Brand user portrait optimization system for cross-border e-commerce personalized recommendation

Abstract

The invention discloses a brand user portrait optimization system for cross-border e-commerce personalized recommendation, and relates to the technical field of e-commerce. The precision problem is solved. The system specifically comprises a data collection module which is used for collecting user information, user purchase information and commodity information; the commodity attribute classification module is used for recording commodity information and classifying commodities; the figure portrait module is used for establishing a multi-dimensional figure portrait according to the classification conditions of the user information, the user purchase information and the commodity information; and a matching module. According to the method, the portrait of the user is established, the established dimensions correspond to the classification types of the commodities, the type to which the commodity belongs and the matching degree of the type to which the commodity belongs are given to each commodity, and finally the favorite degree of the user in each dimension is determined according to the information of the commodities purchased by the user in the past, so that a portrait form similar to a multi-dimensional ability radar map is formed. Therefore, the favorite preference of the user can be determined, and the favorite degree of each favorite type of the user can also be determined.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and particularly to a brand user portrait optimization system for personalized recommendation in cross-border e-commerce. Background Art

[0002] As a vital force in the development of China's foreign trade and an important trend in the development of international trade, new business forms and models with cross-border e-commerce as the highlight have entered a period of rapid growth. In cross-border e-commerce, how to establish and optimize user portraits is a reliable basis for targeted personalized product recommendations based on user portraits in the later stage to increase the conversion rate.

[0003] After retrieval, a patent with the Chinese patent publication number CN114511378A discloses a user portrait system for e-commerce repurchase behavior based on big data, including: a data collection module connected to a central control module for obtaining purchase record data and other interaction data of users within a preset time period; a feature extraction module connected to the central control module for extracting features of purchased products, behavior features, and other features of the collected users; a clustering analysis module connected to the central control module for performing clustering analysis based on the feature extraction results; a portrait optimization module connected to the central control module for optimizing the constructed initial user portrait based on the feature extraction results and the clustering analysis results to obtain an optimized user portrait; a display module connected to the central control module for displaying the collected, processed, analyzed data, and the optimized user portrait.

[0004] The above patent has the following deficiencies: It simply classifies features based on users' past purchase and repurchase records to establish user portraits, and each category in the clustering feature classification has an equal weight form. After classification, it is impossible to determine the degree of users' preference for each product (i.e., each favorite type), resulting in low accuracy of the portrait. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art, and to propose a brand user portrait optimization system for personalized recommendation in cross-border e-commerce.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A brand user portrait optimization system for personalized recommendation in cross-border e-commerce includes:

[0008] A data collection module for collecting user information, user purchase information, and product information;

[0009] A product attribute classification module for collecting product information and classifying products;

[0010] A person portrait module that creates a multi-dimensional portrait of a person based on user information, user purchase information, and the classification of product information;

[0011] A matching module that matches based on the person portrait and product classification, and recommends products to users in descending order of matching degree.

[0012] Preferably, the classification logic of the product attribute classification module includes the following steps:

[0013] A1: Prepare all classification types of product attributes in advance;

[0014] A2: When entering products, personnel classify products according to the attributes of the products in combination with the classification types, and each product can be classified into multiple classification types;

[0015] A3: Assign a matching degree to each classification type to which the product belongs according to the matching situation between the product and the corresponding classification type;

[0016] A4: Combine the basic product information, the type categories of product classification, and the matching degree of each type category to form multi-dimensional radar matching data for product categories.

[0017] Furthermore, in the multi-dimensional portrait of a person in the person portrait module, the dimensions of the person portrait correspond one by one to all the classification types of product attributes in step A1, and its specific logic includes the following steps:

[0018] B1: Obtain the degree of preference R' of the person under the past purchase records according to the past purchase records;

[0019] B2: Classify the preferences of the person according to the basic information of the person, and obtain the degree of preference R'' of the person for each classification;

[0020] B3: Then calculate the final degree of preference R = f(R', R'') of the person according to R' and R'', where f is a binary mathematical function, which is any one of the sum function R = k'R' + k''R'' and the product function R = k'R' + k''R'', and k' and k'' are the importance degrees of R' and R'' respectively, preset by the administrator.

[0021] Based on the above solution: Step B1 includes the following steps:

[0022] B11: Obtain the past purchase records of the user, which include the purchased product M i , the purchase times CM i and the purchase frequency PM of the same product i , M i represents the i-th product in the past purchase records, CM i is the purchase times of the i-th product, PMi is the purchase frequency of the i-th commodity;

[0023] B12: Obtain the category M to which the previously purchased commodity belongs through the commodity attribute classification module i o and the matching degree QM of the category i o , M i o is the o category in the i-th commodity, and QM i o is the o classification matching degree in the i-th commodity;

[0024] B13: Then calculate the preference degree R of the person for the o category based on the previous purchase records according to the formula o ′, where k i is the participation weight of the i-th commodity, which is determined by the purchase times CM i and the purchase frequency PM i of the i-th commodity together;

[0025] B14: Repeat steps B11 - B13 until the preference degrees R′ of all commodity categories of the person based on the purchase records are determined.

[0026] In a better solution of the foregoing solution: in step B13, k i = f(CM i ) + f(PM i ) or k i = f(CM i ) × f(PM i ), where f is a scaling function that scales CM i and PM i to the interval (0, 1).

[0027] As a further solution of the present invention: step B2 includes the following steps:

[0028] B21: Collect the basic information of different persons, and classify the persons into multiple types according to the differences in the basic information;

[0029] B22: For each type of person, establish the preference degree R″ of this type for each commodity category.

[0030] Meanwhile, a person portrait update logic is built into the person portrait module, which includes the following steps:

[0031] C1: Set a regular update time period σT, and the system obtains the commodity purchase times C and commodity purchase frequency F corresponding to each multi-dimensional portrait in real time;

[0032] C2: Set the purchase times trigger threshold σC and the purchase frequency trigger threshold σF;

[0033] C3: Compare according to the last updated time node, the user's purchase times C with the trigger threshold σC, and the user's purchase frequency F with the trigger threshold σF to determine whether to trigger an update.

[0034] As a preferred embodiment of the present invention: In the step C3, it includes the following situations:

[0035] C31: When the time since the last update has not reached the regular update time period k″σT and during this period, both the user's purchase times C and the purchase frequency F have not reached k″σC and k″σF, no update is triggered;

[0036] C32: When the time since the last update has not reached the regular update time period k″σT and during this period, the user's purchase times C > k″σC or the purchase frequency of the product F > k″σF, an update is triggered;

[0037] C33: When the time since the last update reaches the regular update time period k″σT, an update is triggered.

[0038] Meanwhile, in the steps C1 - C3, k″ is the update frequency follow-up coefficient, which is determined by the user's purchasing power.

[0039] As a more optimal solution of the present invention: In the steps C1 - C3, the determination logic of k″ is as follows:

[0040] D1: Obtain the previous n + 1 update nodes and calculate the average purchase amount H of the user between every two adjacent update nodes i , H i represents the ratio of the total amount between the i-th update node and the (i + 1)-th update node calculated forward from the current time node to the time difference between the two nodes;

[0041] D2: Then according to the formula where f′ is the time - related function of H i and f′(i) is a positive number, and f′ is a monotonically decreasing function.

[0042] The beneficial effects of the present invention are:

[0043] 1. In the present invention, for the establishment of the user portrait, the established dimensions correspond to the classification types of the products, and for each product, its belonging type and the matching degree of the belonging type are given. Finally, according to the information of the user's previous purchased products, the degree of preference of the user for each dimension is determined, forming a portrait form similar to a multi - dimensional ability radar chart, so that both the user's preference can be determined and the degree of preference of the user for each preference type can be determined.

[0044] 2. For the establishment of the user's preference degree for each type in the present invention, a comprehensive algorithm of the preference degree under the purchase record and the preference degree under the user's basic information classification is adopted, so that the user portrait can be associated with the user's own situation and the user's past operation situation, further increasing the portrait accuracy.

[0045] 3. In the present invention, for the comprehensive algorithm of the preference degree under the purchase record and the preference degree under the user's basic information classification, importance is introduced, so that the administrator can adjust the importance according to actual needs, and update the focus of the user portrait between the user's own situation and the past operation situation, further increasing the pertinence.

[0046] 4. In the present invention, the portrait of a person can be updated according to the time period, the number of purchases, and the purchase frequency, so as to further increase the accuracy of the portrait.

[0047] 5. In the present invention, for the threshold control of the time period, the number of purchases, and the purchase frequency, a follow-up coefficient of the update frequency is introduced, which can determine the update frequency of the user portrait according to the user's purchasing power. Therefore, high-frequency updates can be realized for important users to increase the pertinence of subsequent recommendations, and low-frequency updates can be performed for ordinary customers to reduce the resource occupancy of the system.

[0048] 6. In the present invention, for the follow-up coefficient of the update frequency, a time-correlation function is used to control the weight size, so that the determination of the final follow-up coefficient of the update frequency can be more matched with the purchasing power in the near future while taking into account the past purchase situation, increasing the matching accuracy with the user's behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is the architecture diagram of the brand user portrait optimization system for cross-border e-commerce personalized recommendation proposed by the present invention;

[0050] Figure 2 It is the logic diagram of the portrait establishment of the brand user portrait optimization system for cross-border e-commerce personalized recommendation proposed by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0051] The technical solutions of the present invention will be further described in detail below in conjunction with the specific embodiments.

[0052] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions from beginning to end. The embodiments described below by referring to the drawings are exemplary and are only used to explain the present invention and should not be construed as a limitation of the present invention.

[0053] Embodiment 1:

[0054] Brand user portrait optimization system for cross-border e-commerce personalized recommendation, including:

[0055] A data collection module, which is used to collect user information, user purchase information and product information;

[0056] A commodity attribute classification module, which collects commodity information and classifies commodities;

[0057] The character portrait module creates a multi-dimensional character portrait based on the classification of user information, user purchase information, and product information;

[0058] The matching module matches character portraits and product categories, and recommends products to users in descending order of matching degree.

[0059] The classification logic of the commodity attribute classification module includes the following steps:

[0060] A1: Prepare all classification types of product attributes in advance;

[0061] A2: When entering products, personnel classify the products according to their attributes and classification types. Each product can be classified into multiple classification types.

[0062] A3: According to the matching between the product and the corresponding category type, a matching degree is assigned to each category type to which the product belongs;

[0063] A4: Combine the basic information of the product, the types of the product classification, and the matching degree of each type to form multi-dimensional radar matching data of the product types.

[0064] For example, four classification types A, B, C, and D are prepared in advance. Now a product M is entered, and the personnel determine that its type is A, B, and D, and its matching degree with A, B, and D is classified as 10%, 50%, and 70%. Then the final multi-dimensional radar data of the product is M = (10% A, 50% B, 70% C, JM), where JM is the basic information of the product.

[0065] In the multi-dimensional character portrait of the character portrait module, the dimensions of the character portrait correspond one-to-one to all classification types of the commodity attributes in step A1, and its specific logic includes the following steps:

[0066] B1: Obtain the character's preference level R' based on the previous purchase records;

[0067] B2: Classify the character's preferences according to the character's basic information, and obtain the character's preference level R" for each category;

[0068] B3: Then, calculate the final preference degree R of the person according to R' and R'' as R = f(R', R''), where f is a binary mathematical function, which can be either a sum function R = k'R' + k''R'' or a product function R = k'R' + k''R'', and k' and k'' are the importance degrees of R' and R'' respectively, preset by the administrator.

[0069] The B1 step includes the following steps:

[0070] B11: Obtain the user's past purchase records, which include the purchased product M i , the number of purchases CM i and the purchase frequency PM of the same product i , where M i represents the i-th product in the past purchase records, CM i is the number of purchases of the i-th product, and PM i is the purchase frequency of the i-th product;

[0071] B12: Obtain the category M to which the past purchased product belongs and the matching degree QM of the category through the product attribute classification module i o , where M i o is the o category in the i-th product, and QM i o is the o category matching degree of the i-th product; i o B13: Then, calculate the preference degree R' of the person for the o category under the past purchase records according to the formula

[0072] , where k is the participation weight of the i-th product, which is jointly determined by the number of purchases CM o of the i-th product and the purchase frequency PM i of the i-th product; i and the purchase frequency PM i of the i-th product;

[0073] B14: Repeat steps B11 - B13 until the preference degrees R' of all product categories of the person under the purchase records are determined.

[0074] In the B13 step, k i = f(CM i ) + f(PM i ) or k i = f(CM i ) × f(PM i ), where f is a scaling function that scales CM i and PM i to the interval (0, 1).

[0075] The step B2 comprises the following steps:

[0076] B21: Collect basic information of different characters, and classify the characters into multiple types according to the different basic information;

[0077] B22: For each type of person, establish the preference level R″ of that type for each product category.

[0078] In this embodiment, when establishing a user portrait, dimensions are established corresponding to the classification types of goods, and each product is assigned its type and the matching degree of its type. Finally, the user's preference for each dimension is determined based on the user's previous purchase information, forming a portrait in the form of a multi-dimensional capability radar chart, thereby determining the user's preference and the user's preference for each favorite type.

[0079] At the same time, in order to establish the user's preference for each type, a comprehensive algorithm is used that combines the preference under the purchase record and the preference under the user's basic information classification, so that the user portrait can be associated with the user's own situation and the user's previous operations, further increasing the portrait's accuracy.

[0080] In addition, importance is introduced into the comprehensive algorithm of the preference level under purchase records and the preference level under user basic information classification, so that administrators can adjust the importance according to actual needs to update the focus of user portraits between their own situation and past operation conditions, further increasing the targeting.

[0081] Embodiment 2:

[0082] A brand user portrait optimization system for personalized recommendation of cross-border e-commerce, this embodiment makes the following improvements on the basis of embodiment 1: the character portrait module has a built-in character portrait update logic, which includes the following steps:

[0083] C1: Set a regular update time period σT, and the system obtains the number of user product purchases C and product purchase frequency F corresponding to each multi-dimensional portrait in real time;

[0084] C2: Set the purchase count trigger threshold σC and the purchase frequency trigger threshold σF;

[0085] C3: Determine whether to trigger an update based on the time of the last update, the number of user purchases C and the trigger threshold σC, and the user purchase frequency F and the trigger threshold σF.

[0086] The C3 step includes the following:

[0087] C31: When the time since the last update has not reached the regular update time period k″σT and during this period, neither the number of user purchases C nor the purchase frequency F has reached k″σC and k″σF, no update is triggered;

[0088] C32: When the time since the last update has not reached the regular update time period k″σT and during this period, the number of user purchases C > k″σC or the purchase frequency of goods F > k″σF, an update is triggered;

[0089] C33: When the time since the last update reaches the regular update time period k″σT, an update is triggered.

[0090] In the steps C1 - C3 described above, k″ is the update frequency follow-up coefficient, which is determined by the purchasing power of the user.

[0091] In the steps C1 - C3 described above, the determination logic of k″ is as follows:

[0092] D1: Obtain the previous n + 1 update nodes, and calculate the average purchase amount H of the user between every two adjacent update nodes i , H i represents the ratio of the total amount between the i-th update node and the (i + 1)-th update node calculated backward from the current time node to the time difference between the two nodes;

[0093] D2: Then, according to the formula where f′ is the time-correlation function of H i , f′(i) is a positive number, and f′ is a monotonically decreasing function.

[0094] In this embodiment, the portrait of the person can be updated according to the time period, the number of purchases, and the purchase frequency, thereby further increasing the accuracy of the portrait.

[0095] In addition, for the threshold control of the time period, the number of purchases, and the purchase frequency, an update frequency follow-up coefficient is introduced. It can determine the update frequency of the user portrait according to the purchasing power of the user. Thus, high-frequency updates can be achieved for important users to increase the pertinence of subsequent recommendations, and low-frequency updates can be performed for ordinary customers to reduce the resource occupation of the system.

[0096] At the same time, for the update frequency follow-up coefficient, a time-correlation function is used to control the weight size, so that the determination of the final update frequency follow-up coefficient can be more matched with the purchasing power in the near future while taking into account the previous purchase situation, increasing the matching accuracy with user behavior.

[0097] As described above, it is only the preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, shall be covered by the protection scope of the present invention.

Claims

1. A brand user portrait optimization system for personalized recommendation in cross-border e-commerce, characterized in that, include: A data collection module, which is used to collect user information, user purchase information and product information; A commodity attribute classification module, which collects commodity information and classifies commodities; The character portrait module creates a multi-dimensional character portrait based on the classification of user information, user purchase information, and product information; The matching module matches character portraits and product categories, and recommends products to users in descending order of matching degree.

2. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 1, wherein The classification logic of the commodity attribute classification module includes the following steps: A1: Prepare all classification types of product attributes in advance; A2: When entering products, personnel classify the products according to their attributes and classification types. Each product can be classified into multiple classification types. A3: According to the matching between the product and the corresponding classification type, a matching degree is assigned to each classification type to which the product belongs; A4: Combine the basic information of the product, the types of the product classification, and the matching degree of each type to form multi-dimensional radar matching data of the product types.

3. The brand user profile optimization system for personalized recommendation in cross-border e-commerce according to claim 2, wherein In the multi-dimensional character portrait of the character portrait module, the dimensions of the character portrait correspond one-to-one to all classification types of the commodity attributes in step A1, and its specific logic includes the following steps: B1: Obtain the character's preference level R' based on the previous purchase records; B2: Classify the character's preferences according to the character's basic information, and obtain the character's preference level R" for each category; B3: Then calculate the final degree of character preference R = f(R′, R″) based on R′ and R″, where f is a binary mathematical function, which is any one of the sum function R = k′R′+k″R″ and the product function R = k′R′+k″R″, where k′ and k″ are the importance of R′ and R″ respectively, which are preset by the administrator.

4. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 3, characterized in that The step B1 comprises the following steps: B11: Obtain the user's past purchase records, which include the purchased product M i , the number of purchases CM i and the purchase frequency PM of the same product i , M i represents the i-th product in the past purchase records, CM i is the number of purchases of the i-th product, PM i is the purchase frequency of the i-th product; B12: Obtain the category M to which the previously purchased product belongs through the product attribute classification module i o and the matching degree QM of the category i o , M i o is the o category in the i-th product, and QM i o is the o category matching degree in the i-th product; B13: Then, calculate the preference degree R of the person for category o under the past purchase records according to the formula o ′, where k i is the participation weight of the i-th commodity, which is jointly determined by the purchase times CM i and the purchase frequency PM i of the i-th commodity; B14: Repeat steps B11-B13 until the preference level R′ of all commodity categories of the person under the purchase record is determined.

5. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 4, characterized in that In the step B13, k i = f(CM i ) + f(PM i ) or k i = f(CM i ) × f(PM i ), where f is a scaling function that scales CM i and PM i to the interval (0, 1).

6. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 3, wherein The step B2 comprises the following steps: B21: Collect basic information of different characters, and classify the characters into multiple types according to the different basic information; B22: For each type of person, establish the preference level R″ of that type for each product category.

7. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 1, characterized in that, The character portrait module has a built-in character portrait update logic, which includes the following steps: C1: Set a regular update time period σT, and the system obtains the number of user product purchases C and product purchase frequency F corresponding to each multi-dimensional portrait in real time; C2: Set the purchase count trigger threshold σC and the purchase frequency trigger threshold σF; C3: Determine whether to trigger an update based on the time of the last update, the number of user purchases C and the trigger threshold σC, and the user purchase frequency F and the trigger threshold σF.

8. The brand user portrait optimization system for personalized recommendation in cross-border e-commerce according to claim 7, wherein The C3 step, This includes the following situations: C31: When the period of time since the last update has not reached the regular update period k″σT and the number of purchases C and the purchase frequency F of the user during this period have not reached k″σC and k″σF, the update is not triggered; C32: Update is triggered when the time since the last update has not reached the regular update time period k″σT and during this period the number of user purchases C > k″σC or the commodity purchase frequency F > k″σF; C33: Update is triggered when the time since the last update reaches the regular update time period k″σT.

9. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 8, wherein In the steps C1 - C3, k″ is a follow-up coefficient for the update frequency, which is determined by the purchasing power of the user.

10. The brand user portrait optimization system for cross-border e-commerce personalized recommendation according to claim 9, characterized in that, In the steps C1 - C3, the determination logic of k″ is as follows: D1: Obtain the previous n + 1 updated nodes, and calculate the average purchase amount H of users between every two adjacent updated nodes i , H i represents the ratio of the total amount between the i-th updated node and the (i + 1)-th updated node calculated by pushing forward from the current time node to the time difference between the two nodes; D2: Then, according to the formula where f' is the time-correlation function of H i and f'(i) is a positive number, and f' is a monotonically decreasing function.

Images