User portrait intelligent optimization method and analysis management system based on cloud computing and AI

Abstract

The invention belongs to the field of user portrait analysis, and particularly relates to a user portrait intelligent optimization method and an analysis management system based on cloud computing and AI, which can more quickly obtain the preference of a user by constructing a user portrait through a cloud platform, perform data cleaning on redundant information in the user portrait, reduce the computing pressure of the cloud platform, and improve the user experience. Meanwhile, user data in the user portrait is encrypted to ensure the safety of user information; a pushing strategy is updated for the user portrait, and meanwhile the user purchase degree and the user triggering sensitive behavior frequency in the user portrait are stored; compared with the prior art, the method has the advantages that the pushing result is more accurate, the types of pushed commodities are various, the judgment on sensitive behaviors of the user is more accurate, and disputes between the user and merchants are reduced.

Description

Technical Field

[0001] The present invention belongs to the field of user portrait analysis, and specifically relates to a user portrait intelligent optimization method and analysis management system based on cloud computing and AI. Background Art

[0002] User portrait analysis refers to the construction of a virtual typical user model by collecting, integrating and analyzing various types of user data (such as behavior, preferences, needs, background, etc.). User portraits can help the system deeply understand the characteristics of the target user group and guide systematic methods such as product design and marketing strategy formulation. At the same time, user behavior can also be monitored through user portraits.

[0003] In the personalized recommendation systems of current cross-border e-commerce platforms, user profile construction primarily relies on statistical analysis of historical behavioral data and product tag matching algorithms. Product recommendation systems are typically established by collecting explicit behavioral data such as user clicks, favorites, and purchase records, combined with static attributes such as product categories, price ranges, and basic functional parameters. Traditional recommendation systems are unable to quickly and effectively filter content that users are not interested in, nor can they quickly update recommendation information based on changes in users' browsing or purchasing habits. Furthermore, feedback on changes in user behavior is not timely enough, and user behavior cannot be accurately constrained based on user behavior feedback in user profiles. Therefore, an intelligent optimization method for platform recommendations based on user profiles and a user analysis and management system are urgently needed to address these issues. Summary of the Invention

[0004] The purpose of this invention is to propose a user portrait intelligent optimization method based on cloud computing and AI to solve one or more technical problems existing in the prior art and at least provide a beneficial option or create conditions.

[0005] S100: Obtain basic user information on the platform and construct user profiles based on the basic user information; S200: organizing user profiles based on the cloud platform, and pushing products to users based on user behavior information in the user profiles and historical user behavior information; S300: Cleaning redundant information in the user portrait and encrypting user data in the user portrait; S400: Calculate and update the push strategy based on the user profile, and save the user's purchase degree and the number of times the user triggers sensitive behaviors in the user profile; S500: Output the push strategy and the number of times the user triggers sensitive behaviors to score and classify the user.

[0006] Furthermore, in step S100, user basic information is obtained through the cross-border e-commerce platform, and the user basic information includes: user preference information, user behavior information, the user preference information is defined by the number of user browsing times and the number of user interactions, and the user behavior information includes browsing behavior information and interaction behavior information. A user portrait is constructed by combining the user preference information and the user behavior information with the user basic information. The user basic information is the user-defined setting parameters and the user number, and the user push products and the preliminary user behavior score are determined by the user portrait.

[0007] Furthermore, in step S200, cloud user update data is obtained through the cloud platform, and the user portrait is updated through the user update data. In the user portrait, the user browses the product type as com_1, and the product type similar to the product com_1 is collaboratively filtered as com_2. It is recorded whether the user chooses to browse the products of the product com_2, and the number of times the user browses the products com_1 and com_2 is recorded. The new product type recommended to the user is defined as com_3 based on the recorded number of times the product com_2 is browsed, and the product is pushed according to the new browsing product type of the user in the same way as the method. The push method is used to sort the pushed products according to the number of views and click-through rate through a logistic regression algorithm.

[0008] Preferably, the logistic regression algorithm is combined with the convolutional neural algorithm to analyze the page views and click-through rates to push products, and similar products or products that users may like are pushed through AI analysis to increase the types of push and ensure the rationality of the push.

[0009] Furthermore, in step S300, the product push strategy is obtained by combining the user preference information and user behavior information in the user portrait with the user feedback information after the product is pushed through the logistic regression algorithm, recording the page views and click-through rates as well as the browsed product categories, filtering the redundant data in the user portrait, and recording the repeated product categories in the recorded page views and click-through rates as the browsing weight of the changed product category by filtering the product categories, and deleting the redundantly recorded page views and click-through rates in the cloud platform to reduce the space occupied by the cloud platform for calculating user portrait data, and recording the traceability channel of the redundantly recorded data in the background system.

[0010] Furthermore, it is characterized in that the specific method for encrypting user data in the user portrait comprises the following steps: dividing the user portrait data into a plurality of nodes by date, and forming a node graph from each node; In the node graph, each node records two attribute values: the number of user visits and the amount of system data accesses. Each time a user visit is recorded, an edge is established in the node graph. The amount of system data accesses during this user visit is recorded on this edge. The commodity type of the user visit is all commodities. The amount of system data accesses is the number of bytes of data read by the node through this edge. The amount of system data accesses is the number of bytes used by the system to update the user profile based on the number of user visits. In the node graph, the set of nodes is recorded as Nset, the number of nodes is n, the sequence number of the node is i, i∈[1,n], the element with sequence number i in Nset is Nset(i), and each edge of each node in the node graph is connected to the adjacent nodes in the node graph; Based on the encryption vector of each node, the method for determining the data access rights between nodes is as follows: The set of encryption vectors obtained by each node through the encryption module is used as the encryption vector set; For any node, let it be Nset(i), let the number of edges of the node be s, let the sequence number of the edges of the node be t, t∈[1,s], let the edge with sequence number t in the edges of Nset(i) be E(i,t), let the data access volume on E(i,t) be B(i,t), let the encrypted vector of the node be Vec(i), Vec(i) be an s-dimensional vector, let the sequence number of the dimension in Vec(i) be t, and let the value of the dimension with sequence number t in Vec(i) be Vec(i,t); Calculate the arithmetic mean of the values ​​of each dimension in Vec(i) as the screening value, and record the screening value as η(i). Compare the values ​​of the dimensions of each sequence number in Vec(i) with the screening value η(i). Filter out the sequence numbers of the dimensions in Vec(i) whose values ​​are less than the screening value η(i) as the screening sequence number set; The screening ratio value calculated according to Vec(i) is recorded as P(i). The calculation formula of P(i) is: ; The function exp is an exponential function with the natural number e as the base, thereby obtaining the screening ratio value P(i); Calculate the average system data access volume on each edge of node Nset(i), and denote the average data access volume on each edge of node Nset(i) as bf(i). The calculation formula of bf(i) is: ; According to the sequence number in the filter sequence number set, select the edges with corresponding sequence numbers from the edges of node Nset(i) to form the edge set to be filtered. Any edge in the edge set to be filtered is denoted as E(i, tb). The sequence number of edge E(i, tb) in the edges of Nset(i) is tb, tb∈[1,s], and the data access volume on E(i, tb) is B(i, tb). Determine whether the constraint B(i, tb)\[bf(i)* P(i)]>1 is met. If so, delete the data reading permission of edge E(i, tb) on node Nset(i). That is, the node originally connected to Nset(i) through edge E(i, tb) loses the permission to read the data of node Nset(i).

[0011] Preferably, the user behavior is obtained from the cloud platform, and the number of user behaviors is monitored and compared with the amount of bytes read. When it is found that the number of user behaviors is not equal to the amount of bytes read, the cloud platform is immediately disconnected from reading the user behavior, and the user data is encrypted and detected to prevent user data leakage.

[0012] Furthermore, in step S400, the method for updating the push strategy by user portrait is as follows: randomly select a user as the main user UsA, the merchant who conducts transactions with the main user UsA as the slave user denoted as UsB, and the set of all slave users as the purchase list of the main user UsA list (UsA), list (UsA) = {M }, j1∈[1, L], where j1 is the serial number of the slave user UsB in the purchase list list(UsA) of UsA. The order of the slave user UsB is based on the purchase time of the master user UsA. If the master user purchases from the same slave user multiple times, the first purchase is used as the ordering basis, and the order is based on the purchase time. represents the j1th element in the purchase list, L represents the number of slave users UsB in list (UsA), k represents the number of different items purchased by the same slave user, M It is represented as the kth item purchased from the user at position j1; S401, set a variable j2 as the first traversal variable, initialize the value of j2 to 1, if j2≤L, use list( ) indicates that the master user UsA is in the slave user Purchase list, list( )=(M ), in this range, j1 and j2 are variables, and j2=j1, E represents the master user UsA in the slave user The total number of purchased items, get the purchase list list (UsA) and list ( ), set a variable as the main intersection quantity for each element in the purchase list of the main user UsA, and use the number of elements in the intersection as the main intersection quantity F(j2) of the j2th element in list(UsA); S402, set a variable j3 as the second traversal variable, initialize the value of i3 to 1, create an empty set as the purchase list listls, set a variable as the total intersection F, initialize the value of the total intersection F to 0, the j3 traverses all slave users UsB, defines the slave user is the main user traversed by j3, ∈ , obtained from the user List of products listls( ) and from the user List of products listls( ) of the intersection TSub( , ), the intersection TSub( , ) as the sub-intersection component TSub(j3), and the sum of the sub-intersection component TSub(j3) and the sub-intersection total F as the new sub-intersection total F. If there are elements in TSub(j3) that do not exist in FEOLs (UsA), then these elements are added to the set PL; S403: The purchase degree λ is 1. The arithmetic mean of the number of elements in the product list of each slave user in the PL is obtained as the purchase mean Efans. The purchase degree λ of all elements of the master user in the PL is calculated. The method for calculating the purchase degree λ is as follows: λ= +1; Where j4 is the subscript of the set PL; merge the elements in PL into list(UsA), L is the total number of all slave users, if the purchase degree λ is not 1, calculate the intersection coefficient ACrs, set a variable j5 as the serial number of the middle element in FEOLs(UsA), where the intersection coefficient ACrsi5 of the i5th element in FEOLs(UsA) is calculated as: ; in represents the mutual attention of the j5th element in FEOLs(UsA), and They represent the main intersection and sub-intersection of the j5th element in list(UsA), m1 is the cumulative variable, U( ) represents the value ratio of the set of all SCrs values The average value of the large values ​​is used, and the set of each intersection coefficient ACrs is used as the intersection space zAC. The intersection space zAC is output and used as the push strategy.

[0013] Preferably, through the intersection space zAC, the purchased items and purchasing preferences of the main user UsA can be effectively quantified, the multiple purchasing behaviors between each merchant user and the main user in the main user's purchasing behavior can be quantified, and the secondary intersection volume can be found among the items of multiple merchants purchased by the main user to further refine the main user's purchasing preferences. At the same time, based on the fact that there are still certain items that the main user is interested in besides the secondary intersection volume and the main intersection volume, these types of items are also added to the intersection space as a reference, which is more helpful for subsequent update push decisions and deepens the mining of the main user's product preferences compared with traditional personal preference analysis.

[0014] Furthermore, a method for obtaining a user behavior coefficient from a purchase list is as follows: obtaining purchase records between a user UsB and a list (UsA) within a period of time, wherein the period of time is a value ranging from 1 to 180 days, and the product information is product information posted by the user; setting a user behavior coefficient Sml for each element in the list (UsA), wherein the initial value of the user behavior coefficient Sml is 0; If a user performs at least one of the following actions: returning a product, not paying after placing an order, abnormal return behavior, or multiple times, a Boolean variable is set as a negative label NFlag for the user's purchase information and the value is FALSE; if a user performs at least one of the following actions: unfollowing, disliking, or reporting, a Boolean variable is set as a negative label NFlag for the user's purchase information and the value is TRUE; Initialize the value of the user behavior coefficient Sml to 0; filter out the co-viewed information RMsg of the main user UsA with the negative label NFLag as the reference purchase information, and calculate the main user behavior coefficient Sml for each user who browses the reference purchase information of the same product: if the negative label NFLag of the reference purchase information for the user corresponding to an element in list(UsA) is equal to the negative label NFLag of the main user UsA, then add 1 to the main user behavior coefficient Sml of the element; otherwise, subtract 1 from the main user behavior coefficient Sml of the element; traverse all reference purchases and calculate the main user behavior coefficient Sml for each element in the corresponding list(UsA), and output the user behavior coefficient score.

[0015] A user portrait intelligent analysis and management system based on cloud computing and AI, the system comprising: a cloud computing platform, a processor and a memory. The cloud computing platform organizes and updates user portraits by acquiring user behavior. The memory stores data output from the cloud computing platform and the processor. When the processor executes the computer program, it can implement any step in the user portrait intelligent optimization method based on cloud computing and AI in the above method.

[0016] The beneficial effects of the present invention are as follows: building a user profile through a cloud platform can more quickly obtain user preferences, clean up redundant information in the user profile, reduce the computing pressure of the cloud platform, and encrypt the user data in the user profile to ensure user information security; By updating the push strategy for the user portrait and saving the user's purchase degree and the number of times the user triggers sensitive behaviors in the user portrait, the push results are more accurate, the types of pushed products are more diverse, and the judgment of user sensitive behaviors is more accurate, reducing the occurrence of disputes between users and merchants. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other features of the present invention will become more apparent through a detailed description of the embodiments shown in conjunction with the accompanying drawings. In the drawings of the present invention, the same reference numerals represent the same or similar elements. Obviously, the drawings described below are only some embodiments of the present invention. It is possible for a person skilled in the art to derive other drawings based on these drawings without inventive effort. In the drawings: Figure 1 Shown is a flow chart of a user portrait intelligent optimization method based on cloud computing and AI. DETAILED DESCRIPTION

[0018] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments and features in the embodiments of this application can be combined with each other unless there is a conflict.

[0019] like Figure 1 As shown, a user portrait intelligent optimization method based on cloud computing and AI includes the following steps: S100: Obtain basic user information on the platform and construct user profiles based on the basic user information; S200: organizing user profiles based on the cloud platform, and pushing products to users based on user behavior information in the user profiles and historical user behavior information; S300: Cleaning redundant information in the user portrait and encrypting user data in the user portrait; S400: Calculate and update the push strategy based on the user profile, and save the user's purchase degree and the number of times the user triggers sensitive behaviors in the user profile; S500: Output the push strategy and the number of times the user triggers sensitive behaviors to score and classify the user.

[0020] Furthermore, in step S100, user basic information is obtained through the cross-border e-commerce platform, and the user basic information includes: user preference information, user behavior information, the user preference information is defined by the number of user browsing times and the number of user interactions, and the user behavior information includes browsing behavior information and interaction behavior information. A user portrait is constructed by combining the user preference information and the user behavior information with the user basic information. The user basic information is the user-defined setting parameters and the user number, and the user push products and the preliminary user behavior score are determined by the user portrait.

[0021] Furthermore, in step S200, cloud user update data is obtained through the cloud platform, and the user portrait is updated through the user update data. In the user portrait, the user browses the product type as com_1, and the product type similar to the product com_1 is collaboratively filtered as com_2. It is recorded whether the user chooses to browse the products of the product com_2, and the number of times the user browses the products com_1 and com_2 is recorded. The new product type recommended to the user is defined as com_3 based on the recorded number of times the product com_2 is browsed, and the product is pushed according to the new browsing product type of the user in the same way as the method. The push method is used to sort the pushed products according to the number of views and click-through rate through a logistic regression algorithm.

[0022] Preferably, the logistic regression algorithm is combined with the convolutional neural algorithm to analyze the page views and click-through rates to push products, and similar products or products that users may like are pushed through AI analysis to increase the types of push and ensure the rationality of the push.

[0023] Furthermore, in step S300, the product push strategy is obtained by combining the user preference information and user behavior information in the user portrait with the user feedback information after the product is pushed through the logistic regression algorithm, recording the page views and click-through rates as well as the browsed product categories, filtering the redundant data in the user portrait, and recording the repeated product categories in the recorded page views and click-through rates as the browsing weight of the changed product category by filtering the product categories, and deleting the redundantly recorded page views and click-through rates in the cloud platform to reduce the space occupied by the cloud platform for calculating user portrait data, and recording the traceability channel of the redundantly recorded data in the background system.

[0024] Furthermore, it is characterized in that the specific method for encrypting user data in the user portrait comprises the following steps: dividing the user portrait data into a plurality of nodes by date, and forming a node graph from each node; In the node graph, each node records two attribute values: the number of user visits and the amount of system data accesses. Each time a user visit is recorded, an edge is established in the node graph. The amount of system data accesses during this user visit is recorded on this edge. The commodity type of the user visit is all commodities. The amount of system data accesses is the number of bytes of data read by the node through this edge. The amount of system data accesses is the number of bytes used by the system to update the user profile based on the number of user visits. In the node graph, the set of nodes is recorded as Nset, the number of nodes is n, the sequence number of the node is i, i∈[1,n], the element with sequence number i in Nset is Nset(i), and each edge of each node in the node graph is connected to the adjacent nodes in the node graph; Based on the encryption vector of each node, the method for determining the data access rights between nodes is as follows: The set of encryption vectors obtained by each node through the encryption module is used as the encryption vector set; For any node, let it be Nset(i), let the number of edges of the node be s, let the sequence number of the edges of the node be t, t∈[1,s], let the edge with sequence number t in the edges of Nset(i) be E(i,t), let the data access volume on E(i,t) be B(i,t), let the encrypted vector of the node be Vec(i), Vec(i) be an s-dimensional vector, let the sequence number of the dimension in Vec(i) be t, and let the value of the dimension with sequence number t in Vec(i) be Vec(i,t); Calculate the arithmetic mean of the values ​​of each dimension in Vec(i) as the screening value, and record the screening value as η(i). Compare the values ​​of the dimensions of each sequence number in Vec(i) with the screening value η(i). Filter out the sequence numbers of the dimensions in Vec(i) whose values ​​are less than the screening value η(i) as the screening sequence number set; The screening ratio value calculated according to Vec(i) is recorded as P(i). The calculation formula of P(i) is: ; The function exp is an exponential function with the natural number e as the base, thereby obtaining the screening ratio value P(i); Calculate the average system data access volume on each edge of node Nset(i), and denote the average data access volume on each edge of node Nset(i) as bf(i). The calculation formula of bf(i) is: ; According to the sequence number in the filter sequence number set, select the edges with corresponding sequence numbers from the edges of node Nset(i) to form the edge set to be filtered. Any edge in the edge set to be filtered is denoted as E(i, tb). The sequence number of edge E(i, tb) in the edges of Nset(i) is tb, tb∈[1,s], and the data access volume on E(i, tb) is B(i, tb). Determine whether the constraint B(i, tb)\[bf(i)* P(i)]>1 is met. If so, delete the data reading permission of edge E(i, tb) on node Nset(i). That is, the node originally connected to Nset(i) through edge E(i, tb) loses the permission to read the data of node Nset(i).

[0025] Preferably, the user behavior is obtained from the cloud platform, and the number of user behaviors is monitored and compared with the amount of bytes read. When it is found that the number of user behaviors is not equal to the amount of bytes read, the cloud platform is immediately disconnected from reading the user behavior, and the user data is encrypted and detected to prevent user data leakage.

[0026] Furthermore, in step S400, the method for updating the push strategy by user portrait is as follows: randomly select a user as the main user UsA, the merchant who conducts transactions with the main user UsA as the slave user denoted as UsB, and the set of all slave users as the purchase list of the main user UsA list (UsA), list (UsA) = {M }, j1∈[1, L], where j1 is the serial number of the slave user UsB in the purchase list list(UsA) of UsA. The order of the slave user UsB is based on the purchase time of the master user UsA. If the master user purchases from the same slave user multiple times, the first purchase is used as the ordering basis, and the order is based on the purchase time. represents the j1th element in the purchase list, L represents the number of slave users UsB in list (UsA), k represents the number of different items purchased by the same slave user, M It is represented as the kth item purchased from the user at position j1; S401, set a variable j2 as the first traversal variable, initialize the value of j2 to 1, if j2≤L, use list( ) indicates that the master user UsA is in the slave user Purchase list, list( )=(M ), in this range, j1 and j2 are variables, and j2=j1, E represents the master user UsA in the slave user The total number of purchased items, get the purchase list list (UsA) and list ( ), set a variable as the main intersection quantity for each element in the purchase list of the main user UsA, and use the number of elements in the intersection as the main intersection quantity F(j2) of the j2th element in list(UsA); S402, set a variable j3 as the second traversal variable, initialize the value of i3 to 1, create an empty set as the purchase list listls, set a variable as the total intersection F, initialize the value of the total intersection F to 0, the j3 traverses all slave users UsB, defines the slave user is the main user traversed by j3, ∈ , obtained from the user List of products listls( ) and from the user List of products listls( ) of the intersection TSub( , ), the intersection TSub( , ) as the sub-intersection component TSub(j3), and the sum of the sub-intersection component TSub(j3) and the sub-intersection total F as the new sub-intersection total F. If there are elements in TSub(j3) that do not exist in FEOLs (UsA), then these elements are added to the set PL; S403: The purchase degree λ is 1. The arithmetic mean of the number of elements in the product list of each slave user in the PL is obtained as the purchase mean Efans. The purchase degree λ of all elements of the master user in the PL is calculated. The method for calculating the purchase degree λ is as follows: λ= +1; Where j4 is the subscript of the set PL; merge the elements in PL into list(UsA), L is the total number of all slave users, if the purchase degree λ is not 1, calculate the intersection coefficient ACrs, set a variable j5 as the serial number of the middle element in FEOLs(UsA), where the intersection coefficient ACrsi5 of the i5th element in FEOLs(UsA) is calculated as: ; in represents the mutual attention of the j5th element in FEOLs(UsA), and They represent the main intersection and sub-intersection of the j5th element in list(UsA), m1 is the cumulative variable, U( ) represents the value ratio of the set of all SCrs values The average value of the large values ​​is used, and the set of each intersection coefficient ACrs is used as the intersection space zAC. The intersection space zAC is output and used as the push strategy.

[0027] Preferably, through the intersection space zAC, the purchased items and purchasing preferences of the main user UsA can be effectively quantified, the multiple purchasing behaviors between each merchant user and the main user in the main user's purchasing behavior can be quantified, and the secondary intersection volume can be found among the items of multiple merchants purchased by the main user to further refine the main user's purchasing preferences. At the same time, based on the fact that there are still certain items that the main user is interested in besides the secondary intersection volume and the main intersection volume, these types of items are also added to the intersection space as a reference, which is more helpful for subsequent update push decisions and deepens the mining of the main user's product preferences compared with traditional personal preference analysis.

[0028] Furthermore, a method for obtaining a user behavior coefficient from a purchase list is as follows: obtaining purchase records between a user UsB and a list (UsA) within a period of time, wherein the period of time is a value ranging from 1 to 180 days, and the product information is product information posted by the user; setting a user behavior coefficient Sml for each element in the list (UsA), wherein the initial value of the user behavior coefficient Sml is 0; If a user performs at least one of the following actions: returning a product, not paying after placing an order, abnormal return behavior, or multiple times, a Boolean variable is set as a negative label NFlag for the user's purchase information and the value is FALSE; if a user performs at least one of the following actions: unfollowing, disliking, or reporting, a Boolean variable is set as a negative label NFlag for the user's purchase information and the value is TRUE; Initialize the value of the user behavior coefficient Sml to 0; filter out the co-viewed information RMsg of the main user UsA with the negative label NFLag as the reference purchase information, and calculate the main user behavior coefficient Sml for each user who browses the reference purchase information of the same product: if the negative label NFLag of the reference purchase information for the user corresponding to an element in list(UsA) is equal to the negative label NFLag of the main user UsA, then add 1 to the main user behavior coefficient Sml of the element; otherwise, subtract 1 from the main user behavior coefficient Sml of the element; traverse all reference purchases and calculate the main user behavior coefficient Sml for each element in the corresponding list(UsA), and output the user behavior coefficient score.

[0029] A user portrait intelligent analysis and management system based on cloud computing and AI, the system comprising: a cloud computing platform, a processor and a memory. The cloud computing platform organizes and updates user portraits by acquiring user behavior. The memory stores data output from the cloud computing platform and the processor. When the processor executes the computer program, it can implement any step in the user portrait intelligent optimization method based on cloud computing and AI in the above method.

[0030] Preferably, in the above steps, user information of the main user is obtained, and user portrait features of the target user are determined based on a preset portrait algorithm; The push strategy will be updated based on the user profile of the primary user and the purchase behavior of the primary user and secondary users; The update push strategy is used to modify the homepage push results and modify the user search results; optionally, the product category preference push strategy and the modified category preference prediction results both include preference push probabilities corresponding to multiple product categories; the product purchase method prediction results and the modified purchase method prediction results both include prediction probabilities corresponding to multiple purchase methods; the product category preference AI and convolutional neural network participate in the calculation, the product push strategy neural network and the product purchase method prediction neural network are all trained through a training data set including multiple training user features and corresponding device data annotations, channel data annotations and prediction target annotations.

[0031] It can be seen that through the above optional embodiments, the user's product category preference push strategy is determined through intersection screening calculation and neural network algorithm, so as to more accurately calculate the user portrait features, thereby assisting in the subsequent realization of more intelligent and reasonable user portrait construction and user product recommendations, thereby improving the recommendation effect.

[0032] The processor may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete component gate circuits or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the system, connecting various sub-areas of the entire system using various interfaces and lines.

[0033] The memory can be used to store the computer programs and / or modules, and the processor implements various functions of the system by running or executing the computer programs and / or modules stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0034] Although the present invention has been described in considerable detail and with particularity with respect to several embodiments, it is not intended to limit the present invention to any of these details or embodiments or any particular embodiment, so as to effectively encompass the intended scope of the present invention. In addition, the present invention has been described above with respect to embodiments foreseen by the inventors for the purpose of providing a useful description, and those insubstantial modifications of the present invention that are not currently foreseen may still represent equivalent modifications of the present invention.

Claims

1. The user portrait intelligent optimization method based on cloud computing and AI is characterized by: The method comprises the following steps: S100: Obtain basic user information on the platform and construct user profiles based on the basic user information; S200: organizing user profiles based on the cloud platform, and pushing products to users based on user behavior information in the user profiles and historical user behavior information; S300: Cleaning redundant information in the user portrait and encrypting user data in the user portrait; S400: Calculate and update the push strategy based on the user profile, and save the user's purchase degree and the number of times the user triggers sensitive behaviors in the user profile; S500: Output the push strategy and the number of times the user triggers sensitive behaviors to score and classify the user.

2. The user portrait intelligent optimization method based on cloud computing and AI according to claim 1 is characterized in that: In step S100, user basic information is obtained through the cross-border e-commerce platform. The user basic information includes: user preference information and user behavior information. The user preference information is defined by the number of user browsing times and the number of user interactions. The user behavior information includes browsing behavior information and interaction behavior information. A user profile is constructed by combining the user preference information and the user behavior information with the user basic information. The user basic information is the user-defined setting parameters and the user number. The user's pushed products and the preliminary user behavior score are determined by the user profile.

3. The user portrait intelligent optimization method based on cloud computing and AI according to claim 1 is characterized in that: In step S200, cloud user update data is obtained through the cloud platform, and the user portrait is updated based on the user update data. In the user portrait, the user's browsing product type is defined as com_1, and the similar product type is defined as com_2 through collaborative filtering of product com_1. It is recorded whether the user chooses to browse the products of product com_2, and the number of times the user browses product com_1 and product com_2 is recorded. A new product type is recommended to the user based on the recorded number of times the product com_2 is browsed, which is defined as com_3. According to the new browsing product type of the user, products are pushed according to the method. The push method is used to sort the pushed products according to the number of views and click-through rate through a logistic regression algorithm.

4. The user portrait intelligent optimization method based on cloud computing and AI according to claim 1 is characterized in that: In step S300, the product push strategy is obtained by combining the user preference information and user behavior information in the user portrait with the user feedback information after the product is pushed through the logistic regression algorithm, recording the page views and click-through rates as well as the types of browsed products, filtering the redundant data in the user portrait, and recording the repeated product types in the recorded page views and click-through rates as the browsing weight of the changed product type by filtering the product types. The redundantly recorded page views and click-through rates are deleted in the cloud platform to reduce the space occupied by the cloud platform for calculating user portrait data, and the redundantly recorded data is recorded in the traceability channel in the background system.

5. The user portrait intelligent optimization method based on cloud computing and AI according to claim 4 is characterized in that: The specific method for encrypting user data in a user portrait comprises the following steps: dividing the user portrait data into a plurality of nodes by date, and forming a node graph from the nodes; In the node graph, each node records two attribute values: the number of user visits and the amount of system data accesses. Each time a user visit is recorded, an edge is established in the node graph. The amount of system data accesses during this user visit is recorded on this edge. The commodity type of the user visit is all commodities. The amount of system data accesses is the number of bytes of data read by the node through this edge. The amount of system data accesses is the number of bytes used by the system to update the user profile based on the number of user visits. In the node graph, the set of nodes is recorded as Nset, the number of nodes is n, the sequence number of the node is i, i∈[1,n], the element with sequence number i in Nset is Nset(i), and each edge of each node in the node graph is connected to the adjacent nodes in the node graph; Based on the encryption vector of each node, the method for determining the data access rights between nodes is as follows: The set of encryption vectors obtained by each node through the encryption module is used as the encryption vector set; For any node, let it be Nset(i), let the number of edges of the node be s, let the sequence number of the edges of the node be t, t∈[1,s], let the edge with sequence number t in the edges of Nset(i) be E(i,t), let the data access volume on E(i,t) be B(i,t), let the encrypted vector of the node be Vec(i), Vec(i) be an s-dimensional vector, let the sequence number of the dimension in Vec(i) be t, and let the value of the dimension with sequence number t in Vec(i) be Vec(i,t); Calculate the arithmetic mean of the values ​​of each dimension in Vec(i) as the screening value, and record the screening value as η(i). Compare the values ​​of the dimensions of each sequence number in Vec(i) with the screening value η(i). Filter out the sequence numbers of the dimensions in Vec(i) whose values ​​are less than the screening value η(i) as the screening sequence number set; The screening ratio value calculated according to Vec(i) is recorded as P(i). The calculation formula of P(i) is: ； The function exp is an exponential function with the natural number e as the base, thereby obtaining the screening ratio value P(i); Calculate the average system data access volume on each edge of node Nset(i), and denote the average data access volume on each edge of node Nset(i) as bf(i). The calculation formula of bf(i) is: ； According to the sequence number in the filter sequence number set, select the edges with corresponding sequence numbers from the edges of node Nset(i) to form the edge set to be filtered. Any edge in the edge set to be filtered is denoted as E(i, tb). The sequence number of edge E(i, tb) in the edges of Nset(i) is tb, tb∈[1,s], and the data access volume on E(i, tb) is B(i, tb). Determine whether the constraint B(i, tb)\[bf(i)* P(i)]>1 is met. If so, delete the data reading permission of edge E(i, tb) on node Nset(i). That is, the node originally connected to Nset(i) through edge E(i, tb) loses the permission to read the data of node Nset(i).

6. The user portrait intelligent optimization method based on cloud computing and AI according to claim 1 is characterized in that: In step S400, the method of updating the push strategy by user portrait is as follows: randomly select a user as the main user UsA, the merchant who conducts transactions with the main user UsA as the secondary user denoted as UsB, and the set of all secondary users as the purchase list of the main user UsA list (UsA), list (UsA) = {M }, j1∈[1, L], where j1 is the serial number of the slave user UsB in the purchase list list(UsA) of UsA. The order of the slave user UsB is based on the purchase time of the master user UsA. If the master user purchases from the same slave user multiple times, the first purchase is used as the ordering basis, and the order is based on the purchase time. represents the j1th element in the purchase list, L represents the number of slave users UsB in list (UsA), k represents the number of different items purchased by the same slave user, M It is represented as the kth item purchased from the user at position j1; S401, set a variable j2 as the first traversal variable, initialize the value of j2 to 1, if j2≤L, use list( ) indicates that the master user UsA is in the slave user Purchase list, list( )=(M ), in this range, j1 and j2 are variables, and j2=j1, E represents the master user UsA in the slave user The total number of purchased items, get the purchase list list (UsA) and list ( ), set a variable as the main intersection quantity for each element in the purchase list of the main user UsA, and use the number of elements in the intersection as the main intersection quantity F(j2) of the j2th element in list(UsA); S402, set a variable j3 as the second traversal variable, initialize the value of i3 to 1, create an empty set as the purchase list listls, set a variable as the total intersection F, initialize the value of the total intersection F to 0, the j3 traverses all slave users UsB, defines the slave user is the main user traversed by j3, ∈ , obtained from the user List of products listls( ) and from the user List of products listls( ) of the intersection TSub( , ), the intersection TSub( , ) as the sub-intersection component TSub(j3), and the sum of the sub-intersection component TSub(j3) and the sub-intersection total F as the new sub-intersection total F. If there are elements in TSub(j3) that do not exist in FEOLs (UsA), then these elements are added to the set PL; S403: The purchase degree λ is 1. The arithmetic mean of the number of elements in the product list of each slave user in the PL is obtained as the purchase mean Efans. The purchase degree λ of all elements of the master user in the PL is calculated. The method for calculating the purchase degree λ is as follows: λ= +1; Where j4 is the subscript of the set PL; merge the elements in PL into list(UsA), L is the total number of all slave users, if the purchase degree λ is not 1, calculate the intersection coefficient ACrs, set a variable j5 as the serial number of the middle element in FEOLs(UsA), where the intersection coefficient ACrsi5 of the i5th element in FEOLs(UsA) is calculated as: ； in represents the mutual attention of the j5th element in FEOLs(UsA), and They represent the main intersection and sub-intersection of the j5th element in list(UsA), m1 is the cumulative variable, U( ) represents the value ratio of the set of all SCrs values The average value of the large values ​​is used, and the set of each intersection coefficient ACrs is used as the intersection space zAC. The intersection space zAC is output and used as the push strategy.

7. The user portrait intelligent optimization method based on cloud computing and AI according to claim 6 is characterized in that: The method for obtaining the user behavior coefficient from the purchase list is as follows: obtaining purchase records between the list (UsA) and the slave user UsB within a period of time, wherein the range of the period is 1 to 180 days, and the product information is the product information posted by the slave user; setting the user behavior coefficient Sml for each element in the list (UsA), and the initial value of the user behavior coefficient Sml is 0; If a user performs at least one of the following actions: returning a product, not paying after placing an order, abnormal return behavior, or multiple times, a Boolean variable is set as a negative label NFlag for the user's purchase information and the value is FALSE; if a user performs at least one of the following actions: unfollowing, disliking, or reporting, a Boolean variable is set as a negative label NFlag for the user's purchase information and the value is TRUE; Initialize the value of the user behavior coefficient Sml to 0; filter out the co-viewed information RMsg of the main user UsA with the negative label NFLag as the reference purchase information, and calculate the main user behavior coefficient Sml for each user who browses the reference purchase information of the same product: if the negative label NFLag of the reference purchase information for the user corresponding to an element in list(UsA) is equal to the negative label NFLag of the main user UsA, then add 1 to the main user behavior coefficient Sml of the element; otherwise, subtract 1 from the main user behavior coefficient Sml of the element; traverse all reference purchases and calculate the main user behavior coefficient Sml for each element in the corresponding list(UsA), and output the user behavior coefficient score.

8. User portrait intelligent analysis and management system based on cloud computing and AI, characterized by: The system includes: a cloud computing platform, a processor and a memory. The cloud computing platform organizes and updates user profiles by obtaining user behavior. The memory stores data output from the cloud computing platform and the processor. When the processor executes the computer program, it can implement any step in the user profile intelligent optimization method based on cloud computing and AI in the above method.

Images