WO2019237569A1 - 产品推荐方法、装置、计算机设备和存储介质 - Google Patents

产品推荐方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2019237569A1
WO2019237569A1 PCT/CN2018/108033 CN2018108033W WO2019237569A1 WO 2019237569 A1 WO2019237569 A1 WO 2019237569A1 CN 2018108033 W CN2018108033 W CN 2018108033W WO 2019237569 A1 WO2019237569 A1 WO 2019237569A1
Authority
WO
WIPO (PCT)
Prior art keywords
product
user
data
recommendation
matrix
Prior art date
Application number
PCT/CN2018/108033
Other languages
English (en)
French (fr)
Inventor
金戈
徐亮
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2019237569A1 publication Critical patent/WO2019237569A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Definitions

  • the present application relates to the field of computer technology, and in particular, to a product recommendation method, device, computer device, and storage medium.
  • the user recommendation system is to use the user's historical data information to provide users with product information and suggestions to help users decide what products to buy and simulate sales staff to help customers complete the purchase process.
  • Personalized recommendation is to recommend information and products that users are interested in to users according to their characteristics of interest and purchase behavior.
  • most of the existing user recommendation systems use implicit feedback as input. This has the disadvantage of causing information loss.
  • the existing user recommendation systems also face the problem of sparse data. Therefore, it provides a method that can fully use all the information of all users and accurately Predicting the user's interest in different products, and recommending reasonable products for users has become an urgent problem.
  • the main purpose of this application is to provide a product recommendation method, device, computer equipment, and storage medium that make full use of all the information of all users, accurately predict the degree of interest of users in different products, and recommend reasonable products for users.
  • the product recommendation method of the present application includes:
  • the user data Data for all products of a certain user which is data for all users of a certain product
  • the method for generating the first scoring matrix includes:
  • the user data of a certain user is obtained, and the first scoring matrix is constructed by setting corresponding scoring parameters according to a user's preference level for all products in multiple dimensions.
  • step of generating the second scoring matrix includes:
  • the product data of a certain product is obtained, and the corresponding rating parameters are set according to the applicability of the product to all users in multiple dimensions to construct the second rating matrix.
  • first scoring matrix is input to a preset first depth matrix decomposition model for calculation to obtain user characteristics
  • second scoring matrix is input to a preset second depth matrix decomposition model.
  • the step of inputting the user characteristics and product characteristics into a preset formula and calculating the similarity value of the user characteristics and the product characteristics includes:
  • the similarity value of the user feature and the product feature is calculated, where p i is the user feature and q j is the product feature, Is the similarity value.
  • the step of recommending the product to the user includes:
  • the method for separately performing data expansion processing on the first scoring matrix and the second scoring matrix includes:
  • Data expansion is performed by extracting the first and second scoring matrices.
  • the product recommendation device proposed in this application includes:
  • a construction unit configured to obtain user data of a user and product data of a product, and obtain a first scoring matrix of the user on the product according to the user data and a second scoring matrix of the product on the user according to the product data
  • the user data is data of a certain user for all products
  • the product data is data of a certain product for all users
  • a first calculation unit configured to input the first scoring matrix into a preset first depth matrix decomposition model and perform calculation to obtain user characteristics, and input the second scoring matrix into a preset second depth matrix decomposition Product characteristics are calculated in the model;
  • a second calculation unit configured to input the user characteristics and product characteristics into a preset formula and perform calculation to obtain a similarity value between the user characteristics and the product characteristics
  • a judging unit configured to judge whether the similarity value is greater than a preset value
  • An execution unit is configured to recommend the product to the user when the similarity value is greater than a preset value.
  • the computer device includes a memory and a processor, and the memory stores computer-readable instructions, and is characterized in that, when the processor executes the computer-readable instructions, implements the steps of the foregoing method.
  • the computer non-volatile readable storage medium stores computer-readable instructions, and is characterized in that, when the computer-readable instructions are executed by a processor, the steps of the foregoing method are implemented.
  • the similarity value of the user characteristics and the product characteristics obtained according to the product recommendation method in the present application When the similarity value is larger, it indicates that the user is more interested in the product.
  • the value When the value is set, the user is recommended to the user, thereby realizing the full use of the user's multiple information and product characteristic information, accurately predicting the user's interest in the product, recommending a reasonable product for the user, and improving the user experience.
  • FIG. 1 is a schematic diagram of steps of a product recommendation method in an embodiment of the present application
  • FIG. 2 is a schematic diagram of steps of a product recommendation method in another embodiment of the present application.
  • FIG. 3 is a schematic structural diagram of a product recommendation device according to an embodiment of the present application.
  • FIG. 4 is a schematic structural diagram of a product recommendation device in another embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of an execution unit of a product recommendation device according to an embodiment of the present application.
  • FIG. 6 is a schematic block diagram of a computer device according to an embodiment of the present application.
  • the product recommendation method in this embodiment includes:
  • Step S1 obtaining user data of a certain user and product data of a certain product, and obtaining a first rating matrix of the user for the product based on the user data and a second rating matrix of the product for the user based on the product data, so
  • the user data is data of a certain user for all products
  • the product data is data of a certain product for all users
  • Step S2 inputting the first scoring matrix into a preset first depth matrix decomposition model for calculation and obtaining user characteristics, and inputting the second scoring matrix into a preset second depth matrix decomposition model for calculation. Get product characteristics;
  • Step S3 input the user characteristics and product characteristics into a preset formula and perform calculation to obtain a similarity value between the user characteristics and the product characteristics;
  • Step S4 determining whether the similarity value is greater than a preset value
  • Step S5 if yes, recommend the product to the user.
  • step S1 user data of a user and product data of a product are obtained, where the above products are different types of insurance products of an insurance company, and the product data of the above product includes the income situation applicable to the user for all users , Applicable age, applicable occupation and other product characteristics information; user data includes the user's evaluation of all products after purchase, the number of purchased products, the number of times to browse the product, and the time to browse the product and other information.
  • the corresponding rating parameters are constructed to obtain the first score.
  • Matrix and set the corresponding scoring parameters according to the product's applicable income situation, applicable age, and applicability of multiple dimensions of applicable occupations to construct a second scoring matrix.
  • step S2 the above-mentioned first scoring matrix is input into a preset first depth matrix decomposition model and calculated to obtain user features, where the user features are matrices.
  • the preset first depth matrix decomposition model needs to be trained.
  • the preset first depth matrix decomposition model is trained by using a specified amount of the first scoring matrix and the user characteristics corresponding to the first scoring matrix as The sample data is obtained through training.
  • the first depth matrix decomposition model is used to calculate user features.
  • the above-mentioned second scoring matrix is input into a preset second depth matrix decomposition model for calculation to obtain product features, and the above-mentioned product features are also matrices.
  • the preset second depth matrix decomposition model also needs to be trained.
  • the preset second depth matrix decomposition model is trained by passing a specified amount of the second scoring matrix and the product corresponding to the second scoring matrix.
  • Features are obtained by training as sample data.
  • the second depth matrix decomposition model is used to calculate product features
  • step S3 the above-mentioned user characteristics and product characteristics are input into a preset formula for calculation to obtain a calculation result.
  • the preset formula needs to input both the user characteristics and the product characteristics, wherein the preset formula is p i is the above-mentioned user characteristic, q j is the above-mentioned product characteristic, Is the similarity value between the above user characteristics and the above product characteristics, and the calculation result is That is, the similarity value between the user characteristics and the product characteristics.
  • step S4 regarding the obtained similarity values of the above-mentioned user characteristics and the above-mentioned product characteristics, a larger similarity value indicates that the user is more interested in the product. Therefore, whether the product needs to be recommended to the user can be determined according to the magnitude of the similarity value, so a corresponding preset value can be set to determine whether the above-mentioned similarity value is greater than the preset value.
  • step S5 when the above-mentioned similarity value is greater than a preset value, the product is recommended to the user, thereby realizing the full use of the user's various information and product characteristic information, and accurately predicting the user's interest in the product. , Recommend reasonable products for users and improve user experience.
  • the method for generating the first scoring matrix includes:
  • the user data of a certain user is obtained, and the first scoring matrix is constructed by setting corresponding scoring parameters according to a user's preference level for all products in multiple dimensions.
  • the above-mentioned first scoring matrix is generated in a specific manner, including firstly obtaining user data of a certain user, according to multiple dimensions of the user's evaluation of all products after purchase, the number of products purchased, the number of times the product is viewed, and the time of browsing the product
  • the above-mentioned first scoring matrix is constructed by scoring parameters corresponding to the preference setting.
  • the user's evaluation after purchasing a product can be subdivided into different levels of evaluation, such as very like, like, generally like, hate, and very hate.
  • the corresponding rating can be
  • the parameters are set to different scoring parameters such as 5, 4, 3, 2, and 1 in order to set the scoring parameters according to the user's evaluation of a certain product.
  • the corresponding rating parameters For the number of purchases of the product, you can set the corresponding rating parameters to 5, 4, 3, 2 and 1, etc. according to different purchases such as four or more purchases, three purchases, two purchases, one purchase, and no purchases.
  • Different scoring parameters set the scoring parameters according to the quantity of the product purchased.
  • the above methods can be used to set different scoring parameters, which will not be repeated here.
  • the user's rating parameters for a certain product in the above multiple dimensions to obtain a rating vector the user's rating parameters for all products are constructed to obtain the above-mentioned first rating matrix.
  • the method for generating the second scoring matrix includes:
  • the product data of a certain product is obtained, and the corresponding rating parameters are set according to the applicability of the product to all users in multiple dimensions to construct the second rating matrix.
  • the method for generating the above-mentioned second scoring matrix is specific, including first obtaining product data of a certain product, and setting corresponding scoring parameters according to the applicability of the product to all users, such as income, applicable age, and applicable occupations. Construct the above-mentioned second scoring matrix.
  • the sales price of a product is different, and the users of the applicable income situation will be different. For example, the price of a product is high.
  • the corresponding scoring parameters will be different.
  • For users with strong spending power, The product's rating parameter for users is set to 1, and for users with weak consumption capabilities, the product's rating parameter for users is set to 0.
  • the applicable age is 20 to 30 years old.
  • the product's rating parameter for the user is set to 1.
  • the product ’s rating parameter for that user is set to 0.
  • the above methods can be used to set different scoring parameters, which will not be repeated here. For the above-mentioned multiple dimensions, setting the product's rating parameters for a certain user to obtain a rating vector, and constructing the product's rating parameters for all users to obtain the above-mentioned second rating matrix.
  • a product recommendation method in another embodiment the inputting the first scoring matrix into a preset first depth matrix decomposition model for calculation to obtain user characteristics, and inputting the second scoring matrix Before step S2, where calculation is performed to obtain product characteristics in a preset second depth matrix decomposition model, includes:
  • Step 201 Perform data expansion processing on the first and second scoring matrices, respectively.
  • calculation is performed by inputting the first scoring matrix into a preset first depth matrix decomposition model and inputting the second scoring matrix into a preset second depth matrix decomposition.
  • data expansion processing can be performed on the first and second scoring matrices to expand the size of the matrix and increase the size of the training set. The purpose is to avoid overfitting and improve the first depth matrix decomposition.
  • the accuracy of the model and the prediction of the second depth matrix factorization model For example, for the first scoring matrix, a specific method for performing data expansion processing is to extract a part of the matrix from the first scoring matrix as an expanded first expansion matrix, and then input the first expansion matrix and the first scoring matrix to a first depth together.
  • the calculation is performed in the matrix decomposition model; the method of performing data expansion processing on the second scoring matrix is the same as the above method, and is not repeated here.
  • the step S3 of inputting the user characteristics and product characteristics into a preset formula and calculating a similarity value of the user characteristics and the product characteristics includes:
  • the similarity value of the user feature and the product feature is calculated, where p i is the user feature and q j is the product feature, Is the similarity value.
  • the similarity values of the above-mentioned user characteristics and the above-mentioned product characteristics need to be calculated according to a preset formula.
  • the above-mentioned preset formula is Where p i is the above-mentioned user characteristics and q j is the above-mentioned product characteristics, It is the similarity value of the above-mentioned user characteristics and the above-mentioned product characteristics, and the above-mentioned similarity value is used to indicate the degree of interest of the user to the product.
  • the step S5 of recommending the product to the user includes:
  • Step S51 match the similarity value with a preset recommendation level table, where the recommendation level table includes a correspondence relationship between different similarity value ranges and recommendation levels;
  • Step S52 output a recommendation level according to the matching result
  • Step S53 Recommend the product to the user according to the recommendation level.
  • the similarity value of the user's interest in the product can also be matched with a preset recommendation level table.
  • the above recommendation level table includes different similarity value ranges and recommendation levels. Corresponding relationship, where The larger the value, the higher the degree of interest of the user in the product; the corresponding recommendation level is output according to the matching result, wherein the recommendation can be different recommendation levels such as strong recommendation and recommendation, and the user is recommended to the user according to the above recommendation level.
  • the product recommendation device in this embodiment includes:
  • the construction unit 10 is configured to obtain user data of a certain user and product data of a certain product, and obtain a first rating matrix of the user based on the user data and a second rating of the user based on the product data.
  • Matrix the user data is data of a certain user for all products
  • the product data is data of a certain product for all users;
  • a first calculation unit 20 is configured to input the first scoring matrix into a preset first depth matrix decomposition model and perform calculation to obtain user characteristics, and input the second scoring matrix to a preset second depth matrix. Product characteristics are calculated in the decomposition model;
  • a second calculation unit 30 configured to input the user characteristics and product characteristics into a preset formula and perform calculation to obtain a similarity value between the user characteristics and the product characteristics;
  • a judging unit 40 configured to judge whether the similarity value is greater than a preset value
  • the execution unit 50 is configured to recommend the product to the user when the similarity value is greater than a preset value.
  • Obtain user data of a user and product data of a product where the above products are different types of insurance products of an insurance company, and the product data of the above product includes the income situation, applicable age and applicable age of the product for all users Occupational and other product characteristics information; user data includes the user ’s evaluation of all products after purchase, the number of products purchased, the number of times the product was viewed, and the time spent browsing the product.
  • the constructing unit 10 constructs the corresponding rating parameters according to the user's preference for all products after product purchase, the number of purchased products, the number of times the product is browsed, and the degree of preference for multiple dimensions of the time of product browsing.
  • the first scoring matrix is constructed by setting corresponding scoring parameters according to the product's applicable income situation, applicable age, and applicability of multiple dimensions of applicable occupations to all users.
  • the first calculation unit 20 inputs the first scoring matrix into a preset first depth matrix decomposition model and performs calculation to obtain user features, where the user features are matrices.
  • the preset first depth matrix decomposition model needs to be trained.
  • the preset first depth matrix decomposition model is trained by using a specified amount of the first scoring matrix and the user characteristics corresponding to the first scoring matrix as The sample data is obtained through training.
  • the first depth matrix decomposition model is used to calculate user features.
  • the above-mentioned second scoring matrix is input into a preset second depth matrix decomposition model for calculation to obtain product features, and the above-mentioned product features are also matrices.
  • the preset second depth matrix factorization model also needs to be trained.
  • the preset second depth matrix factorization model is trained by passing a specified amount of the second rating matrix and the product corresponding to the first person rating matrix. Features are obtained by training as sample data.
  • the second depth matrix decomposition model is used to calculate product features.
  • the second calculation unit 30 inputs the above-mentioned user characteristics and product characteristics into a preset formula and performs calculation to obtain a calculation result.
  • the preset formula needs to input both the user characteristics and the product characteristics.
  • p i is the above-mentioned user characteristic
  • q j is the above-mentioned product characteristic
  • the calculation result is That is, the similarity value between the user characteristics and the product characteristics.
  • a larger similarity value indicates that the user is more interested in the product. Therefore, whether the product needs to be recommended to the user can be determined according to the magnitude of the similarity value, so a corresponding preset value can be set, and the determination unit 40 determines whether the above-mentioned similarity value is greater than the preset value.
  • the execution unit 50 recommends the product to the user, thereby realizing the full use of the user's various information and product characteristic information, and accurately predicting the user's degree of interest in the product. Users recommend reasonable products to improve user experience.
  • the method for generating the first scoring matrix includes:
  • the user data of a certain user is obtained, and the first scoring matrix is constructed by setting corresponding scoring parameters according to a user's preference level for all products in multiple dimensions.
  • the above-mentioned first scoring matrix is generated in a specific manner, including firstly obtaining user data of a certain user, according to multiple dimensions of the user's evaluation of all products after purchase, the number of products purchased, the number of times the product is viewed, and the time of browsing the product.
  • the above-mentioned first scoring matrix is constructed by scoring parameters corresponding to the preference setting.
  • the user's evaluation after purchasing a product can be subdivided into different levels of evaluation, such as very like, like, generally like, hate, and very hate.
  • the corresponding rating can be The parameters are set to different scoring parameters such as 5, 4, 3, 2, and 1 in order to set the scoring parameters according to the user's evaluation of a certain product.
  • the corresponding rating parameters For the number of purchases of the product, you can set the corresponding rating parameters to 5, 4, 3, 2 and 1, etc. according to different purchases such as four or more purchases, three purchases, two purchases, one purchase, and no purchases.
  • Different scoring parameters set the scoring parameters according to the quantity of the product purchased.
  • the above methods can be used to set different scoring parameters, which will not be repeated here.
  • the user's rating parameters for a certain product in the above multiple dimensions to obtain a rating vector the user's rating parameters for all products are constructed to obtain the above-mentioned first rating matrix.
  • the method for generating the second scoring matrix includes:
  • the product data of a certain product is obtained, and the corresponding rating parameters are set according to the preference degree of the product to all users in multiple dimensions to construct the second rating matrix.
  • the method for generating the above-mentioned second scoring matrix is specific, including first obtaining product data of a certain product, and setting corresponding scoring parameters according to the applicability of the product to all users, such as income, applicable age, and applicable occupations. Construct the above-mentioned second scoring matrix.
  • the sales price of a product is different, and the users of the applicable income situation will be different. For example, the price of a product is high.
  • the corresponding scoring parameters will be different.
  • For users with strong spending power, The product's rating parameter for users is set to 1, and for users with weak consumption capabilities, the product's rating parameter for users is set to 0.
  • the applicable age is 20 to 30 years old.
  • the product's rating parameter for the user is set to 1.
  • the product ’s rating parameter for that user is set to 0.
  • the above methods can be used to set different scoring parameters, which will not be repeated here. For the above-mentioned multiple dimensions, setting the product's rating parameters for a certain user to obtain a rating vector, and constructing the product's rating parameters for all users to obtain the above-mentioned second rating matrix.
  • the product recommendation device in another embodiment further includes:
  • the expansion unit 201 is configured to perform data expansion processing on the first and second scoring matrices, respectively.
  • the expansion unit 201 performs data expansion processing on the first and second scoring matrices to expand the size of the matrix and increase the size of the training set. The purpose is to avoid overfitting and increase the first depth.
  • the accuracy of the matrix factorization model and the prediction of the second depth matrix factorization model For example, for the first scoring matrix, a specific method for performing data expansion processing is to extract a part of the matrix from the first scoring matrix as an expanded first expansion matrix, and then input the first expansion matrix and the first scoring matrix to a first depth together.
  • the calculation is performed in the matrix decomposition model; the method of performing data expansion processing on the second scoring matrix is the same as the above method, and is not repeated here.
  • the second calculation unit 30 is specifically configured to be based on a formula The similarity value of the user feature and the product feature is calculated, where p i is the user feature and q j is the product feature, Is the similarity value.
  • the similarity values of the above-mentioned user characteristics and the above-mentioned product characteristics need to be calculated according to a preset formula.
  • the above-mentioned preset formula is Where p i is the above-mentioned user characteristics and q j is the above-mentioned product characteristics, It is the similarity value of the above-mentioned user characteristics and the above-mentioned product characteristics, and the above-mentioned similarity value is used to indicate the degree of interest of the user to the product.
  • the execution unit 50 includes:
  • a first matching module 51 configured to match the similarity value with a preset recommendation level table, where the recommendation level table includes a correspondence relationship between different similarity value ranges and recommendation levels;
  • a second matching module 52 configured to output a recommendation level according to the matching result
  • the recommendation module 53 is configured to recommend the product to the user according to the recommendation level.
  • the first matching module 51 may also match the similarity values that the user is interested in the product with a preset recommendation level table, and the above recommendation level table includes different similarities Correspondence between value range and recommendation level, where The larger the value, the higher the degree of interest of the user in the product; the second matching module 52 outputs the corresponding recommendation level according to the matching result, wherein the recommendation can be different recommendation levels such as strong recommendation and recommendation, and the recommendation module 53 is based on The above recommendation level recommends the product to the user, and realizes a more humane recommendation of a reasonable product for the user.
  • an embodiment of the present application further provides a computer device.
  • the computer device may be a server, and its internal structure may be as shown in FIG.
  • the computer device includes a processor, a memory, a network interface, and a database connected through a system bus.
  • the computer design processor is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system, computer-readable instructions, and a database.
  • the memory provides an environment for operating systems and computer-readable instructions in a non-volatile storage medium.
  • the database of the computer equipment is used to preset data such as product recommendation methods.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer-readable instructions are executed by a processor to implement a product recommendation method.
  • the above processor executes the steps of the above product recommendation method: obtaining user data of a certain user and product data of a certain product, and obtaining a first rating matrix of the user for the product based on the above user data, and obtaining a product to the user based on the above product data.
  • the second scoring matrix the user data is data of a user for all products, and the product data is the data of a product for all users;
  • the first scoring matrix is input into a preset first depth matrix decomposition model.
  • the user characteristics are calculated, and the second scoring matrix is input to a preset second depth matrix decomposition model to calculate the product characteristics.
  • the user characteristics and product characteristics are input to a preset formula to calculate the user characteristics. Similarity value with the above product characteristics; determine whether the above similarity value is greater than a preset value; if so, recommend the product to the user.
  • the method includes: obtaining user data of a certain user, and constructing the first scoring matrix according to a user's preference for all products in multiple dimensions by setting corresponding scoring parameters.
  • the method for generating the above-mentioned second scoring matrix includes: obtaining product data of a certain product, and setting the corresponding scoring parameters according to the application degree of the product to all users in multiple dimensions to construct the second scoring matrix.
  • the above-mentioned first scoring matrix is input into a preset first depth matrix decomposition model for calculation to obtain user characteristics
  • the above-mentioned second scoring matrix is input into a preset second depth matrix decomposition model.
  • the method includes: performing data expansion processing on the first scoring matrix and the second scoring matrix, respectively.
  • the step of inputting the user characteristics and product characteristics into a preset formula and calculating the similarity value of the user characteristics and the product characteristics includes: according to the formula The similarity values of the above-mentioned user characteristics and the above-mentioned product characteristics are calculated, where p i is the above-mentioned user characteristic and q j is the above-mentioned product characteristic, Is the similarity value.
  • the step of recommending the product to the user includes: matching the similarity value with a preset recommendation level table, and the recommendation level table includes a corresponding relationship between different similarity value ranges and recommendation levels; A recommendation level is output according to the matching result; the product is recommended to the user according to the above recommendation level.
  • the method for performing data expansion processing on the first and second scoring matrices includes: performing data expansion in a manner of extracting a partial matrix from the first and second scoring matrices.
  • FIG. 6 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied.
  • An embodiment of the present application further provides a computer non-volatile readable storage medium, which stores computer-readable instructions.
  • a product recommendation method is implemented, specifically: obtaining a certain The user's user data and the product data of a product, and the user's first rating matrix for the product and the product's second rating matrix for the user based on the user data.
  • Product data is data for all users of a certain product; input the first scoring matrix into a preset first depth matrix decomposition model to calculate user characteristics, and input the second scoring matrix to Product characteristics are calculated in a preset second depth matrix decomposition model; the user characteristics and product characteristics are input into a preset formula and calculated to obtain similarity values between the user characteristics and the product characteristics; and the similarity values are judged Whether it is greater than the preset value; if so, recommend the product to the user.
  • the computer non-volatile readable storage medium includes the method for generating the first scoring matrix, including: obtaining user data of a user, and constructing the corresponding scoring parameters according to a user's preference level for all products in multiple dimensions.
  • the first scoring matrix described above.
  • the method for generating the above-mentioned second scoring matrix includes: obtaining product data of a certain product, and setting the corresponding scoring parameters according to the application degree of the product to all users in multiple dimensions to construct the second scoring matrix.
  • the above-mentioned first scoring matrix is input into a preset first depth matrix decomposition model for calculation to obtain user characteristics
  • the above-mentioned second scoring matrix is input into a preset second depth matrix decomposition model.
  • the method includes: performing data expansion processing on the first scoring matrix and the second scoring matrix, respectively.
  • the step of inputting the user characteristics and product characteristics into a preset formula and calculating the similarity value of the user characteristics and the product characteristics includes: according to the formula The similarity values of the above-mentioned user characteristics and the above-mentioned product characteristics are calculated, where p i is the above-mentioned user characteristic and q j is the above-mentioned product characteristic, Is the similarity value.
  • the step of recommending the product to the user includes: matching the similarity value with a preset recommendation level table, and the recommendation level table includes a corresponding relationship between different similarity value ranges and recommendation levels; A recommendation level is output according to the matching result; the product is recommended to the user according to the above recommendation level.
  • the method for performing data expansion processing on the first and second scoring matrices includes: performing data expansion in a manner of extracting a partial matrix from the first and second scoring matrices.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory can include random access memory (RAM) or external cache memory.
  • RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
  • the similarity values of the above-mentioned user characteristics and the above-mentioned product characteristics obtained according to the product recommendation method in this application.
  • the similarity value is larger, the user is more interested in the product.
  • the similarity value is When the value is greater than the preset value, the product is recommended to the user, thereby realizing the full use of the user's various information and product characteristic information, accurately predicting the user's interest in the product, recommending a reasonable product for the user, and improving the user Experience.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种产品推荐方法、装置、计算机设备和存储介质，该方法包括：获取用户数据和产品数据，并根据用户数据得到用户对产品的第一评分矩阵以及根据产品数据得到产品对用户的第二评分矩阵（S1）；将第一评分矩阵输入到预设的第一深度矩阵分解模型中计算得到用户特征，将第二评分矩阵输入到预设的第二深度矩阵分解模型中计算得到产品特征（S2）；将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和产品特征的相似度值（S3）；判断所述相似度值是否大于预设值（S4）；若是，则向该用户推荐该产品（S5）。

Description

产品推荐方法、装置、计算机设备和存储介质
本申请要求于2018年6月13日提交中国专利局、申请号为2018106078629，申请名称为“产品推荐方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及到计算机技术领域，特别是涉及到一种产品推荐方法、装置、计算机设备和存储介质。
背景技术
随着社会的快速发展，用户推荐系统是利用用户的历史数据信息向用户提供商品信息和建议，帮助用户决定应该购买什么产品，模拟销售人员帮助客户完成购买过程。个性化推荐是根据用户的兴趣特点和购买行为，向用户推荐用户感兴趣的信息和商品。但是现有的用户推荐系统大多采用隐式反馈作为输入，这样的缺点在于会造成信息丢失；现有的用户推荐系统还面临数据稀疏问题，因此提供一种能充分利用所有用户所有信息，准确地预测出用户对不同产品的感兴趣程度，为用户推荐合理的产品成为亟待解决的问题。
技术问题
本申请的主要目的为提供一种产品推荐方法、装置、计算机设备和存储介质，充分利用所有用户所有信息，准确地预测出用户对不同产品的感兴趣程度，为用户推荐合理的产品。
技术解决方案
为了实现上述本申请的目的，本申请的产品推荐方法，包括：
获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产 品特征的相似度值；
判断所述相似度值是否大于预设值；
若是，则向该用户推荐该产品。
进一步地，所述第一评分矩阵的生成方法，包括：
获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到所述第一评分矩阵。
进一步地，所述第二评分矩阵的生成方法骤，包括：
获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到所述第二评分矩阵。
进一步地，所述将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征的步骤之前，包括：
对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
进一步地，所述将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值的步骤，包括：
根据公式
Figure PCTCN2018108033-appb-000001
计算得到所述用户特征和所述产品特征的相似度值，其中p i为所述用户特征，q j为所述产品特征，
Figure PCTCN2018108033-appb-000002
为所述相似度值。
进一步地，所述向该用户推荐该产品的步骤，包括：
将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级表包括不同相似度值范围与推荐等级的对应关系；
根据匹配结果输出推荐等级；
根据所述推荐等级向该用户推荐该产品。
进一步地，所述对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理的方法，包括：
以抽取所述第一评分矩阵以及第二评分矩阵中部分矩阵的方式进行数据扩充。
本申请提出的产品推荐装置，包括：
构造单元，用于获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
第一计算单元，用于将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
第二计算单元，用于将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
判断单元，用于判断所述相似度值是否大于预设值；
执行单元，用于当所述相似度值大于预设值时，则向该用户推荐该产品。
本申请提出的计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现上述方法的步骤。
本申请提出的计算机非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现上述方法的步骤。
有益效果
根据本申请中的产品推荐方法得到的所述用户特征和所述产品特征的相似度值，当相似度值的越大，表示该用户对该产品越感兴趣，当所述相似度值大于预设值时，则向该用户推荐该产品，从而实现了充分利用用户的多种信息和产品特性信息，准确地预测出用户对产品的感兴趣程度，为用户推荐合理的产品，提高用户体验。
附图说明
图1为本申请一实施例中的产品推荐方法的步骤示意图；
图2为本申请另一实施例中的产品推荐方法的步骤示意图；
图3为本申请一实施例中的产品推荐装置的结构示意图；
图4为本申请另一实施例中的产品推荐装置的结构示意图；
图5为本申请一实施例中的产品推荐装置的执行单元的结构示意图；
图6为本申请一实施例的计算机设备的结构示意框图。
本申请目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
本发明的最佳实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
参照图1，本实施例中的产品推荐方法，包括：
步骤S1，获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到 用户对产品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
步骤S2，将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
步骤S3，将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
步骤S4，判断所述相似度值是否大于预设值；
步骤S5，若是，则向该用户推荐该产品。
在步骤S1中，获取某个用户的用户数据和某个产品的产品数据，其中上述产品为保险公司不同类型的保险产品，上述某个产品的产品数据包括该产品对所有用户所适用的收入情况、适用年龄以及适用职业等多种产品特性信息；用户数据包括该用户对所有产品在购买之后的评价、购买产品的数量、浏览产品的次数以及浏览产品的时间等多种信息。对于上述用户数据和产品数据，根据用户对所有产品在产品购买之后的评价、购买产品的数量、浏览产品的次数以及浏览产品的时间多个维度的偏好程度设置对应的评分参数构造得到第一评分矩阵，并根据产品对所有用户所适用的收入情况、适用年龄以及适用职业多个维度的适用程度设置对应的评分参数构造得到第二评分矩阵。
在步骤S2中，将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，其中上述用户特征为矩阵。上述预设的第一深度矩阵分解模型需要进行训练，上述预设的第一深度矩阵分解模型进行训练的方式为通过指定量的第一评分矩阵，以及上述第一评分矩阵所对应的用户特征作为样本数据进行训练所得，上述第一深度矩阵分解模型用于计算用户特征。将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征，其中上述产品特征也为矩阵。上述预设的第二深度矩阵分解模型也需要进行训练，上述预设的第二深度矩阵分解模型进行训练的方式为通过指定量的第二评分矩阵，以及在上述第二评分矩阵所对应的产品特征作为样本数据进行训练所得，上述第二深度矩阵分解模型用于计算产品特征。
在步骤S3中，将上述用户特征和产品特征输入到预设的公式中进行计算得到计算结果，预设的公式需要同时输入用户特征和产品特征，其中上述预设的公式为
Figure PCTCN2018108033-appb-000003
p i为上述用户特征，q j为上述产品特征，
Figure PCTCN2018108033-appb-000004
为上述用户特征和上述产品特征的相似度值，计算结果为
Figure PCTCN2018108033-appb-000005
即上述用户特征和上述产品特征的相似度值。
在步骤S4中，对于得到的上述用户特征和上述产品特征的相似度值，当相似度值的越大，表示该用户对该产品越感兴趣。因此可以根据相似度值的大小判断是否需要向该用户推荐该产品，因此可以设置相应的预设值，判断上述相似度值是否大于预设值。
在步骤S5中，当上述相似度值大于预设值时，则向该用户推荐该产品，从而实现了充分利用用户的多种信息和产品特性信息，准确地预测出用户对产品的感兴趣程度，为用户推荐合理的产品，提高用户体验。
本实施例中的产品推荐方法，所述第一评分矩阵的生成方法，包括：
获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到所述第一评分矩阵。
上述第一评分矩阵的生成方式具体，包括将先获取某个用户的用户数据，根据用户所有产品在购买之后的评价、购买产品的数量、浏览产品的次数以及浏览产品的时间等多个维度的偏好程度设置对应的评分参数构造得到上述第一评分矩阵。例如，用户在将某个产品购买之后的评价可以细分为非常喜欢、喜欢、一般喜欢、讨厌以及非常讨厌等不同等级的评价，对于用户对某个产品不同等级的评价，可以将对应的评分参数依次设置为5、4、3、2以及1等不同评分参数，实现根据用户对某个产品的评价设置评分参数。对于购买该产品的数量，可以根据购买四次以上、购买三次、购买二次、购买一次以及未购买等不同的购买次数，将对应的评分参数依次设置为5、4、3、2以及1等不同评分参数，实现根据购买该产品的数量设置评分参数。同理，对于浏览该产品的次数以及浏览该产品的时间等信息，均可以采用上述方法来设置不同的评分参数，在此不再赘述。对于在上述多个维度下设置用户对某个产品的评分参数得到评分向量，将用户对所有产品的评分参数构造得到上述第一评分矩阵。
本实施例中的产品推荐方法，所述第二评分矩阵的生成方法，包括：
获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到所述第二评分矩阵。
上述第二评分矩阵的生成方式具体，包括将先获取某个产品的产品数据，根据该产品对所有用户所适用的收入情况、适用年龄以及适用职业等多个维度的适用程度设置对应的评分参数构造得到上述第二评分矩阵。某个产品的销售价格不同，其所适用的收入情况的用户将不同，例如某个产品的价格很高，对于不同消费水平的用户，对应的评分参数将不同，对于消费能力强的用户，将产品对用户的评分参数设置为1，对于消费能力弱的用户，将产品对用户的评分参数设置为0。对于某个产品的适用年龄为20岁到30岁，对于处于20岁到30岁年龄段的用户，则将产品对该用户的评分参数设置为1，对于不处于20岁到30岁年 龄段的用户，则将产品对该用户的评分参数设置为0。同理，对于适用职业等信息，均可以采用上述方法来设置不同的评分参数，在此不再赘述。对于在上述多个维度下设置产品对某个用户的评分参数得到评分向量，将产品对所有用户的评分参数构造得到上述第二评分矩阵。
参照图2，另一实施例中的产品推荐方法，所述将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征的步骤S2之前，包括：
步骤201，对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
对于上述第一评分矩阵以及第二评分矩阵，在将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算以及将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算之前，可以对上述第一评分矩阵以及第二评分矩阵分别进行数据扩充处理，扩充矩阵大小，增加训练集大小，其目的在于避免过拟合的发生，从而提高第一深度矩阵分解模型以及第二深度矩阵分解模型预测的准确性。例如，对于第一评分矩阵，进行数据扩充处理的具体方式为在第一评分矩阵中抽取部分矩阵作为扩充的第一扩充矩阵，再将第一扩充矩阵与第一评分矩阵一起输入到第一深度矩阵分解模型中进行计算；对第二评分矩阵进行数据扩充处理的方法和上述方法相同，在此不再赘述。
本实施例中的产品推荐方法，所述将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值的步骤S3，包括：
根据公式
Figure PCTCN2018108033-appb-000006
计算得到所述用户特征和所述产品特征的相似度值，其中p i为所述用户特征，q j为所述产品特征，
Figure PCTCN2018108033-appb-000007
为所述相似度值。
对于输入的用户特征和产品特征，需要根据预设的公式进行计算得到上述用户特征和上述产品特征的相似度值，上述预设的公式为
Figure PCTCN2018108033-appb-000008
其中p i为上述用户特征，q j为上述产品特征，
Figure PCTCN2018108033-appb-000009
为上述用户特征和上述产品特征的相似度值，上述相似度值用以表示用户对产品感兴趣的程度，当相似度值越大，说明用户对产品感兴趣的程度越高。
本实施例中的产品推荐方法，所述向该用户推荐该产品的步骤S5，包括：
步骤S51，将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级表包括不同相似度值范围与推荐等级的对应关系；
步骤S52，根据匹配结果输出推荐等级；
步骤S53，根据所述推荐等级向该用户推荐该产品。
对于输出的上述用户特征和上述产品特征的相似度值，还可以将用户对产品感兴趣的 相似度值与预设的推荐等级表进行匹配，上述推荐等级表包括不同相似度值范围与推荐等级的对应关系，其中
Figure PCTCN2018108033-appb-000010
的值越大，表示用户对该产品的感兴趣的程度越高；根据匹配结果输出对应的推荐等级，其中推荐可以为强烈推荐和推荐等不同的推荐等级，根据上述推荐等级向该用户推荐该产品，实现更人性化的为用户推荐合理的产品。
参照图3，本实施例中的产品推荐装置，包括：
构造单元10，用于获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
第一计算单元20，用于将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
第二计算单元30，用于将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
判断单元40，用于判断所述相似度值是否大于预设值；
执行单元50，用于当所述相似度值大于预设值时，则向该用户推荐该产品。
获取某个用户的用户数据和某个产品的产品数据，其中上述产品为保险公司不同类型的保险产品，上述某个产品的产品数据包括该产品对所有用户所适用的收入情况、适用年龄以及适用职业等多种产品特性信息；用户数据包括该用户对所有产品在购买之后的评价、购买产品的数量、浏览产品的次数以及浏览产品的时间等多种信息。对于上述用户数据和产品数据，构造单元10根据用户对所有产品在产品购买之后的评价、购买产品的数量、浏览产品的次数以及浏览产品的时间多个维度的偏好程度设置对应的评分参数构造得到第一评分矩阵，并根据产品对所有用户所适用的收入情况、适用年龄以及适用职业多个维度的适用程度设置对应的评分参数构造得到第二评分矩阵。
第一计算单元20将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，其中上述用户特征为矩阵。上述预设的第一深度矩阵分解模型需要进行训练，上述预设的第一深度矩阵分解模型进行训练的方式为通过指定量的第一评分矩阵，以及上述第一评分矩阵所对应的用户特征作为样本数据进行训练所得，上述第一深度矩阵分解模型用于计算用户特征。将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征，其中上述产品特征也为矩阵。上述预设的第二深度矩阵分解模型也需要进行训练，上述预设的第二深度矩阵分解模型进行训练的方式为通过指定量的第二评 分矩阵，以及在上述第人评分矩阵所对应的产品特征作为样本数据进行训练所得，上述第二深度矩阵分解模型用于计算产品特征。
第二计算单元30将上述用户特征和产品特征输入到预设的公式中进行计算得到计算结果，预设的公式需要同时输入用户特征和产品特征，其中上述预设的公式为
Figure PCTCN2018108033-appb-000011
p i为上述用户特征，q j为上述产品特征，
Figure PCTCN2018108033-appb-000012
为上述用户特征和上述产品特征的相似度值，计算结果为
Figure PCTCN2018108033-appb-000013
即上述用户特征和上述产品特征的相似度值。
对于得到的上述用户特征和上述产品特征的相似度值，当相似度值的越大，表示该用户对该产品越感兴趣。因此可以根据相似度值的大小判断是否需要向该用户推荐该产品，因此可以设置相应的预设值，判断单元40判断上述相似度值是否大于预设值。
当上述相似度值大于预设值时，执行单元50则向该用户推荐该产品，从而实现了充分利用用户的多种信息和产品特性信息，准确地预测出用户对产品的感兴趣程度，为用户推荐合理的产品，提高用户体验。
本实施例中的产品推荐装置，所述第一评分矩阵的生成方法，包括：
获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到所述第一评分矩阵。
上述第一评分矩阵的生成方式具体，包括将先获取某个用户的用户数据，根据用户所有产品在购买之后的评价、购买产品的数量、浏览产品的次数以及浏览产品的时间等多个维度的偏好程度设置对应的评分参数构造得到上述第一评分矩阵。例如，用户在将某个产品购买之后的评价可以细分为非常喜欢、喜欢、一般喜欢、讨厌以及非常讨厌等不同等级的评价，对于用户对某个产品不同等级的评价，可以将对应的评分参数依次设置为5、4、3、2以及1等不同评分参数，实现根据用户对某个产品的评价设置评分参数。对于购买该产品的数量，可以根据购买四次以上、购买三次、购买二次、购买一次以及未购买等不同的购买次数，将对应的评分参数依次设置为5、4、3、2以及1等不同评分参数，实现根据购买该产品的数量设置评分参数。同理，对于浏览该产品的次数以及浏览该产品的时间等信息，均可以采用上述方法来设置不同的评分参数，在此不再赘述。对于在上述多个维度下设置用户对某个产品的评分参数得到评分向量，将用户对所有产品的评分参数构造得到上述第一评分矩阵。
本实施例中的产品推荐装置，所述第二评分矩阵的生成方法，包括：
获取某个产品的产品数据，根据产品对所有用户在多个维度的偏好程度设置对应的评分参数构造得到所述第二评分矩阵。
上述第二评分矩阵的生成方式具体，包括将先获取某个产品的产品数据，根据该产品对所有用户所适用的收入情况、适用年龄以及适用职业等多个维度的适用程度设置对应的评分参数构造得到上述第二评分矩阵。某个产品的销售价格不同，其所适用的收入情况的用户将不同，例如某个产品的价格很高，对于不同消费水平的用户，对应的评分参数将不同，对于消费能力强的用户，将产品对用户的评分参数设置为1，对于消费能力弱的用户，将产品对用户的评分参数设置为0。对于某个产品的适用年龄为20岁到30岁，对于处于20岁到30岁年龄段的用户，则将产品对该用户的评分参数设置为1，对于不处于20岁到30岁年龄段的用户，则将产品对该用户的评分参数设置为0。同理，对于适用职业等信息，均可以采用上述方法来设置不同的评分参数，在此不再赘述。对于在上述多个维度下设置产品对某个用户的评分参数得到评分向量，将产品对所有用户的评分参数构造得到上述第二评分矩阵。
参照图4，另一实施例中的产品推荐装置，还包括：
扩充单元201，用于对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
对于上述第一评分矩阵以及第二评分矩阵，在将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算以及将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算之前，扩充单元201对上述第一评分矩阵以及第二评分矩阵分别进行数据扩充处理，扩充矩阵大小，增加训练集大小，其目的在于避免过拟合的发生，从而提高第一深度矩阵分解模型以及第二深度矩阵分解模型预测的准确性。例如，对于第一评分矩阵，进行数据扩充处理的具体方式为在第一评分矩阵中抽取部分矩阵作为扩充的第一扩充矩阵，再将第一扩充矩阵与第一评分矩阵一起输入到第一深度矩阵分解模型中进行计算；对第二评分矩阵进行数据扩充处理的方法和上述方法相同，在此不再赘述。
本实施例中的产品推荐装置，所述第二计算单元30具体用于根据公式
Figure PCTCN2018108033-appb-000014
计算得到所述用户特征和所述产品特征的相似度值，其中p i为所述用户特征，q j为所述产品特征，
Figure PCTCN2018108033-appb-000015
为所述相似度值。
对于输入的用户特征和产品特征，需要根据预设的公式进行计算得到上述用户特征和上述产品特征的相似度值，上述预设的公式为
Figure PCTCN2018108033-appb-000016
其中p i为上述用户特征，q j为上述产品特征， 为上述用户特征和上述产品特征的相似度值，上述相似度值用以表示用户对产品感兴趣的程度，当相似度值越大，说明用户对产品感兴趣的程度越高。
参照图5，本实施例中的产品推荐装置，所述执行单元50包括：
第一匹配模块51，用于将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级 表包括不同相似度值范围与推荐等级的对应关系；
第二匹配模块52，用于根据匹配结果输出推荐等级；
推荐模块53，用于根据所述推荐等级向该用户推荐该产品。
对于输出的上述用户特征和上述产品特征的相似度值，第一匹配模块51还可以将用户对产品感兴趣的相似度值与预设的推荐等级表进行匹配，上述推荐等级表包括不同相似度值范围与推荐等级的对应关系，其中
Figure PCTCN2018108033-appb-000018
的值越大，表示用户对该产品的感兴趣的程度越高；第二匹配模块52根据匹配结果输出对应的推荐等级，其中推荐可以为强烈推荐和推荐等不同的推荐等级，推荐模块53根据上述推荐等级向该用户推荐该产品，实现更人性化的为用户推荐合理的产品。
参照图6，本申请实施例中还提供一种计算机设备，该计算机设备可以是服务器，其内部结构可以如图6所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于预设产品推荐方法等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令被处理器执行时以实现产品推荐方法。
上述处理器执行上述产品推荐方法的步骤：获取某个用户的用户数据和某个产品的产品数据，并根据上述用户数据得到用户对产品的第一评分矩阵以及根据上述产品数据得到产品对用户的第二评分矩阵，上述用户数据为某个用户对所有产品的数据，上述产品数据为某个产品对所有用户的数据；将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；将上述用户特征和产品特征输入到预设的公式中进行计算得到上述用户特征和上述产品特征的相似度值；判断上述相似度值是否大于预设值；若是，则向该用户推荐该产品。
上述计算机设备，上述第一评分矩阵的生成方法，包括：获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到上述第一评分矩阵。
在一个实施例中，上述第二评分矩阵的生成方法，包括：获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到上述第二评分矩阵。
在一个实施例中，上述将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征的步骤之前，包括：对上述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
在一个实施例中，上述将上述用户特征和产品特征输入到预设的公式中进行计算得到上述用户特征和上述产品特征的相似度值的步骤，包括：根据公式
Figure PCTCN2018108033-appb-000019
计算得到上述用户特征和上述产品特征的相似度值，其中p i为上述用户特征，q j为上述产品特征，
Figure PCTCN2018108033-appb-000020
为上述相似度值。
在一个实施例中，上述向该用户推荐该产品的步骤，包括：将上述相似度值与预设的推荐等级表进行匹配，上述推荐等级表包括不同相似度值范围与推荐等级的对应关系；根据匹配结果输出推荐等级；根据上述推荐等级向该用户推荐该产品。
在一个实施例中，上述对上述第一评分矩阵和第二评分矩阵分别进行数据扩充处理的方法，包括：以抽取上述第一评分矩阵以及第二评分矩阵中部分矩阵的方式进行数据扩充。
本领域技术人员可以理解，图6中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定。
本申请一实施例还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，计算机可读指令被处理器执行时实现一种产品推荐方法，具体为：获取某个用户的用户数据和某个产品的产品数据，并根据上述用户数据得到用户对产品的第一评分矩阵以及根据上述产品数据得到产品对用户的第二评分矩阵，上述用户数据为某个用户对所有产品的数据，上述产品数据为某个产品对所有用户的数据；将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；将上述用户特征和产品特征输入到预设的公式中进行计算得到上述用户特征和上述产品特征的相似度值；判断上述相似度值是否大于预设值；若是，则向该用户推荐该产品。
上述计算机非易失性可读存储介质，将上述第一评分矩阵的生成方法，包括：获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到上述第一评分矩阵。
在一个实施例中，上述第二评分矩阵的生成方法，包括：获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到上述第二评分矩阵。
在一个实施例中，上述将上述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将上述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征的步骤之前，包括：对上述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
在一个实施例中，上述将上述用户特征和产品特征输入到预设的公式中进行计算得到上述用户特征和上述产品特征的相似度值的步骤，包括：根据公式
Figure PCTCN2018108033-appb-000021
计算得到上述用户特征和上述产品特征的相似度值，其中p i为上述用户特征，q j为上述产品特征，
Figure PCTCN2018108033-appb-000022
为上述相似度值。
在一个实施例中，上述向该用户推荐该产品的步骤，包括：将上述相似度值与预设的推荐等级表进行匹配，上述推荐等级表包括不同相似度值范围与推荐等级的对应关系；根据匹配结果输出推荐等级；根据上述推荐等级向该用户推荐该产品。
在一个实施例中，上述对上述第一评分矩阵和第二评分矩阵分别进行数据扩充处理的方法，包括：以抽取上述第一评分矩阵以及第二评分矩阵中部分矩阵的方式进行数据扩充。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储与一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的和实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可以包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限，RAM一多种形式可得，诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双速据率SDRAM(SSRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
综上所述，根据本申请中的产品推荐方法得到的上述用户特征和上述产品特征的相似度值，当相似度值的越大，表示该用户对该产品越感兴趣，当上述相似度值大于预设值时，则向该用户推荐该产品，从而实现了充分利用用户的多种信息和产品特性信息，准确地预测出用户对产品的感兴趣程度，为用户推荐合理的产品，提高用户体验。
以上所述仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申 请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种产品推荐方法，其特征在于，包括：
    获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
    将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
    将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
    判断所述相似度值是否大于预设值；
    若是，则向该用户推荐该产品。
  2. 根据权利要求1所述的产品推荐方法，其特征在于，所述第一评分矩阵的生成方法，包括：
    获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到所述第一评分矩阵。
  3. 根据权利要求1所述的产品推荐方法，其特征在于，所述第二评分矩阵的生成方法，包括：
    获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到所述第二评分矩阵。
  4. 根据权利要求1所述的产品推荐方法，其特征在于，所述将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征的步骤之前，包括：
    对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
  5. 根据权利要求1所述的产品推荐方法，其特征在于，所述将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值的步骤，包括：
    根据公式
    Figure PCTCN2018108033-appb-100001
    计算得到所述用户特征和所述产品特征的相似度值，其中p i为所述用户特征，q j为所述产品特征，
    Figure PCTCN2018108033-appb-100002
    为所述相似度值。
  6. 根据权利要求1所述的产品推荐方法，其特征在于，所述向该用户推荐该产品的步 骤，包括：
    将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级表包括不同相似度值范围与推荐等级的对应关系；
    根据匹配结果输出推荐等级；
    根据所述推荐等级向该用户推荐该产品。
  7. 根据权利要求4所述的产品推荐方法，其特征在于，所述对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理的方法，包括：
    以抽取所述第一评分矩阵以及第二评分矩阵中部分矩阵的方式进行数据扩充。
  8. 一种产品推荐装置，其特征在于，包括：
    构造单元，用于获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
    第一计算单元，用于将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
    第二计算单元，用于将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
    判断单元，用于判断所述相似度值是否大于预设值；
    执行单元，用于当所述相似度值大于预设值时，则向该用户推荐该产品。
  9. 根据权利要求8所述的产品推荐装置，其特征在于，还包括：
    扩充单元，用于对所述第一评分矩阵和第二评分矩阵分别进行数据扩充处理。
  10. 根据权利要求8所述的产品推荐装置，其特征在于，所述执行单元包括：
    第一匹配模块，用于将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级表包括不同相似度值范围与推荐等级的对应关系；
    第二匹配模块，用于根据匹配结果输出推荐等级；
    推荐模块，用于根据所述推荐等级向该用户推荐该产品。
  11. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现产品推荐方法，该产品推荐方法包括：
    获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产 品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
    将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
    将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
    判断所述相似度值是否大于预设值；
    若是，则向该用户推荐该产品。
  12. 根据权利要求11所述的计算机设备，其特征在于，所述第一评分矩阵的生成方法，包括：
    获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到所述第一评分矩阵。
  13. 根据权利要求11所述的计算机设备，其特征在于，所述第二评分矩阵的生成方法，包括：
    获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到所述第二评分矩阵。
  14. 根据权利要求11所述的计算机设备，其特征在于，所述将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值的步骤，包括：
    根据公式
    Figure PCTCN2018108033-appb-100003
    计算得到所述用户特征和所述产品特征的相似度值，其中p i为所述用户特征，q j为所述产品特征，
    Figure PCTCN2018108033-appb-100004
    为所述相似度值。
  15. 根据权利要求11所述的计算机设备，其特征在于，所述向该用户推荐该产品的步骤，包括：
    将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级表包括不同相似度值范围与推荐等级的对应关系；
    根据匹配结果输出推荐等级；
    根据所述推荐等级向该用户推荐该产品。
  16. 一种计算机非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现产品推荐方法，该产品推荐方法包括：
    获取某个用户的用户数据和某个产品的产品数据，并根据所述用户数据得到用户对产 品的第一评分矩阵以及根据所述产品数据得到产品对用户的第二评分矩阵，所述用户数据为某个用户对所有产品的数据，所述产品数据为某个产品对所有用户的数据；
    将所述第一评分矩阵输入到预设的第一深度矩阵分解模型中进行计算得到用户特征，以及将所述第二评分矩阵输入到预设的第二深度矩阵分解模型中进行计算得到产品特征；
    将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值；
    判断所述相似度值是否大于预设值；
    若是，则向该用户推荐该产品。
  17. 根据权利要求16所述的计算机非易失性可读存储介质，其特征在于，所述第一评分矩阵的生成方法，包括：
    获取某个用户的用户数据，根据用户对所有产品在多个维度的偏好程度设置对应的评分参数构造得到所述第一评分矩阵。
  18. 根据权利要求16所述的计算机非易失性可读存储介质，其特征在于，所述第二评分矩阵的生成方法，包括：
    获取某个产品的产品数据，根据产品对所有用户在多个维度的适用程度设置对应的评分参数构造得到所述第二评分矩阵。
  19. 根据权利要求16所述的计算机非易失性可读存储介质，其特征在于，所述将所述用户特征和产品特征输入到预设的公式中进行计算得到所述用户特征和所述产品特征的相似度值的步骤，包括：
    根据公式
    Figure PCTCN2018108033-appb-100005
    计算得到所述用户特征和所述产品特征的相似度值，其中p i为所述用户特征，q j为所述产品特征，
    Figure PCTCN2018108033-appb-100006
    为所述相似度值。
  20. 根据权利要求16所述的计算机设备，其特征在于，所述向该用户推荐该产品的步骤，包括：
    将所述相似度值与预设的推荐等级表进行匹配，所述推荐等级表包括不同相似度值范围与推荐等级的对应关系；
    根据匹配结果输出推荐等级；
    根据所述推荐等级向该用户推荐该产品。
PCT/CN2018/108033 2018-06-13 2018-09-27 产品推荐方法、装置、计算机设备和存储介质 WO2019237569A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810607862.9 2018-06-13
CN201810607862.9A CN109034941B (zh) 2018-06-13 2018-06-13 产品推荐方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2019237569A1 true WO2019237569A1 (zh) 2019-12-19

Family

ID=64609539

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/108033 WO2019237569A1 (zh) 2018-06-13 2018-09-27 产品推荐方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN109034941B (zh)
WO (1) WO2019237569A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110083688B (zh) * 2019-05-10 2022-03-25 北京百度网讯科技有限公司 搜索结果召回方法、装置、服务器及存储介质
CN110598040B (zh) * 2019-09-17 2023-10-10 上海喜马拉雅科技有限公司 专辑召回方法、装置、设备及存储介质
CN113344673A (zh) * 2021-06-28 2021-09-03 平安信托有限责任公司 产品推送方法、装置、电子设备及存储介质
CN115880076A (zh) * 2022-04-20 2023-03-31 北京中关村科金技术有限公司 信托产品推荐方法、装置及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102426686A (zh) * 2011-09-29 2012-04-25 南京大学 一种基于矩阵分解的互联网信息产品推荐方法
CN102663627A (zh) * 2012-04-26 2012-09-12 焦点科技股份有限公司 个性化推荐方法
CN104021163A (zh) * 2014-05-28 2014-09-03 深圳市盛讯达科技股份有限公司 产品推荐系统及方法
CN107292648A (zh) * 2016-04-01 2017-10-24 株式会社理光 一种用户行为分析方法及装置
CN107562795A (zh) * 2017-08-01 2018-01-09 广州市香港科大霍英东研究院 基于异构信息网络的推荐方法及装置

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103246672B (zh) * 2012-02-09 2016-06-08 中国科学技术大学 对用户进行个性化推荐的方法和装置
CN106570008B (zh) * 2015-10-09 2020-03-27 阿里巴巴集团控股有限公司 推荐方法及装置
CN107578270A (zh) * 2017-08-03 2018-01-12 中国银联股份有限公司 一种金融标签的构建方法、装置及计算设备
CN107563841B (zh) * 2017-08-03 2021-02-05 电子科技大学 一种基于用户评分分解的推荐系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102426686A (zh) * 2011-09-29 2012-04-25 南京大学 一种基于矩阵分解的互联网信息产品推荐方法
CN102663627A (zh) * 2012-04-26 2012-09-12 焦点科技股份有限公司 个性化推荐方法
CN104021163A (zh) * 2014-05-28 2014-09-03 深圳市盛讯达科技股份有限公司 产品推荐系统及方法
CN107292648A (zh) * 2016-04-01 2017-10-24 株式会社理光 一种用户行为分析方法及装置
CN107562795A (zh) * 2017-08-01 2018-01-09 广州市香港科大霍英东研究院 基于异构信息网络的推荐方法及装置

Also Published As

Publication number Publication date
CN109034941B (zh) 2023-03-31
CN109034941A (zh) 2018-12-18

Similar Documents

Publication Publication Date Title
WO2019237569A1 (zh) 产品推荐方法、装置、计算机设备和存储介质
JP6843882B2 (ja) 履歴ログからの学習と、ｅｔｌツール内のデータアセットに関するデータベースオペレーションの推奨
Westerlund New simple tests for panel cointegration
US20190311395A1 (en) Estimating click-through rate
CN111080398A (zh) 商品推荐方法、装置、计算机设备和存储介质
CN106251174A (zh) 信息推荐方法及装置
Gallant et al. The relative efficiency of method of moments estimators
US20210366006A1 (en) Ranking of business object
WO2019232892A1 (zh) 预测投保风险概率的方法、装置、计算机设备和存储介质
Valkov Fitted finite volume method for a generalized Black–Scholes equation transformed on finite interval
CN108205775A (zh) 一种业务对象的推荐方法、装置和客户端
CN106022869A (zh) 一种消费对象的推荐方法及装置
US10937070B2 (en) Collaborative filtering to generate recommendations
CN110428295A (zh) 商品推荐方法和系统
Pintoux et al. The Dothan pricing model revisited
US20230245210A1 (en) Knowledge graph-based information recommendation
KR101635283B1 (ko) 행렬 분해 모델 기반 데이터 분석 방법 및 장치
CN114385930A (zh) 一种兴趣点推荐方法及系统
Uddin et al. A note on numerical solution of a linear black-scholes model
CN110659701B (zh) 信息处理方法、装置、电子设备及介质
CN107169837B (zh) 用于辅助搜索的方法、装置、电子设备及计算机可读介质
CN112925982B (zh) 用户重定向方法及装置、存储介质、计算机设备
US20140324523A1 (en) Missing String Compensation In Capped Customer Linkage Model
US20140324524A1 (en) Evolving a capped customer linkage model using genetic models
US12051025B2 (en) Artificial intelligence-enabled metadata optimizer

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18922354

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18922354

Country of ref document: EP

Kind code of ref document: A1