Brand marketing system based on big data

Abstract

The invention relates to the technical field of brand marketing analysis, in particular to a brand marketing system based on big data, which comprises a commodity data acquisition module, a commodity image feature analysis module and a commodity image feature analysis module, the commodity image feature analysis module extracts image contour features of the marketing brand and the commodity image in the current data, sets a commodity similarity threshold, calculates the Euclidean distance between different image contour features, and deletes similar commodities in the current data according to the Euclidean distance and the commodity similarity threshold. Similar commodities are different in brand, the same in appearance and similar in function, and by deleting the similar commodities, it is avoided that similar commodities appear in the marketing commodity set and the browsing commodity set, and follow-up analysis of the related commodities by the commodity correlation analysis module is affected.

Description

Technical Field

[0001] The present invention relates to the technical field of brand marketing analysis, and in particular to a brand marketing system based on big data. Background Art

[0002] In the current e-commerce environment, marketing brands usually use e-commerce platforms to sell products, which can obtain massive e-commerce data. Due to the wide variety of products and different uses, users often buy other suitable products together during the purchase process;

[0003] At present, brand marketing systems mainly rely on analyzing existing e-commerce market data to determine the products that the current marketing brand needs to replenish inventory and provide suggestions for the brand to produce different products. However, if we only rely on analyzing existing e-commerce market data, we cannot analyze the correlation between different products due to the potential matching relationship between different products in the sales process and the association logic behind user purchasing behavior. This makes it impossible for brands to accurately grasp market demand when formulating production strategies, which easily leads to a disconnect between production and actual market demand. In view of this, we propose a brand marketing system based on big data. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of being unable to analyze the association between different commodities in a marketing brand.

[0005] To achieve the above-mentioned purpose, the present invention provides a brand marketing system based on big data, including a commodity data collection module, a commodity image feature analysis module and a commodity correlation analysis module;

[0006] The commodity data collection module sends identity authentication information and browsing data acquisition information to the e-commerce platform. After receiving the identity authentication information, the e-commerce platform verifies the identity authentication information. After the verification is passed, a connection is established with the e-commerce platform, and the corresponding user browsing record data is queried according to the browsing data acquisition information.

[0007] The product image feature analysis module uses the order generation time as the boundary to define historical data and current data, and is used to perceive the product information corresponding to the marketing brand. It uses a feature extraction method to extract image contour features of the marketing brand and the product images in the current data, and defines them as a marketing image contour feature set and a product image contour feature set, respectively. It calculates the Euclidean distance between different image contour features, and uses a scatter plot analysis method to set a product similarity threshold.

[0008] If the Euclidean distance between two image contour features is greater than the product similarity threshold, the two image contour features are determined to be similar products, and the image contour feature corresponding to any similar product is randomly deleted. Otherwise, the two image contour features in the marketing image contour features are determined not to be similar products, and both image contour features are retained.

[0009] If all the corresponding commodities in the user's historical data are included in the commodities in the marketing commodity set, the commodity correlation analysis module defines the corresponding user as an accurate user, senses the marketing commodity set and the browsed commodity set corresponding to the accurate user, and calculates the intersection and difference of the browsed commodity set of each accurate user and the commodities in the marketing commodity set;

[0010] According to the number of times different product combinations appear in the marketing product set, the correlation between different products in the marketing brand is calculated, the products corresponding to the difference set are defined as missing products, and the correlation between the missing products and different products in the marketing product set is calculated.

[0011] As a further improvement of the present technical solution, the working principle of the commodity data collection module for verifying the identity authentication information is as follows: the e-commerce platform compares the identity authentication information sent by the commodity data collection module with the legal key list pre-stored in the e-commerce platform database character by character. If they are completely consistent, the verification is passed and a connection is established; if there is a difference, an error message is returned and the connection is refused.

[0012] As a further improvement of the technical solution, the feature extraction method in the commodity image feature analysis module determines the strength and direction of the edge by calculating the gradient of the commodity image in the horizontal and vertical directions, and extracts the image contour features in the commodity image;

[0013] The feature extraction method uses two convolution kernels, one for detecting horizontal edges and the other for detecting vertical edges. These two convolution kernels are convolved with the image to obtain the approximate gradient values ​​of the product image in the horizontal and vertical directions. The gradient amplitude and direction of each pixel are calculated based on the gradient value. The specific calculation formula is as follows:

[0014] Horizontal convolution kernel for feature extraction ;

[0015] Vertical convolution kernel for feature extraction ;

[0016] For each pixel in the product image , its gradient approximation in the horizontal direction and the vertical gradient approximation It is obtained by convolving the convolution kernel with the pixel and its surrounding pixels;

[0017] Taking the horizontal direction as an example, the calculation formula is: ;

[0018] in, It is the product image in pixels The grayscale value at , the calculation in the vertical direction is the same;

[0019] After obtaining the approximate values ​​of the horizontal and vertical gradients, the pixel points are calculated using the following formula The gradient magnitude and direction :

[0020] Gradient Magnitude: ;

[0021] direction: .

[0022] As a further improvement of the technical solution, the product image feature analysis module calculates the Euclidean distance between two image contour features using the following calculation formula:

[0023] Perceive the corresponding image contour features of two product images and , and The Euclidean distance between The calculation formula is: .

[0024] As a further improvement of the technical solution, the working principle of setting the commodity similarity threshold by the scatter plot analysis method in the commodity image feature analysis module is as follows:

[0025] The scatter plot analysis method is to display each data point in the data set in the form of coordinates on a two-dimensional plane. In the product similarity analysis, the Euclidean distance between the contour features of different product images is used as the coordinate value. By observing the distribution of scatter points, the aggregation trend and degree of dispersion of the data can be intuitively seen.

[0026] As a further improvement of the technical solution, the specific steps of determining the commodity similarity threshold are as follows:

[0027] For each Euclidean distance data point , calculate its local density , the formula is:

[0028] ;

[0029] in Yes and The Euclidean distance between is a cutoff distance used to control the range of density calculation. is the total number of data points;

[0030] Calculate the distance from each point to the closest point among the points with higher density than it ,Right now ;

[0031] Will and Plot on a scatter plot and look for high local density and greater distance The point is the product similarity threshold.

[0032] As a further improvement of this technical solution, the commodity association analysis module perceives the marketing commodity set as ,have Precise user, The browsed product collection of precise users is , , then calculate the two commodities and The steps of the correlation index are as follows:

[0033] Step 1: Calculate the products and Number of times they appear in the intersection at the same time , for every precise user ,like and ,but Add 1 and traverse all After the precise users, value;

[0034] Step 2: Calculate the included products The number of intersection sets , that is, for all precise users ,like ,but Add 1;

[0035] Products and The correlation index is: , the correlation index calculation formula reflects the In the case of The higher the probability of occurrence, the stronger the correlation between the two.

[0036] As a further improvement of this technical solution, the commodity association analysis module perceives the marketing commodity set as , a precise user, The browsed product collection of precise users is , then the difference set corresponding to the accurate user , the elements in the difference set are the missing products;

[0037] For each missing item and products in the marketing product collection , statistics on the browsing history of all accurate users, missing products and marketing products The number of co-occurrences is recorded as The specific calculation method is: traverse all the browsed product sets of accurate users, if a set contains and ,but Add 1;

[0038] Calculate correlation: The total number of known accurate users is , missing items and marketing products Relevance The calculation formula is: .

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] In a brand marketing system based on big data, the image contour features of the marketing brand and the product images in the current data are respectively extracted through a product image feature analysis module, and a product similarity threshold is set, and the Euclidean distance between different image contour features is calculated. According to the Euclidean distance and the product similarity threshold, similar products in the current data are deleted, so as to construct a marketing product set and a browsing product set. Similar products are products with different brands but the same appearance and similar functions. By deleting similar products, similar products are avoided from appearing in the marketing product set and the browsing product set, which affects the subsequent product correlation analysis module to analyze the mutually related products.

[0041] Then, the product correlation analysis module is used to analyze the intersection and difference of the products in the marketing product set and the browsed product set. The correlation between different products in the marketing brand is analyzed based on the intersection, and the correlation between products that do not exist in the marketing brand is analyzed based on the difference. Through different correlations, the brand can reasonably arrange production resources according to the demand for related products, adjust the production proportion and production plan of different products, increase production input for product combinations with strong correlation and strong demand; appropriately reduce production for products with low correlation and low demand to avoid waste of resources and inventory backlogs. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The overall module schematic diagram of the present invention is

[0043] Figure 2 This is a schematic diagram of a commodity image feature analysis module of the present invention;

[0044] Figure 3 This is a schematic diagram of the commodity correlation analysis module of the present invention.

[0045] The meaning of each number in the figure is:

[0046] 100. Commodity data collection module; 200. Commodity image feature analysis module; 300. Commodity correlation analysis module. DETAILED DESCRIPTION

[0047] 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.

[0048] A brand marketing system based on big data, comprising a commodity data collection module 100, a commodity image feature analysis module 200 and a commodity relevance analysis module 300;

[0049] The commodity data collection module 100 sends identity authentication information and browsing data acquisition information to the e-commerce platform. After receiving the identity authentication information, the e-commerce platform verifies the identity authentication information. After the verification is passed, the commodity data collection module 100 establishes a connection with the e-commerce platform;

[0050] At the same time, the e-commerce platform obtains information based on the browsing data, queries the corresponding user browsing record data in its database, and transmits the queried user browsing record data to the commodity data collection module 100 according to the established connection;

[0051] User browsing history data includes information such as product category (electronic products, clothing, and food, etc.), product name, number of product views, product brand, product image, and product price.

[0052] The working principle of the commodity data collection module 100 for verifying identity authentication information is as follows:

[0053] The e-commerce platform compares the identity authentication information sent by the commodity data collection module 100 with the legal key list pre-stored in the e-commerce platform database character by character. If they are completely consistent, the verification is passed and the connection is established; if there is a difference, an error message is returned and the connection is refused.

[0054] The product image feature analysis module 200 is used to perceive the product information corresponding to the marketing brand (product information includes product images, product prices, etc.). The product information corresponding to the marketing brand is provided by the staff corresponding to the marketing brand. The feature extraction method is used to extract the image contour features of the product images in the marketing brand and current data (the current data is based on the order generation time. Before the order is generated, the user's browsing record data on the e-commerce platform is the historical data. After the order is generated, the corresponding browsing record is the current data), and they are defined as a marketing image contour feature set and a product image contour feature set, respectively.

[0055] The feature extraction method in the product image feature analysis module 200 determines the strength and direction of the edge by calculating the gradient of the product image in the horizontal and vertical directions, thereby extracting the image contour features in the product image;

[0056] The feature extraction method uses two convolution kernels, one for detecting horizontal edges and the other for detecting vertical edges. These two convolution kernels are convolved with the product image to obtain the approximate gradient values ​​of the image in the horizontal and vertical directions. The gradient amplitude and direction of each pixel are calculated based on the gradient value. The specific working principle is as follows:

[0057] Horizontal convolution kernel for feature extraction ;

[0058] Vertical convolution kernel for feature extraction ;

[0059] For each pixel in the product image , its gradient approximation in the horizontal direction and the vertical gradient approximation It is obtained by convolving the convolution kernel with the pixel and its surrounding pixels;

[0060] Taking the horizontal direction as an example, the calculation formula is: ;

[0061] in, It is the product image in pixels The grayscale value at , the calculation in the vertical direction is the same;

[0062] After obtaining the approximate values ​​of the horizontal and vertical gradients, the pixel points are calculated using the following formula The gradient magnitude and direction :

[0063] Gradient Amplitude: ;

[0064] direction: .

[0065] The product image feature analysis module 200 calculates the Euclidean distances between different image profile features in the marketing image profile feature set and the product image profile feature set, and uses a scatter plot analysis method to set a product similarity threshold;

[0066] If the Euclidean distance between two image contour features in the marketing image contour feature set is greater than the product similarity threshold, the two image contour features in the marketing image contour feature set are determined to be similar products, and the image contour feature corresponding to any one of the similar products is randomly deleted;

[0067] Otherwise, it will be determined that the two image contour features in the marketing image contour features are not similar products, and both image contour features will be retained;

[0068] Similarly, the product image contour feature set uses the same method to delete the image contour features corresponding to similar products, thereby constructing a marketing product set and a browsing product set respectively.

[0069] The calculation formula for calculating the Euclidean distance between two image contour features in the commodity image feature analysis module 200 is as follows:

[0070] Perceive the corresponding image contour features of two product images and , and The Euclidean distance between The calculation formula is: .

[0071] The working principle of setting the commodity similarity threshold by the scatter plot analysis method in the commodity image feature analysis module 200 is as follows:

[0072] The scatter plot analysis method is to display each data point in the data set in the form of coordinates on a two-dimensional plane. In the product similarity analysis, the Euclidean distance between the contour features of different product images is used as the coordinate value. By observing the distribution of scatter points, the aggregation trend and dispersion degree of the data can be intuitively seen;

[0073] The Euclidean distance of the image contour features of similar products is relatively small, and they will be concentrated in a certain area of ​​the scatter plot; while the Euclidean distance of dissimilar products is larger, and they will be distributed in other areas of the scatter plot. To set the product similarity threshold, you need to find a suitable position in the scatter plot and divide the scattered points into two areas, "similar" and "dissimilar". The Euclidean distance value corresponding to this position is the product similarity threshold;

[0074] The specific steps to determine the product similarity threshold are as follows:

[0075] For each Euclidean distance data point , calculate its local density , the formula is ,in Yes and The Euclidean distance between is a cutoff distance used to control the range of density calculation. is the total number of data points;

[0076] Calculate the distance from each point to the closest point among the points with higher density than it ,Right now ;

[0077] Will and Plot on a scatter plot and look for high local density and greater distance The point is the product similarity threshold.

[0078] If all the corresponding products in the user's historical data are included in the products in the marketing product set, the user is marked and the marked user is defined as an accurate user;

[0079] When all the products corresponding to the user's historical data are included in the marketing product set, it means that the user's past consumption behavior is highly consistent with the products provided by the brand, which means that the user has shown a clear purchase tendency and interest in the brand's products. Compared with other users, they are more likely to purchase products in the marketing product set again. They are the brand's most potential target customer group, so they are marked as accurate users;

[0080] The historical data of accurate users provides valuable reference for brand product development and optimization. By analyzing the types of goods purchased by accurate users, usage frequency and other information, brands can gain a deep understanding of user needs and pain points, and then optimize existing product functions, develop new products that better meet market demand, and enhance the brand's market competitiveness.

[0081] The product association analysis module 300 is used to sense the marketing product set and the browsed product set corresponding to the accurate user, and analyze the intersection of the products in the browsed product set of each accurate user and the marketing product set. The intersection of the products is again determined to be similar products through the similarity threshold in the product image feature analysis module 200. When the products in the user's browsed product set and the marketing product set are similar products, the products are determined to be the same product. According to the number of times different product combinations appear in the marketing product set, the association between different products in the marketing brand is calculated;

[0082] By calculating the correlation between products, we can find out which products are often browsed by users at the same time, so that similar products with high correlation can be sold together or displayed together. For example, if we find that users often browse coffee and coffee cups at the same time, we can launch a set of coffee and coffee cups to increase the average order value and sales.

[0083] At the same time, in order to plan precise marketing activities: understanding the relevance of products can help to carry out more precise marketing activities for different user groups. For users who have purchased cameras, relevant photography accessories such as lenses and memory cards can be recommended to improve the accuracy and effectiveness of marketing activities, and increase user response rate and purchase conversion rate.

[0084] The perceived marketing product set is ,have Precise user, The browsed product collection of precise users is , , then calculate any two products in the marketing product set and The steps of the correlation index are as follows:

[0085] Step 1: Calculate the products and Number of times they appear in the intersection at the same time , for every precise user ,like and ,but Add 1 and traverse all After the precise users, value;

[0086] Step 2: Calculate the included products The number of intersection sets , that is, for all precise users ,like ,but Add 1;

[0087] Products and The correlation index is: , the correlation index calculation formula reflects the In the case of The higher the probability of occurrence, the stronger the correlation between the two.

[0088] When considering multiple product combinations, such as , , First, calculate the number of times these three products appear in the intersection at the same time. , and contains goods , The number of intersection sets Etc., and then similarly calculate the correlation index Etc., to measure the correlation between multiple commodities.

[0089] After obtaining the browsed product set and the marketing product set of each accurate user, the product association analysis module 300 calculates the difference between the two. The products in the difference are defined as missing products. The missing products reflect that the accurate user has browsed the products but the marketing product set has not yet covered them.

[0090] The product relevance analysis module 300 analyzes the difference between the product set browsed by each accurate user and the marketing product set, defines the product corresponding to the difference as a missing product, and calculates the relevance between the missing product and different products in the marketing product set;

[0091] The perceived marketing product set is , a precise user, The browsed product collection of precise users is , then the difference set corresponding to the accurate user , the elements in the difference set are the missing products;

[0092] For each missing item and products in the marketing product collection , statistics on the browsing history of all accurate users, missing products and marketing products The number of co-occurrences is recorded as The specific calculation method is to traverse the browsed product collections of all accurate users. If a collection contains and ,but Add 1;

[0093] Calculate correlation: The total number of known accurate users is , missing items and marketing products Relevance The calculation formula is: ;

[0094] By quantifying the correlation between missing products and marketing products, companies can make targeted decisions on whether to introduce new missing products and how to adjust the promotion strategies of existing marketing products to better meet the needs of targeted users and improve marketing effectiveness.

[0095] 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 big data, characterized in that: It comprises a commodity data collection module (100), a commodity image feature analysis module (200) and a commodity correlation analysis module (300); The commodity data collection module (100) sends identity authentication information and browsing data acquisition information to the e-commerce platform. After receiving the identity authentication information, the e-commerce platform verifies the identity authentication information. After the verification is passed, a connection is established with the e-commerce platform, and corresponding user browsing record data is queried according to the browsing data acquisition information. The product image feature analysis module (200) uses the order generation time as the boundary to define historical data and current data, and is used to perceive product information corresponding to the marketing brand, uses a feature extraction method to extract image contour features of the marketing brand and product images in the current data, and defines them as a marketing image contour feature set and a product image contour feature set, respectively, calculates the Euclidean distance between different image contour features, and uses a scatter plot analysis method to set a product similarity threshold; If the Euclidean distance between two image contour features is greater than the product similarity threshold, the two image contour features are determined to be similar products, and the image contour feature corresponding to any similar product is randomly deleted. Otherwise, the two image contour features in the marketing image contour features are determined not to be similar products, and both image contour features are retained. If all the corresponding commodities in the user's historical data are included in the commodities in the marketing commodity set, the commodity correlation analysis module (300) defines the corresponding user as an accurate user, senses the marketing commodity set and the browsed commodity set corresponding to the accurate user, and calculates the intersection and difference of each accurate user's browsed commodity set and the commodities in the marketing commodity set; According to the number of times different product combinations appear in the marketing product set, the correlation between different products in the marketing brand is calculated, the products corresponding to the difference set are defined as missing products, and the correlation between the missing products and different products in the marketing product set is calculated.

2. The brand marketing system based on big data according to claim 1, characterized in that: The working principle of the commodity data collection module (100) for verifying identity authentication information is as follows: the e-commerce platform compares the identity authentication information sent by the commodity data collection module (100) with the legal key list pre-stored in the e-commerce platform database character by character. If they are completely consistent, the verification is passed and a connection is established; if there is a difference, an error message is returned and the connection is refused.

3. The brand marketing system based on big data according to claim 2, characterized in that: The feature extraction method in the commodity image feature analysis module (200) determines the strength and direction of the edge by calculating the gradient of the commodity image in the horizontal and vertical directions, thereby extracting the image contour features in the commodity image; The feature extraction method uses two convolution kernels, one for detecting horizontal edges and the other for detecting vertical edges. These two convolution kernels are convolved with the image to obtain the approximate gradient values ​​of the product image in the horizontal and vertical directions. The gradient amplitude and direction of each pixel are calculated based on the gradient value. The specific calculation formula is as follows: Horizontal convolution kernel for feature extraction ; Vertical convolution kernel for feature extraction ; For each pixel in the product image , its gradient approximation in the horizontal direction and the vertical gradient approximation It is obtained by convolving the convolution kernel with the pixel and its surrounding pixels; Taking the horizontal direction as an example, the calculation formula is: ; in, The product image in pixels The grayscale value at , the calculation in the vertical direction is the same; After obtaining the approximate values ​​of the horizontal and vertical gradients, the pixel points are calculated using the following formula The gradient magnitude and direction : Gradient Magnitude: ; direction: .

4. The brand marketing system based on big data according to claim 1, characterized in that: The product image feature analysis module (200) calculates the Euclidean distance between two image contour features using the following calculation formula: Perceive the corresponding image contour features of two product images and , and The Euclidean distance between The calculation formula is: .

5. The brand marketing system based on big data according to claim 4 is characterized by: The working principle of setting the commodity similarity threshold by the scatter plot analysis method in the commodity image feature analysis module (200) is as follows: The scatter plot analysis method is to display each data point in the data set in the form of coordinates on a two-dimensional plane. In the product similarity analysis, the Euclidean distance between the contour features of different product images is used as the coordinate value. By observing the distribution of scatter points, the aggregation trend and degree of dispersion of the data can be intuitively seen.

6. The brand marketing system based on big data according to claim 5, characterized in that: The specific steps to determine the product similarity threshold are as follows: For each Euclidean distance data point , calculate its local density , the formula is: ； in Yes and The Euclidean distance between is a cutoff distance used to control the range of density calculation. is the total number of data points; Calculate the distance from each point to the closest point among the points with higher density than it ,Right now ; Will and Plot on a scatter plot and look for high local density and greater distance The point is the product similarity threshold.

7. The brand marketing system based on big data according to claim 6, characterized in that: The commodity association analysis module (300) senses the marketing commodity set as ,have Precise user, The browsed product collection of precise users is , , then calculate the two commodities and The steps of the correlation index are as follows: Step 1: Calculate the products and Number of times they appear in the intersection at the same time , for every precise user ,like and ,but Add 1 and traverse all After the precise users, value; Step 2: Calculate the included products The number of intersection sets , that is, for all precise users ,like ,but Add 1; Products and The correlation index is: , the correlation index calculation formula reflects the In the case of The higher the probability of occurrence, the stronger the correlation between the two.

8. The brand marketing system based on big data according to claim 7, characterized in that: The commodity association analysis module (300) senses the marketing commodity set as , a precise user, The browsed product collection of precise users is , then the difference set corresponding to the accurate user , the elements in the difference set are the missing products; For each missing item and products in the marketing product collection , statistics on the browsing history of all accurate users, missing products and marketing products The number of co-occurrences is recorded as The specific calculation method is: traverse all the browsed product sets of accurate users, if a set contains and ,but Add 1; Calculate correlation: The total number of known accurate users is , missing items and marketing products Relevance The calculation formula is: .

Images