A data management method and system for algorithm recommendation service

Abstract

The present invention relates to the field of data processing technology, and specifically discloses a data management method and system for an algorithm recommendation service, including: obtaining all behavioral data of all users on a network platform within a preset time, and performing data processing on all behavioral data of all users to obtain all real behavioral data of each user; obtaining all similar recommended users of each user based on all real behavioral data of all users; obtaining a recommendation matrix of each user based on each user and all corresponding similar recommended users; obtaining the best recommendation subject of each user based on the recommendation matrix of each user, and obtaining the algorithm recommendation service result based on the best recommendation subject of all users. The present invention accurately obtains the best recommendation subject of each user through all user data of the network platform, optimizes and iterates the recommendation algorithm according to the best recommendation subject of all users, and improves the accuracy of the algorithm recommendation service.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a data management method and system for an algorithm recommendation service. Background Art

[0002] At present, with the development of the Internet, recommendation algorithms are receiving more and more attention. Recommendation algorithms are algorithms that can recommend personalized content to users based on their behavioral data and basic information. Nowadays, various types of network platforms have applied recommendation algorithms to recommend personalized content to users.

[0003] However, how to efficiently manage the data involved in algorithm recommendation services within the network platform and improve the accuracy of the recommendation algorithm has become an urgent problem, and there is currently no way to solve this problem.

[0004] Therefore, the present invention proposes a data management method and system for algorithm recommendation services. Summary of the invention

[0005] The present invention provides a data management method and system for an algorithm recommendation service, which are used to process all behavioral data of all users on a network platform within a preset time, obtain all real behavioral data of each user, facilitate subsequent accurate quantification of interest values, obtain all similar recommended users of each user based on all real behavioral data of all users, accurately screen out the remaining users with similar interests to each user from all users, facilitate subsequent construction of a recommendation matrix, obtain each user's recommendation matrix based on each user and all corresponding similar recommended users, facilitate determination of the best recommendation subject for each user, optimize and iterate the recommendation algorithm based on the best recommendation subjects of all users, and improve the accuracy of the algorithm recommendation service.

[0006] The present invention provides a data management method for an algorithm recommendation service, comprising:

[0007] S1: Obtain all behavioral data of all users on the network platform within a preset time, and process all behavioral data of all users to obtain all real behavioral data of each user;

[0008] S2: Obtain all similar recommended users for each user based on all real behavior data of all users;

[0009] S3: Obtain a recommendation matrix for each user based on each user and all corresponding similar recommended users;

[0010] S4: The best recommendation subject for each user is obtained based on the recommendation matrix of each user, and the algorithm recommendation service result is obtained based on the best recommendation subject of all users.

[0011] Preferably, the data management method for algorithm recommendation service, S1: obtaining all behavior data of all users on the network platform within a preset time, and performing data processing on all behavior data of all users to obtain all real behavior data of each user, including:

[0012] Obtain all behavioral data of all users on the network platform with algorithm recommendation services within a preset time, including clicks, browsing, collections, and purchases;

[0013] All behavioral data of all users are processed to obtain all real behavioral data of each user.

[0014] Preferably, the data management method of the algorithm recommendation service processes all the behavior data of all users to obtain all the real behavior data of each user, including:

[0015] From all the behavior data of each user, obtain the subsequent behavior of the user after each click behavior. If the subsequent behavior of the user after the click behavior is not an exit behavior, the corresponding click behavior is regarded as a real click behavior, and the real click behavior of the corresponding user is regarded as the first real behavior data of the corresponding user;

[0016] From all the behavior data of each user, the browsing time of each browsing behavior of the user is obtained. If the browsing time of a single browsing behavior of the user is greater than the preset browsing time, the corresponding browsing behavior is regarded as the real browsing behavior, and the real browsing behavior of the corresponding user is regarded as the second real behavior data of the corresponding user;

[0017] From all the behavior data of each user, obtain the subsequent behavior of the user after each collection behavior. If the subsequent behavior of the user after the collection behavior is not to cancel the collection, the corresponding collection behavior is regarded as the real collection behavior, and the real collection behavior of the corresponding user is regarded as the third real behavior data of the corresponding user;

[0018] In all the behavior data of each user, the number of stars of the purchase evaluation given by the user after each purchase behavior is obtained. If the number of stars of the purchase evaluation given by the user after the purchase behavior is greater than the preset number of stars, the corresponding purchase behavior is regarded as a real purchase behavior, and the real purchase behavior of the corresponding user is regarded as the fourth real behavior data of the corresponding user;

[0019] The real behavior data of the user includes the first real behavior data, the second real behavior data, the third real behavior data and the fourth real behavior data of the user.

[0020] Preferably, the data management method for algorithm recommendation service, S2: obtaining all similar recommended users of each user based on all real behavior data of all users, including:

[0021] Obtain all recommended subjects on the network platform, and filter out all real behavior data of each user for each recommended subject from all real behavior data of each user;

[0022] Based on the total number of behaviors of each user for each type of real behavior data of each recommendation subject, assign a value to each type of real behavior data of each user for each recommendation subject, and obtain the assigned value of each type of real behavior data of each user for each recommendation subject;

[0023] Based on the numerical values ​​assigned by each user to each type of real behavior data of each recommendation subject, the interest value of each user for each recommendation subject is obtained, which is:

[0024]

[0025] Among them, δ is the user's interest value for a single recommendation subject, S is the sum of the user's assigned values ​​of all categories of real behavior data for the corresponding recommendation subject, and s 1 is the value assigned to the user's real behavior data category of the corresponding recommendation subject, which is click, s 2 is the value assigned to the user's real behavior data category of the corresponding recommendation subject, which is browsing, s 3 is the value assigned to the user for the corresponding recommendation subject’s real behavior data category of collection, s 4 Assign a numerical value to the user's actual behavior data category of the corresponding recommendation subject, which is purchase;

[0026] Based on the interest values ​​of all users in all recommended subjects, all similar recommended users of each user are obtained.

[0027] Preferably, the data management method of the algorithm recommendation service obtains all similar recommended users of each user based on the interest values ​​of all users to all recommended subjects, including:

[0028] Each recommended subject involved in all real behavior data of each user is regarded as the involved subject of the corresponding user;

[0029] All the same involved subjects existing in the involved subjects of the two users are regarded as all the intersection subjects of the two users;

[0030] Based on all the intersection subjects of two users and each user's interest value for each intersection subject, the similarity recommendation value between two users with intersection subjects is obtained, which is:

[0031]

[0032] Among them, ε is the similarity recommendation value between two users, n is the total number of intersection entities between two users, δ aiis the interest value of the first user of the two users currently being calculated in the i-th intersection subject between the two users currently being calculated, δ a0 is the mean of the interest values ​​of the first user among the two users currently being calculated for all the intersection entities between the two users currently being calculated, δ bi is the interest value of the second user of the two users currently being calculated in the i-th intersection subject between the two users currently being calculated, δ b0 is the average of the interest values ​​of the second user of the two users currently being calculated for all the intersection entities between the two users currently being calculated;

[0033] Based on the similar recommendation values ​​between all two users with intersection subjects, all similar recommended users of each user are obtained.

[0034] Preferably, the data management method for the algorithm recommendation service obtains all similar recommended users of each user based on the similar recommendation values ​​between all two users with intersection subjects, including:

[0035] If the similarity recommendation value between two users with an intersection subject is greater than a preset recommendation value threshold, the corresponding two users are determined to be similar recommended users to each other;

[0036] Get all similar recommended users for each user.

[0037] Preferably, the data management method for algorithm recommendation service, S3: obtaining a recommendation matrix for each user based on each user and all corresponding similar recommended users, including:

[0038] According to the similar recommendation values ​​from large to small, all similar recommended users of each user are ordinal-defined starting from 1 to obtain the user ordinal definition result of each user;

[0039] All subjects involved in each user and all similar recommended users of each user are taken as reference subjects of each user, and all reference subjects of each user are ordinal-defined starting from 1 according to the total number of clicks from large to small, to obtain the subject ordinal definition result of each user;

[0040] Based on the user ordinal definition results and subject ordinal definition results of each user and the interest value of each user for each reference subject, the recommendation matrix of each user is obtained, which is:

[0041]

[0042] Among them, E is the user's recommendation matrix, δ 01 is the interest value of a single user to the reference subject with ordinal number 1, δ 02 is the interest value of a single user to the reference subject with ordinal number 2, δ 0mis the interest value of a single user to the reference subject with ordinal number m, m is the total number of all reference subjects of a single user, δ 11 is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number 1 of a single user, δ 12 is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number 2 of a single user, δ 1m is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number m, δ k1 is the interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number 1, δ k2 is the interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number 2 of a single user, δ km The interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number m.

[0043] Preferably, the data management method of the algorithm recommendation service obtains the best recommendation subject for each user based on the recommendation matrix of each user, including:

[0044] Eliminate the entire row of elements in each user's recommendation matrix in turn, and obtain a new matrix after each elimination, which is used as the calculation matrix of the reference subject corresponding to the entire row of elements eliminated;

[0045] From all reference subjects of each user, the reference subject with the largest rank of the corresponding calculation matrix is ​​selected as the best recommended subject for the corresponding user.

[0046] Preferably, the data management method for the algorithm recommendation service obtains the algorithm recommendation service result based on the best recommendation subject of all users, including:

[0047] Obtain basic information of each user, including gender, age, and identity attributes;

[0048] The basic information of each user and the best recommended subject are used as training samples, and the recommendation algorithm adopted by the recommendation algorithm service is optimized and replaced to obtain a new recommendation algorithm for the recommendation algorithm service as the algorithm recommendation service result.

[0049] The present invention provides a data management method system for an algorithm recommendation service, which is used to execute any one of the data management methods for the algorithm recommendation service in embodiments 1 to 9, including:

[0050] The data processing module is used to obtain all the behavior data of all users on the network platform within a preset time, and to process all the behavior data of all users to obtain all the real behavior data of each user;

[0051] A calculation module, used to obtain all similar recommended users for each user based on all real behavior data of all users;

[0052] A construction module is used to obtain a recommendation matrix for each user based on each user and all corresponding similar recommended users;

[0053] The recommendation optimization module is used to obtain the best recommendation subject for each user based on the recommendation matrix of each user, and obtain the algorithm recommendation service result based on the best recommendation subject of all users.

[0054] Compared with the prior art, the beneficial effects of the present invention are as follows: data processing is performed on all behavioral data of all users on the network platform within a preset time to obtain all real behavioral data of each user, which is convenient for subsequent accurate quantification of interest values; all similar recommended users of each user are obtained based on all real behavioral data of all users; the remaining users with similar interests to each user are accurately screened out from all users to facilitate the subsequent construction of a recommendation matrix; a recommendation matrix of each user is obtained based on each user and all corresponding similar recommended users to facilitate determination of the best recommendation subject for each user; the recommendation algorithm is optimized and iterated based on the best recommendation subjects of all users, thereby improving the accuracy of the algorithm recommendation service.

[0055] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structure specifically pointed out in the written application document.

[0056] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0058] Figure 1 A flow chart of a data management method for an algorithm recommendation service in an embodiment of the present invention;

[0059] Figure 2 A schematic diagram of a data management system for an algorithm recommendation service in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0061] Embodiment 1:

[0062] The present invention provides a data management method for an algorithm recommendation service, referring to Figure 1 ,include:

[0063] S1: Obtain all behavioral data of all users on the network platform within a preset time, and process all behavioral data of all users to obtain all real behavioral data of each user;

[0064] S2: Obtain all similar recommended users for each user based on all real behavior data of all users;

[0065] S3: Obtain a recommendation matrix for each user based on each user and all corresponding similar recommended users;

[0066] S4: The best recommendation subject for each user is obtained based on the recommendation matrix of each user, and the algorithm recommendation service result is obtained based on the best recommendation subject of all users.

[0067] In this embodiment, the network platform is a network platform with algorithm recommendation services, such as an online film and television platform, an online novel platform, an online shopping platform, etc.

[0068] In this embodiment, the preset time is a time period preset to obtain all behavior data of all users on the network platform, such as 30 minutes.

[0069] In this embodiment, data processing is the process of processing all behavior data of all users to obtain all real behavior data of each user.

[0070] In this embodiment, the real behavior data refers to the behavior data of each user that can truly reflect the user's behavior intention.

[0071] In this embodiment, similar recommended users are other users with similar interests to each user obtained based on all real behavior data of all users.

[0072] In this embodiment, the recommendation matrix is ​​a matrix containing the interest values ​​of each user and all corresponding similar recommended users to all reference subjects of the corresponding user.

[0073] In this embodiment, the best recommendation subject is the recommendation subject that each user is most interested in, obtained based on the recommendation matrix of each user.

[0074] In this embodiment, the algorithm recommendation service result is the result of optimizing and iterating the recommendation algorithm (in the form of a neural network model) adopted by the recommendation algorithm service by using each user's basic information and the best recommended subject as training samples (i.e., a new neural network model obtained after optimization and iteration).

[0075] The beneficial effects of the above technology are: data processing is performed on all behavioral data of all users on the network platform within a preset time to obtain all real behavioral data of each user, which is convenient for the subsequent accurate quantification of interest values, and all similar recommended users of each user are obtained based on all real behavioral data of all users, and the remaining users with similar interests as each user are accurately screened out from all users to facilitate the subsequent construction of the recommendation matrix, and the recommendation matrix of each user is obtained based on each user and all corresponding similar recommended users, which is convenient for determining the best recommendation subject for each user, and the recommendation algorithm is optimized and replaced according to the best recommendation subjects of all users, thereby improving the accuracy of the algorithm recommendation service.

[0076] Embodiment 2:

[0077] Based on Example 1, the data management method for algorithm recommendation service, S1: obtaining all behavior data of all users on the network platform within a preset time, and performing data processing on all behavior data of all users to obtain all real behavior data of each user, including:

[0078] Obtain all behavioral data of all users on the network platform with algorithm recommendation services within a preset time, including clicks, browsing, collections, and purchases;

[0079] All behavioral data of all users are processed to obtain all real behavioral data of each user.

[0080] The beneficial effect of the above technology is: all behavioral data of all users on the network platform within a preset time are processed to obtain all real behavioral data of each user, which is convenient for subsequent accurate quantification of interest values.

[0081] Embodiment 3:

[0082] On the basis of Example 2, the data management method of the algorithm recommendation service processes all the behavior data of all users to obtain all the real behavior data of each user, including:

[0083] From all the behavior data of each user, obtain the subsequent behavior of the user after each click behavior. If the subsequent behavior of the user after the click behavior is not an exit behavior, the corresponding click behavior is regarded as a real click behavior, and the real click behavior of the corresponding user is regarded as the first real behavior data of the corresponding user;

[0084] From all the behavior data of each user, the browsing time of each browsing behavior of the user is obtained. If the browsing time of a single browsing behavior of the user is greater than the preset browsing time, the corresponding browsing behavior is regarded as the real browsing behavior, and the real browsing behavior of the corresponding user is regarded as the second real behavior data of the corresponding user;

[0085] From all the behavior data of each user, obtain the subsequent behavior of the user after each collection behavior. If the subsequent behavior of the user after the collection behavior is not to cancel the collection, the corresponding collection behavior is regarded as the real collection behavior, and the real collection behavior of the corresponding user is regarded as the third real behavior data of the corresponding user;

[0086] In all the behavior data of each user, the number of stars of the purchase evaluation given by the user after each purchase behavior is obtained. If the number of stars of the purchase evaluation given by the user after the purchase behavior is greater than the preset number of stars, the corresponding purchase behavior is regarded as a real purchase behavior, and the real purchase behavior of the corresponding user is regarded as the fourth real behavior data of the corresponding user;

[0087] The real behavior data of the user includes the first real behavior data, the second real behavior data, the third real behavior data and the fourth real behavior data of the user.

[0088] In this embodiment, the preset browsing time is a time preset to determine whether the browsing behavior of the user is a real browsing behavior, for example, 20 seconds.

[0089] In this embodiment, the subsequent behavior of a single behavior is the behavior of the corresponding user after the single behavior, that is, the next behavior adjacent to the single behavior.

[0090] In this embodiment, the exit behavior is an act of exiting the network platform or exiting the corresponding recommended content (or subject).

[0091] In this embodiment, the purchase evaluation star number is the user's evaluation level for the recommended content or subject purchased in the network platform. The higher the purchase evaluation star number, the higher the evaluation level, and vice versa. The purchase evaluation star number ranges from 1 to 5 stars.

[0092] The beneficial effect of the above technology is: based on the real behavior data screening conditions corresponding to each behavior category, all the behavior data of all users on the network platform within a preset time are screened to obtain all the real behavior data of each user. This embodiment provides a method for processing all the behavior data of all users on the network platform within a preset time to obtain all the real behavior data of each user.

[0093] Embodiment 4:

[0094] Based on Example 1, the data management method for algorithm recommendation service, S2: obtaining all similar recommended users of each user based on all real behavior data of all users, including:

[0095] Obtain all recommended subjects on the network platform, and filter out all real behavior data of each user for each recommended subject from all real behavior data of each user;

[0096] Based on the total number of behaviors of each user for each type of real behavior data of each recommendation subject, assign a value to each type of real behavior data of each user for each recommendation subject, and obtain the assigned value of each type of real behavior data of each user for each recommendation subject;

[0097] Based on the numerical values ​​assigned by each user to each type of real behavior data of each recommendation subject, the interest value of each user for each recommendation subject is obtained, which is:

[0098]

[0099] Among them, δ is the user's interest value for a single recommendation subject, S is the sum of the user's assigned values ​​of all categories of real behavior data for the corresponding recommendation subject, and s 1 is the value assigned to the user's real behavior data category of the corresponding recommendation subject, which is click, s 2 is the value assigned to the user's real behavior data category of the corresponding recommendation subject, which is browsing, s 3 is the value assigned to the user for the corresponding recommendation subject’s real behavior data category of collection, s 4 Assign a numerical value to the user's actual behavior data category of the corresponding recommendation subject, which is purchase;

[0100] Based on the interest values ​​of all users in all recommended subjects, all similar recommended users of each user are obtained.

[0101] In this embodiment, the recommendation subject is a specific subject recommended to users on a network platform with an algorithm recommendation service, for example, the recommendation subject of an online novel platform is novels, and the recommendation subject of an online shopping platform is merchandise.

[0102] In this embodiment, each user's each type of real behavior data for each recommendation subject is assigned a value by taking the product of the total number of behaviors of each user for each type of real behavior data for each recommendation subject and a preset coefficient of the corresponding category of real behavior data (a coefficient preset to assign a value to each type of real behavior data of each recommendation subject by the user, and each category of real behavior data corresponds to a preset coefficient) as the process of assigning the corresponding user to the corresponding category of real behavior data of the corresponding recommendation subject.

[0103] In this embodiment, the assigned value is the result of each user assigning a value to each type of real behavior data of each recommendation subject.

[0104] In this embodiment, the interest value is a numerical value representing the interest level of each user in each recommendation subject.

[0105] The beneficial effects of the above technology are: obtaining the interest values ​​of all users in all recommended subjects based on all real behavior data of all users, realizing the specific quantification of the user's interest in each recommended subject, and obtaining all similar recommended users of each user based on the interest values ​​of all users in all recommended subjects, that is, accurately screening out the remaining users with similar interests to each user from all users, facilitating the construction of the subsequent recommendation matrix. This embodiment provides a method for obtaining the interest values ​​of all users in all recommended subjects based on all real behavior data of all users.

[0106] Embodiment 5:

[0107] On the basis of Example 4, the data management method of the algorithm recommendation service obtains all similar recommended users of each user based on the interest values ​​of all users to all recommended subjects, including:

[0108] Each recommended subject involved in all real behavior data of each user is regarded as the involved subject of the corresponding user;

[0109] All the same involved subjects existing in the involved subjects of the two users are regarded as all the intersection subjects of the two users;

[0110] Based on all the intersection subjects of two users and each user's interest value for each intersection subject, the similarity recommendation value between two users with intersection subjects is obtained, which is:

[0111]

[0112] Among them, ε is the similarity recommendation value between two users, n is the total number of intersection entities between two users, δ ai is the interest value of the first user of the two users currently being calculated in the i-th intersection subject between the two users currently being calculated, δ a0 is the mean of the interest values ​​of the first user among the two users currently being calculated for all the intersection entities between the two users currently being calculated, δ bi is the interest value of the second user of the two users currently being calculated in the i-th intersection subject between the two users currently being calculated, δ b0 is the average of the interest values ​​of the second user of the two users currently being calculated for all the intersection entities between the two users currently being calculated;

[0113] Based on the similar recommendation values ​​between all two users with intersection subjects, all similar recommended users of each user are obtained.

[0114] The beneficial effects of the above technology are: filtering out users with overlapping subjects, and accurately calculating the similar recommendation values ​​between users with overlapping subjects based on the interest values ​​of all users in all recommended subjects, and obtaining all similar recommended users of each user based on the similar recommendation values ​​between users with overlapping subjects, that is, accurately filtering out the remaining users with similar interests to each user from all users, which is convenient for constructing the subsequent recommendation matrix. This embodiment provides a method for obtaining all similar recommended users of each user based on the interest values ​​of all users in all recommended subjects.

[0115] Embodiment 6:

[0116] On the basis of Example 5, the data management method of the algorithm recommendation service obtains all similar recommended users of each user based on the similar recommendation values ​​between all two users with intersection subjects, including:

[0117] If the similarity recommendation value between two users with an intersection subject is greater than a preset recommendation value threshold, the corresponding two users are determined to be similar recommended users to each other;

[0118] Get all similar recommended users for each user.

[0119] In this embodiment, the preset recommendation value threshold is a threshold of similar recommendation values ​​that is preset to determine whether two users with an intersection are similar recommended users to each other.

[0120] The beneficial effect of the above technology is: according to the similar recommendation value between two users with an intersection subject and the preset recommendation value threshold, it is accurately judged whether two users are similar recommended users to each other, which is convenient for the construction of the subsequent recommendation matrix.

[0121] Embodiment 7:

[0122] Based on Example 1, the data management method for algorithm recommendation service, S3: obtaining a recommendation matrix for each user based on each user and all corresponding similar recommended users, including:

[0123] According to the similar recommendation values ​​from large to small, all similar recommended users of each user are ordinal-defined starting from 1 to obtain the user ordinal definition result of each user;

[0124] All subjects involved in each user and all similar recommended users of each user are taken as reference subjects of each user, and all reference subjects of each user are ordinal-defined starting from 1 according to the total number of clicks from large to small, to obtain the subject ordinal definition result of each user;

[0125] Based on the user ordinal definition results and subject ordinal definition results of each user and the interest value of each user for each reference subject, the recommendation matrix of each user is obtained, which is:

[0126]

[0127] Among them, E is the user's recommendation matrix, δ 01 is the interest value of a single user to the reference subject with ordinal number 1, δ 02 is the interest value of a single user to the reference subject with ordinal number 2, δ 0m is the interest value of a single user to the reference subject with ordinal number m, m is the total number of all reference subjects of a single user, δ 11 is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number 1 of a single user, δ 12 is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number 2 of a single user, δ 1m is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number m, δ k1 is the interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number 1, δ k2 is the interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number 2 of a single user, δ km The interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number m.

[0128] The beneficial effect of the above technology is: a recommendation matrix for each user is constructed based on the interest values ​​of each user and all corresponding similar recommended users in each reference subject of the user, which facilitates the acquisition of the best recommended subject for each user. This embodiment provides a method for constructing a recommendation matrix for the corresponding user based on the user ordinal definition results and subject ordinal definition results of each user and the interest value of each user in each reference subject.

[0129] Embodiment 8:

[0130] Based on Example 1, the data management method of the algorithm recommendation service obtains the best recommendation subject for each user based on the recommendation matrix of each user, including:

[0131] Eliminate the entire row of elements in each user's recommendation matrix in turn, and obtain a new matrix after each elimination, which is used as the calculation matrix of the reference subject corresponding to the entire row of elements eliminated;

[0132] From all reference subjects of each user, the reference subject with the largest rank of the corresponding calculation matrix is ​​selected as the best recommended subject for the corresponding user.

[0133] In this embodiment, the calculation matrix is ​​a new matrix obtained by removing the entire row of elements in the recommendation matrix of each user.

[0134] In this embodiment, the entire row of elements in the recommendation matrix of each user is removed in turn to obtain a new matrix after each removal, for example:

[0135] Assuming that the user's recommendation matrix has n rows of elements, delete an entire row of elements in the recommendation matrix each time, and obtain the calculation matrix of the reference subject corresponding to the deleted entire row of elements;

[0136] According to this principle, the recommendation matrix is ​​eliminated n times to obtain n calculation matrices;

[0137] Each calculation matrix has (n-1) rows of elements.

[0138] The beneficial effects of the above technology are: by eliminating the user's recommendation matrix according to the rules, the calculation matrices of different reference subjects are obtained respectively, and the best recommendation subject of each user is screened out based on the rank of the calculation matrix, that is, the best recommendation subject of each user is obtained according to the recommendation matrix of each user, which is convenient for subsequent optimization and iteration of the recommendation algorithm.

[0139] Embodiment 9:

[0140] On the basis of Example 1, the data management method of the algorithm recommendation service obtains the algorithm recommendation service result based on the best recommendation subject of all users, including:

[0141] Obtain basic information of each user, including gender, age, and identity attributes (such as student, accountant, etc.);

[0142] The basic information of each user and the best recommended subject are used as training samples, and the recommendation algorithm adopted by the recommendation algorithm service is optimized and replaced to obtain a new recommendation algorithm for the recommendation algorithm service as the algorithm recommendation service result.

[0143] In this embodiment, the basic information of each user and the best recommended subject are used as training samples to optimize and iterate the recommendation algorithm used by the recommendation algorithm service, including:

[0144] Because the recommendation algorithm exists in the form of a neural network model, during the training process, the basic information of the users in the training samples is used as the model input, and the corresponding best recommendation subject is used as the model output. The model corresponding to the recommendation algorithm is optimized and iterated to obtain a new model as a new recommendation algorithm for the recommendation algorithm service.

[0145] The beneficial effect of the above technology is: the recommendation algorithm is optimized and updated according to the best recommendation subjects of all users, thereby improving the accuracy of the algorithm recommendation service.

[0146] Embodiment 10:

[0147] The present invention provides a data management method system for an algorithm recommendation service, which is used to execute the data management method for an algorithm recommendation service in any one of embodiments 1 to 9, referring to Figure 2 ,include:

[0148] The data processing module is used to obtain all the behavior data of all users on the network platform within a preset time, and to process all the behavior data of all users to obtain all the real behavior data of each user;

[0149] A calculation module, used to obtain all similar recommended users for each user based on all real behavior data of all users;

[0150] A construction module is used to obtain a recommendation matrix for each user based on each user and all corresponding similar recommended users;

[0151] The recommendation optimization module is used to obtain the best recommendation subject for each user based on the recommendation matrix of each user, and obtain the algorithm recommendation service result based on the best recommendation subject of all users.

[0152] The beneficial effects of the above technology are: data processing is performed on all behavioral data of all users on the network platform within a preset time to obtain all real behavioral data of each user, which is convenient for the subsequent accurate quantification of interest values, and all similar recommended users of each user are obtained based on all real behavioral data of all users, and the remaining users with similar interests as each user are accurately screened out from all users to facilitate the subsequent construction of the recommendation matrix, and the recommendation matrix of each user is obtained based on each user and all corresponding similar recommended users, which is convenient for determining the best recommendation subject for each user, and the recommendation algorithm is optimized and replaced according to the best recommendation subjects of all users, thereby improving the accuracy of the algorithm recommendation service.

[0153] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A data management method for an algorithm recommendation service, characterized in that: include: S1: Obtain all behavioral data of all users on the network platform within a preset time, and process all behavioral data of all users to obtain all real behavioral data of each user; S2: Obtain all similar recommended users for each user based on all real behavior data of all users; S3: Obtain a recommendation matrix for each user based on each user and all corresponding similar recommended users; S4: Obtain the best recommendation subject for each user based on the recommendation matrix of each user, and obtain the algorithm recommendation service result based on the best recommendation subjects of all users; Among them, S2: obtains all similar recommended users for each user based on all real behavior data of all users, including: Obtain all recommended subjects on the network platform, and filter out all real behavior data of each user for each recommended subject from all real behavior data of each user; Based on the total number of behaviors of each user for each type of real behavior data of each recommendation subject, assign a value to each type of real behavior data of each user for each recommendation subject, and obtain the assigned value of each type of real behavior data of each user for each recommendation subject; Based on the numerical values ​​assigned by each user to each type of real behavior data of each recommendation subject, the interest value of each user for each recommendation subject is obtained, which is: Where δ is the user's interest value for a single recommendation subject, S is the sum of the values ​​assigned by the user to all categories of the real behavior data of the corresponding recommendation subject, s1 is the value assigned by the user when the real behavior data category of the corresponding recommendation subject is click, s2 is the value assigned by the user when the real behavior data category of the corresponding recommendation subject is browse, s3 is the value assigned by the user when the real behavior data category of the corresponding recommendation subject is favorite, and s4 is the value assigned by the user when the real behavior data category of the corresponding recommendation subject is purchase; Based on the interest values ​​of all users in all recommended subjects, all similar recommended users of each user are obtained; Among them, based on the interest values ​​of all users for all recommended subjects, all similar recommended users of each user are obtained, including: Each recommended subject involved in all real behavior data of each user is regarded as the involved subject of the corresponding user; All the same involved subjects existing in the involved subjects of the two users are regarded as all the intersection subjects of the two users; Based on all the intersection subjects of two users and each user's interest value for each intersection subject, the similarity recommendation value between two users with intersection subjects is obtained, which is: Among them, ε is the similarity recommendation value between two users, n is the total number of intersection entities between two users, δ ai is the interest value of the first user of the two users currently being calculated in the i-th intersection subject between the two users currently being calculated, δ a0 is the mean of the interest values ​​of the first user among the two users currently being calculated for all the intersection entities between the two users currently being calculated, δ bi is the interest value of the second user of the two users currently being calculated in the i-th intersection subject between the two users currently being calculated, δ b0 is the average of the interest values ​​of the second user of the two users currently being calculated for all the intersection entities between the two users currently being calculated; Based on the similar recommendation values ​​between all two users with intersection subjects, all similar recommended users of each user are obtained; Among them, based on the similar recommendation values ​​between all two users with intersection subjects, all similar recommended users of each user are obtained, including: If the similarity recommendation value between two users with an intersection subject is greater than a preset recommendation value threshold, the corresponding two users are determined to be similar recommended users to each other; Get all similar recommended users for each user.

2. The data management method for algorithm recommendation service according to claim 1, characterized in that: S1: Obtain all behavioral data of all users on the network platform within a preset time, and process all behavioral data of all users to obtain all real behavioral data of each user, including: Obtain all behavioral data of all users on the network platform with algorithm recommendation services within a preset time, including clicks, browsing, collections, and purchases; All behavioral data of all users are processed to obtain all real behavioral data of each user.

3. The data management method for algorithm recommendation service according to claim 2, characterized in that: Process all behavior data of all users to obtain all real behavior data of each user, including: From all the behavior data of each user, obtain the subsequent behavior of the user after each click behavior. If the subsequent behavior of the user after the click behavior is not an exit behavior, the corresponding click behavior is regarded as a real click behavior, and the real click behavior of the corresponding user is regarded as the first real behavior data of the corresponding user; From all the behavior data of each user, the browsing time of each browsing behavior of the user is obtained. If the browsing time of a single browsing behavior of the user is greater than the preset browsing time, the corresponding browsing behavior is regarded as the real browsing behavior, and the real browsing behavior of the corresponding user is regarded as the second real behavior data of the corresponding user; From all the behavior data of each user, obtain the subsequent behavior of the user after each collection behavior. If the subsequent behavior of the user after the collection behavior is not to cancel the collection, the corresponding collection behavior is regarded as the real collection behavior, and the real collection behavior of the corresponding user is regarded as the third real behavior data of the corresponding user; In all the behavior data of each user, the number of stars of the purchase evaluation given by the user after each purchase behavior is obtained. If the number of stars of the purchase evaluation given by the user after the purchase behavior is greater than the preset number of stars, the corresponding purchase behavior is regarded as a real purchase behavior, and the real purchase behavior of the corresponding user is regarded as the fourth real behavior data of the corresponding user; The real behavior data of the user includes the first real behavior data, the second real behavior data, the third real behavior data and the fourth real behavior data of the user.

4. The data management method for algorithm recommendation service according to claim 1, characterized in that: S3: Based on each user and all corresponding similar recommended users, a recommendation matrix for each user is obtained, including: According to the similar recommendation values ​​from large to small, all similar recommended users of each user are ordinal-defined starting from 1 to obtain the user ordinal definition result of each user; All subjects involved in each user and all similar recommended users of each user are taken as reference subjects of each user, and all reference subjects of each user are ordinal-defined starting from 1 according to the total number of clicks from large to small, to obtain the subject ordinal definition result of each user; Based on the user ordinal definition results and subject ordinal definition results of each user and the interest value of each user for each reference subject, the recommendation matrix of each user is obtained, which is: Among them, E is the user's recommendation matrix, δ 01 is the interest value of a single user to the reference subject with ordinal number 1, δ 02 is the interest value of a single user to the reference subject with ordinal number 2, δ 0m is the interest value of a single user to the reference subject with ordinal number m, m is the total number of all reference subjects of a single user, δ 11 is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number 1 of a single user, δ 12 is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number 2 of a single user, δ 1m is the interest value of the similar recommended user with ordinal number 1 for the reference subject with ordinal number m, δ k1 is the interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number 1, δ k2 is the interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number 2 of a single user, δ km The interest value of the similar recommended user with ordinal number k for the reference subject with ordinal number m.

5. The data management method for algorithm recommendation service according to claim 1, characterized in that: Based on each user's recommendation matrix, the best recommendation subject for each user is obtained, including: Eliminate the entire row of elements in each user's recommendation matrix in turn, and obtain a new matrix after each elimination, which is used as the calculation matrix of the reference subject corresponding to the entire row of elements eliminated; From all reference subjects of each user, the reference subject with the largest rank of the corresponding calculation matrix is ​​selected as the best recommended subject for the corresponding user.

6. The data management method for algorithm recommendation service according to claim 1, characterized in that: Obtain algorithm recommendation service results based on the best recommendation subjects for all users, including: Obtain basic information of each user, including gender, age, and identity attributes; The basic information of each user and the best recommended subject are used as training samples, and the recommendation algorithm adopted by the recommendation algorithm service is optimized and replaced to obtain a new recommendation algorithm for the recommendation algorithm service as the algorithm recommendation service result.

7. A data management method system for algorithm recommendation service, characterized in that: A data management method for executing an algorithm recommendation service as described in any one of claims 1 to 6, comprising: The data processing module is used to obtain all the behavior data of all users on the network platform within a preset time, and to process all the behavior data of all users to obtain all the real behavior data of each user; A calculation module, used to obtain all similar recommended users for each user based on all real behavior data of all users; A construction module is used to obtain a recommendation matrix for each user based on each user and all corresponding similar recommended users; The recommendation optimization module is used to obtain the best recommendation subject for each user based on the recommendation matrix of each user, and obtain the algorithm recommendation service result based on the best recommendation subject of all users.

Images