is evaluated as qualified, the market clustering result is sent to the user portrait generation and evaluation module;

[0014] The user portrait generation and evaluation module generates a user portrait for each market according to the market clustering result, and calculates the mean square error between the real-time user behavior data and the user portrait, where:

[0015] If the mean square error evaluation fails, adjusting the feature engineering in the data processing module;

[0016] If the mean square error evaluation is qualified, the user portrait is sent to a marketing strategy generation module;

[0017] The marketing strategy generation module generates a personalized marketing strategy based on the user portrait and the user's collaborative filtering.

[0018] As a further improvement of the present technical solution, the data processing module performs feature engineering on historical user behavior data to generate initial user behavior features by filling missing values, identifying and deleting outliers, and standardizing numerical features; using the Pearson correlation coefficient to screen features that are highly correlated with the target variable; performing logarithmic transformation and square root transformation; constructing new features through arithmetic operations of addition and multiplication, and finally generating initial user behavior features.

[0019] As a further improvement of the present technical solution, the market segmentation module includes a cluster analysis unit, which performs cluster analysis on users according to initial user behavior characteristics to generate market clustering results, and classifies samples into K clusters by taking the initial user behavior characteristics as samples and minimizing the total square distance between samples within the group.

[0020] As a further improvement of the technical solution, the market segmentation module includes a clustering result evaluation unit, which evaluates the market clustering result and evaluates the clustering quality through the silhouette coefficient. The silhouette coefficient formula is as follows:

[0021] ,in It is a sample The average distance to other data points in the same cluster, It is a sample The average distance to all data points in the nearest neighbor cluster.

[0022] As a further improvement of the technical solution, the process of the clustering result evaluation unit evaluating the clustering quality according to the silhouette coefficient is as follows:

[0023] Calculate the individual silhouette coefficient for each market clustering result, and then take the average as the silhouette coefficient of the entire cluster;

[0024] A silhouette coefficient threshold is set, and according to the judgment logic of the silhouette coefficient and the silhouette coefficient threshold, it is determined whether to adjust the feature engineering of the data processing module, wherein:

[0025] If the silhouette coefficient is lower than the silhouette coefficient threshold, it means that the silhouette coefficient assessment is unqualified. The features with large differences are analyzed, and the similarity index of the samples in each cluster is calculated; the distance between each sample and the group is recorded; for samples with large differences within the group, the performance of these samples in each feature dimension is counted to generate a feature contribution list; the data features with the highest and lowest contribution in the feature contribution list are sent to the data processing module;

[0026] If the silhouette coefficient is greater than or equal to the silhouette coefficient, it means that the silhouette coefficient evaluation is qualified, and the market clustering result is sent to the user portrait generation and evaluation module.

[0027] As a further improvement of the present technical solution, the user portrait generation and evaluation module includes a user portrait generation unit, which generates a user portrait for each market according to the market clustering result, specifically including:

[0028] The importance score is obtained by calculating the dispersion of features in different clusters; features with high scores indicate that they contribute more to distinguishing different clusters, while features with low scores do the opposite;

[0029] For each cluster, a weighted average method is used to generate a user profile according to the dimension setting of the user profile. For all users belonging to the cluster, the weighted average of each feature is calculated, and the weight is based on the importance score of the feature;

[0030] The weighted average result is used to represent the user profile of each cluster.

[0031] As a further improvement of the present technical solution, the user portrait generation and evaluation module includes a user portrait evaluation unit, which receives the real-time user behavior data from the data collection module, and compares the expected behavior of the user portrait with the actual behavior of the real-time user behavior data, and compares whether they are consistent by calculating the mean square error. The comparison results of these features are calculated for each user, where the mean square error is The formula is as follows:

[0032] ,in is the total number of users, It is The expected feature value of a user in the user portrait, It is The actual behavior characteristic value of a user.

[0033] As a further improvement of the technical solution, the user portrait evaluation unit calculates the mean square error , determine whether to adjust the feature engineering of the data processing module, including:

[0034] Set a Threshold, calculated by comparison With the setting Threshold, where:

[0035] like Less than the set Threshold, indicating that the evaluation is qualified, then the user portrait is sent to the marketing strategy generation module;

[0036] like Greater than or equal to the set Threshold, the data features with the highest and lowest importance scores in the feature contribution list in the user portrait generation unit are sent to the data processing module.

[0037] As a further improvement of the technical solution, the data processing module performs feature engineering adjustments based on the data features with the highest and lowest contribution sent by the clustering result evaluation unit and the data features with the highest and lowest importance scores sent by the user portrait evaluation unit, specifically including:

[0038] Increase the weight of the data features with the highest contribution and remove the data features with the lowest contribution;

[0039] The weight of the data feature with the highest importance score is increased, and the data feature with the lowest importance score is removed.

[0040] As a further improvement of the technical solution, the marketing strategy generation module generates personalized marketing strategies based on user portraits and user collaborative filtering, specifically including:

[0041] Calculate the similarity between the target user and other users;

[0042] Based on the similarity ranking, select a group of users who are most similar to the target user;

[0043] Analyze the historical behaviors and interest preferences of similar users to predict the brands and services that the target user likes. The prediction method is based on the weighted average ratings of neighboring users, with the weights determined by the similarity.

[0044] Generate a personalized recommendation list containing brands and marketing strategies that target users are interested in.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] 1. The brand marketing system based on multi-algorithm fusion uses data collection module, data processing module, market segmentation module and user portrait generation and evaluation module in turn to build a high-precision and high-reliability brand marketing system through multi-algorithm fusion. First, in the data collection module, historical user behavior data and real-time user behavior data are fully captured to ensure the diversity and timeliness of the data; next, the data processing module performs in-depth processing of the data through data cleaning, feature selection, feature transformation and composite feature construction to ensure the quality and consistency of the data and improve the accuracy of subsequent analysis; in the market segmentation module, through clustering analysis and silhouette coefficient evaluation, the system further optimizes the clustering results, maximizes the similarity within the group, and ensures the accuracy and effectiveness of market segmentation; then, after receiving the market clustering results, the user portrait generation and evaluation module generates a high-precision user portrait through weighted average calculation. The marketing strategy generation module based on user portrait and collaborative filtering algorithm designs personalized marketing strategies by calculating user similarity, finding similar user groups, predicting scores and generating recommendation lists, effectively improving the brand's market competitiveness and user satisfaction.

[0047] 2. The data collection module in the brand marketing system based on multi-algorithm fusion ensures the high accuracy and reliability of user clustering and portrait generation through dynamic interaction between the market segmentation module and the user portrait generation and evaluation module; specifically, the market segmentation module selects accurate user groups through clustering analysis and silhouette coefficient evaluation; if the evaluation is unqualified, the data processing module immediately performs feature engineering adjustments to optimize the weights and screening of features and improve the effect of subsequent clustering; if the evaluation is qualified, the clustering results are passed to the user portrait generation and evaluation module; secondly, the user portrait generation and evaluation module further verifies and corrects the accuracy of the user portrait by comparing the mean square error between real-time data and expected behavior data; if the mean square error evaluation is unqualified, the data processing module will also readjust specific features to ensure that the user portrait is closer to reality; this cyclic optimization and feedback mechanism significantly improves the accuracy of the entire system in identifying and understanding user behavior, provides high-quality basic data for the marketing strategy generation module, and ultimately achieves the optimization of personalized marketing strategies, effectively improving the brand's market competitiveness and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the overall module of the present invention;

[0049] Figure 2 It is a schematic diagram of the market segmentation module unit of the present invention;

[0050] Figure 3 Schematic diagram of the user portrait generation and evaluation module unit of the present invention.

[0051] In the figure: 100, data collection module; 200, data processing module; 300, market segmentation module; 301, clustering analysis unit; 302, clustering result evaluation unit; 400, user portrait generation and evaluation module; 401, user portrait generation unit; 402, user portrait evaluation unit; 500, marketing strategy generation module. DETAILED DESCRIPTION

[0052] The following will be combined with the accompanying drawings in the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] See also Figure 1-3 The present invention provides a technical solution: a brand marketing system based on multi-algorithm fusion, including a data collection module 100, a data processing module 200, a market segmentation module 300, a user portrait generation and evaluation module 400 and a marketing strategy generation module 500.

[0054] The data collection module 100 collects historical user behavior data and real-time user behavior data, wherein the historical user behavior data includes but is not limited to e-commerce platform user data, customer service data and website application data; the real-time user behavior data includes but is not limited to real-time transaction data, real-time interaction data and real-time application data.

[0055] The data processing module 200 performs feature engineering on the historical user behavior data to generate initial user behavior features, specifically including:

[0056] Data cleaning: fill in missing values ​​or delete records with a large number of missing values; identify and correct or delete outliers; standardize or normalize numerical features so that each feature value is within a similar scale range to eliminate dimensional differences;

[0057] Feature selection: Use the Pearson correlation coefficient to select features that are highly correlated with the target variable;

[0058] Feature transformation: Logarithmic transformation, square root transformation, standardization and other processes are used to ensure that the features meet the assumptions of the clustering algorithm;

[0059] Constructing composite features: constructing new features through arithmetic operations of addition and multiplication as initial user behavior features.

[0060] Data processing module 200: by filling missing values, identifying and correcting outliers, standardizing and normalizing, etc., the quality and consistency of data are ensured, and the accuracy of subsequent analysis is improved; the Pearson correlation coefficient is used to screen features that are highly correlated with the target variable, reducing feature redundancy and improving the efficiency and accuracy of the model; by logarithmic transformation, square root transformation and standardization, it is ensured that the features meet the assumptions of the clustering algorithm and improve the effectiveness of the clustering results; by constructing new combined features, the feature set of the model is enriched, and the depth and breadth of the analysis are further improved.

[0061] The cluster analysis unit 301 in the market segmentation module 300 performs cluster analysis on users according to the initial user behavior characteristics to generate market clustering results. The initial user behavior characteristics are used as samples to minimize the total square distance between samples in the group to classify the samples into K clusters, specifically including:

[0062] Initialization: Randomly select K initial cluster centers;

[0063] Assign samples: assign each sample to the cluster center closest to it;

[0064] Update cluster centers: Calculate the centroid (average) of all samples in each cluster and use the centroid as the new cluster center;

[0065] Repeat: Repeat the assignment and update steps until the cluster centers no longer change or the maximum number of iterations is reached.

[0066] The clustering result evaluation unit 302 in the market segmentation module 300 evaluates the market clustering result in the clustering analysis unit 301 and evaluates the clustering quality through the silhouette coefficient. The silhouette coefficient formula is as follows:

[0067] ,in It is a sample The average distance to other data points in the same cluster, It is a sample The average distance to all data points in the nearest neighbor cluster.

[0068] The process of clustering result evaluation unit 302 evaluating clustering quality according to the silhouette coefficient is as follows:

[0069] Calculate the individual silhouette coefficient for each clustering result, and then take the average as the silhouette coefficient of the entire cluster;

[0070] Set a predefined satisfactory threshold, for example, the silhouette coefficient threshold is set to 0.5, and judge whether the clustering result is satisfactory based on this threshold;

[0071] The clustering result evaluation unit 302 determines whether to perform feature engineering adjustment on the data processing module 200 according to the judgment logic of the silhouette coefficient and the silhouette coefficient threshold, specifically including:

[0072] If the silhouette coefficient is lower than the silhouette coefficient threshold, it means that the clustering result is not good and the silhouette coefficient evaluation is unqualified. It may be that some features fail to fully distinguish users. Analyze the features with large internal differences in certain groups and determine that the importance of these features needs to be adjusted, that is, increase or adjust the weights of the corresponding features. The analysis process is as follows:

[0073] Calculate the similarity index of the samples in each cluster, such as Euclidean distance, cosine similarity, etc.; record the distance between each sample and the group center (cluster centroid), such as RMSE (root mean square error) to measure the average distance between the sample and the centroid; for samples with large differences within the group, count the performance of these samples in various feature dimensions, and generate a feature contribution list, that is, the difference contribution value of each feature in different samples; send the data features with the highest and lowest contribution in the feature contribution list to the data processing module 200, where:

[0074] The data processing module 200 adjusts the data features with the highest and lowest contribution sent by the clustering result evaluation unit 302 by feature engineering, increases the weight of the data feature with the highest contribution, and increases its influence in the subsequent clustering process by multiplying the feature by a higher weight coefficient; removes the data feature with the lowest contribution to reduce its influence in the subsequent clustering process;

[0075] If the silhouette coefficient is greater than or equal to the silhouette coefficient, it means that the clustering effect is evaluated as qualified, and the user portrait generation and evaluation module 400 is performed.

[0076] The user portrait generation unit 401 in the user portrait generation and evaluation module 400 generates a user portrait for each market according to the market clustering result of the market segmentation module 300, specifically including:

[0077] Receive the market clustering result of the market segmentation module 300, where each clustering label represents a user segment group;

[0078] The importance score is obtained by calculating the dispersion of features in different clusters; features with high scores indicate that they contribute more to distinguishing different clusters, while features with low scores do the opposite;

[0079] For each cluster, a weighted average method is used to generate a user profile according to the dimension setting of the user profile. For all users belonging to the cluster, the weighted average of each feature is calculated, and the weight is based on the importance score of the feature;

[0080] Using the weighted average results, the user profile of each cluster is represented in multiple dimensions, for example:

[0081] Demographic characteristics: Use the weighted average results of demographic characteristics to describe the user's age group, gender distribution, etc.

[0082] Behavioral characteristics: express the user's behavior pattern through the weighted average results of dimensions such as purchase frequency and average purchase amount;

[0083] Interest preference: Analyze the weighted average results of brand category visit frequency and other characteristics to show the user's interest preference.

[0084] The user portrait evaluation unit 402 in the user portrait generation and evaluation module 400 receives the real-time user behavior data from the data collection module 100, and compares the expected behavior of the user portrait with the actual behavior of the real-time user behavior data, and compares whether they are consistent by calculating the mean square error. For example, there are feature A (the expected number of purchases of the user portrait) and feature B (the actual number of purchases of the real-time user behavior data), and the comparison results of these features are calculated for each user, where the mean square error is The formula is as follows:

[0085] ,in is the total number of users, It is The expected feature value of a user in the user portrait, It is The actual behavior characteristic value of a user.

[0086] The user profile evaluation unit 402 calculates the mean square error , determining whether to perform feature engineering adjustment on the data processing module 200, specifically including:

[0087] Set a Threshold, indicating the reasonable error range of user portrait;

[0088] Compare the calculated With the setting Threshold, used to determine whether the accuracy of the image meets the standard, where:

[0089] like Less than the set The threshold value indicates that the evaluation is qualified, indicating that the expected behavior of the user profile is consistent with the actual behavior, and the marketing strategy generation module 500 is performed;

[0090] like Greater than or equal to the set The threshold indicates that there is a large error in the portrait, and the data features with the highest and lowest importance scores in the feature contribution list in the user portrait generation unit 401 are sent to the data processing module 200, where:

[0091] The data processing module 200 will perform feature engineering adjustments on the data features with the highest and lowest importance scores sent by the user portrait generation unit 401, increase the weight of the data feature with the highest importance score, and increase its influence in the subsequent clustering process by multiplying the feature by a higher weight coefficient; and eliminate the data feature with the lowest importance score to reduce its influence in the subsequent clustering process.

[0092] After the data processing module 200 performs feature engineering adjustment on the data features with the highest and lowest contribution sent by the clustering result evaluation unit 302, and performs feature engineering adjustment on the data features with the highest and lowest importance scores sent by the user portrait generation unit 401, the silhouette coefficient and until the assessment is qualified.

[0093] The data processing module 200 ensures the high accuracy and reliability of user clustering and portrait generation through the dynamic interaction between the market segmentation module 300 and the user portrait generation and evaluation module 400. Specifically, first, the market segmentation module 300 selects accurate user groups through cluster analysis and contour coefficient evaluation. If the evaluation is unqualified, the data processing module 200 immediately adjusts the feature engineering, optimizes the weight and screening of the features, and improves the effect of subsequent clustering. If the evaluation is qualified, the clustering results are passed to the user portrait generation and evaluation module 400. Secondly, the user portrait generation and evaluation module 400 further verifies and corrects the accuracy of the user portrait by comparing the mean square error of real-time data with the expected behavior data. If the mean square error evaluation is unqualified, the data processing module 200 will also readjust the specific features to ensure that the user portrait is closer to reality. This cyclic optimization and feedback mechanism significantly improves the accuracy of the entire system in identifying and understanding user behavior, provides high-quality basic data for the marketing strategy generation module 500, and ultimately achieves the optimization of personalized marketing strategies, effectively improving the market competitiveness and user satisfaction of the brand.

[0094] The marketing strategy generation module 500 generates personalized marketing strategies based on user portraits and user collaborative filtering, specifically including:

[0095] Calculate user similarity: Calculate the similarity between the target user and other users. Use the user's behavioral characteristics (such as purchase frequency and purchase amount) and interest preferences (brand category visit frequency) to calculate the similarity. Commonly used methods include cosine similarity and Pearson correlation coefficient.

[0096] Find similar user groups: Select a group of users that are most similar to the target user (i.e., neighboring users) based on similarity ranking;

[0097] Predictive rating: Analyze the historical behaviors (such as purchase records) and interest preferences of these similar users to predict the brands or services that the target user may like. The prediction method is based on the weighted average ratings of neighboring users, with the weights determined by the similarity.

[0098] Generate recommendation list: Generate a personalized recommendation list containing brands and marketing strategies that target users are interested in.

[0099] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A brand marketing system based on multi-algorithm fusion, characterized by: The system comprises a data collection module (100), a data processing module (200), a market segmentation module (300), a user portrait generation and evaluation module (400) and a marketing strategy generation module (500), wherein: The data collection module (100) collects historical user behavior data and real-time user behavior data; The data processing module (200) performs feature engineering on the historical user behavior data, wherein the feature engineering includes data cleaning, feature selection, feature transformation, and constructing composite features to generate initial user behavior features; The market segmentation module (300) performs cluster analysis on users according to the initial user behavior characteristics, generates market clustering results, and uses the silhouette coefficient to evaluate the market clustering results, wherein: If the silhouette coefficient evaluation fails, adjusting the feature engineering in the data processing module (200); If the silhouette coefficient evaluation is qualified, the market clustering result is sent to the user portrait generation and evaluation module (400); The user portrait generation and evaluation module (400) generates a user portrait for each market according to the market clustering result, and calculates the mean square error between the real-time user behavior data and the user portrait, wherein: If the mean square error evaluation fails, adjusting the feature engineering in the data processing module (200); If the mean square error evaluation is qualified, the user portrait is sent to a marketing strategy generation module (500); The marketing strategy generation module (500) generates a personalized marketing strategy based on the user's collaborative filtering according to the user portrait.

2. The brand marketing system based on multi-algorithm fusion according to claim 1 is characterized in that: The data processing module (200) performs feature engineering on the historical user behavior data to generate initial user behavior features by filling in missing values, identifying and deleting outliers, and standardizing numerical features; and uses the Pearson correlation coefficient to screen features that are highly correlated with the target variable; Through logarithmic transformation and square root transformation; new features are constructed through arithmetic operations of addition and multiplication, and finally the initial user behavior features are generated.

3. The brand marketing system based on multi-algorithm fusion according to claim 2 is characterized in that: The market segmentation module (300) comprises a cluster analysis unit (301), which performs cluster analysis on users according to initial user behavior characteristics to generate market clustering results, and classifies samples into K clusters by minimizing the total square distance between samples in a group by taking the initial user behavior characteristics as samples.

4. The brand marketing system based on multi-algorithm fusion according to claim 3 is characterized in that: The market segmentation module (300) comprises a clustering result evaluation unit (302), wherein the clustering result evaluation unit (302) evaluates the market clustering result and evaluates the clustering quality through a silhouette coefficient. The silhouette coefficient formula is as follows: ,in It is a sample The average distance to other data points in the same cluster, It is a sample The average distance to all data points in the nearest neighbor cluster.

5. The brand marketing system based on multi-algorithm fusion according to claim 4 is characterized in that: The process of the clustering result evaluation unit (302) evaluating the clustering quality according to the silhouette coefficient is as follows: Calculate the individual silhouette coefficient for each market clustering result, and then take the average as the silhouette coefficient of the entire cluster; A silhouette coefficient threshold is set, and according to the judgment logic between the silhouette coefficient and the silhouette coefficient threshold, it is determined whether to perform feature engineering adjustment on the data processing module (200), wherein: If the silhouette coefficient is lower than the silhouette coefficient threshold, it means that the silhouette coefficient evaluation is unqualified, analyze the features with large differences, and calculate the similarity index of the samples in each cluster; record the distance between each sample and the group; for samples with large differences within the group, count the performance of these samples in each feature dimension and generate a feature contribution list; send the data features with the highest and lowest contribution in the feature contribution list to the data processing module (200); If the silhouette coefficient is greater than or equal to the silhouette coefficient, it means that the silhouette coefficient evaluation is qualified, and the market clustering result is sent to the user portrait generation and evaluation module (400).

6. The brand marketing system based on multi-algorithm fusion according to claim 5 is characterized in that: The user portrait generation and evaluation module (400) comprises a user portrait generation unit (401), wherein the user portrait generation unit (401) generates a user portrait for each market according to the market clustering result, specifically comprising: The importance score is obtained by calculating the dispersion of features in different clusters; features with high scores indicate that they contribute more to distinguishing different clusters, while features with low scores do the opposite; For each cluster, a weighted average method is used to generate a user profile according to the dimension setting of the user profile. For all users belonging to the cluster, the weighted average of each feature is calculated, and the weight is based on the importance score of the feature; The weighted average result is used to represent the user profile of each cluster.

7. The brand marketing system based on multi-algorithm fusion according to claim 6 is characterized in that: The user portrait generation and evaluation module (400) includes a user portrait evaluation unit (402), which receives the real-time user behavior data from the data collection module (100), and compares the expected behavior of the user portrait with the actual behavior of the real-time user behavior data, and compares whether they are consistent by calculating the mean square error, and calculates the comparison result of these features for each user, wherein the mean square error The formula is as follows: ,in is the total number of users, It is The expected feature value of a user in the user portrait, It is The actual behavior characteristic value of a user.

8. The brand marketing system based on multi-algorithm fusion according to claim 7 is characterized in that: The user profile evaluation unit (402) calculates the mean square error , determining whether to perform feature engineering adjustment on the data processing module (200), specifically including: Set a Threshold, calculated by comparison With the setting Threshold, where: like Less than the set The threshold indicates that the evaluation is qualified, and the user portrait is sent to the marketing strategy generation module (500); like Greater than or equal to the set Threshold, the data features with the highest and lowest importance scores in the feature contribution list in the user portrait generation unit (401) are sent to the data processing module (200).

9. The brand marketing system based on multi-algorithm fusion according to claim 7 is characterized in that: The data processing module (200) performs feature engineering adjustments based on the data features with the highest and lowest contribution sent by the clustering result evaluation unit (302) and the data features with the highest and lowest importance scores sent by the user portrait evaluation unit (402), specifically including: Increase the weight of the data features with the highest contribution and remove the data features with the lowest contribution; The weight of the data feature with the highest importance score is increased, and the data feature with the lowest importance score is removed.

10. The brand marketing system based on multi-algorithm fusion according to claim 8, characterized in that: The marketing strategy generation module (500) generates a personalized marketing strategy based on user portraits and user collaborative filtering, specifically including: Calculate the similarity between the target user and other users; Based on the similarity ranking, select a group of users who are most similar to the target user; Analyze the historical behaviors and interest preferences of similar users to predict the brands and services that the target user likes. The prediction method is based on the weighted average ratings of neighboring users, with the weights determined by the similarity. Generate a personalized recommendation list containing brands and marketing strategies that target users are interested in.

Images