CN111859155A - Item recommendation method, equipment and computer-readable storage medium - Google Patents

Item recommendation method, equipment and computer-readable storage medium Download PDF

Info

Publication number
CN111859155A
CN111859155A CN202010774401.8A CN202010774401A CN111859155A CN 111859155 A CN111859155 A CN 111859155A CN 202010774401 A CN202010774401 A CN 202010774401A CN 111859155 A CN111859155 A CN 111859155A
Authority
CN
China
Prior art keywords
matrix
user
recommendation
item
interest
Prior art date
Legal status (The legal status 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 status listed.)
Granted
Application number
CN202010774401.8A
Other languages
Chinese (zh)
Other versions
CN111859155B (en
Inventor
谭奔
郑文琛
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
WeBank Co Ltd
Original Assignee
WeBank Co Ltd
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 WeBank Co Ltd filed Critical WeBank Co Ltd
Priority to CN202010774401.8A priority Critical patent/CN111859155B/en
Publication of CN111859155A publication Critical patent/CN111859155A/en
Priority to PCT/CN2021/101999 priority patent/WO2022028140A1/en
Application granted granted Critical
Publication of CN111859155B publication Critical patent/CN111859155B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/958—Organisation or management of web site content, e.g. publishing, maintaining pages or automatic linking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • 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
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0282—Rating or review of business operators or products
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • General Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Artificial Intelligence (AREA)
  • General Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The invention discloses an article recommendation method, equipment and a computer-readable storage medium, wherein the method comprises the following steps: calling a local interest matrix in a recommendation system, wherein the local interest matrix is formed by carrying out federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system; calculating the grade of each user on each system article in the recommendation system according to the article clustering matrix and the local interest matrix in the recommendation system; and recommending system articles to each user according to the scores of the users on the system articles. According to the method and the system, the safety of the number of users is ensured through federal training, and the recommended articles are directly related to the interests of the users according to the scores, so that the recommendation is more accurate, and the accurate recommendation of the articles is realized while the safety of user data is ensured.

Description

Item recommendation method, equipment and computer-readable storage medium
Technical Field
The invention relates to the technical field of financial technology (Fintech), in particular to an article recommendation method, device and computer-readable storage medium.
Background
With the continuous development of financial technology (Fintech), especially internet technology and finance, more and more technologies (such as artificial intelligence, big data, cloud storage and the like) are applied to the financial field, but the financial field also puts higher requirements on various technologies, and the recommendation of articles is safer and more accurate if the requirements are met.
Currently, article recommendation generally includes two ways, namely single system recommendation and joint multiple system recommendation, for a single system, recommendation is performed based on user behavior data collected by the system, and for a joint multiple system, user behavior data collected by each system are mutually transmitted to perform joint recommendation. However, the user behavior data collected by a single system is limited, and the user requirements are difficult to reflect comprehensively, so that the recommendation is inaccurate; and the transmission of the user behavior data collected by a plurality of systems is combined, so that the leakage of the user behavior data is easily caused, and the safety of the user data is influenced.
Therefore, how to implement accurate recommendation of articles while ensuring the security of user data is a technical problem to be solved urgently at present.
Disclosure of Invention
The invention mainly aims to provide an article recommendation method, equipment and a computer readable storage medium, and aims to solve the technical problem of how to realize accurate recommendation of articles while ensuring the safety of user data in the prior art.
In order to achieve the above object, the present invention provides an article recommendation method, including the steps of:
calling a local interest matrix in a recommendation system, wherein the local interest matrix is formed by carrying out federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system;
calculating the grade of each user on each system article in the recommendation system according to the article clustering matrix and the local interest matrix in the recommendation system;
and recommending system articles to each user according to the scores of the users on the system articles.
Optionally, the step of recommending system items to each user according to the score of each user on each system item includes:
ranking the scores of the users on the system articles, determining the target scores arranged in the front preset position corresponding to the users, and respectively executing the following steps on the target scores arranged in the front preset position corresponding to the users:
determining target objects corresponding to the target scores, and searching for objects to be recommended in the target objects, wherein the objects to be recommended do not carry comment information corresponding to the user;
and recommending the item to be recommended to the user.
Optionally, the step of calculating, according to the item clustering matrix and the local interest matrix in the recommendation system, a score of each user on each system item in the recommendation system includes:
determining an interest row of each user in the local interest matrix, and performing product operation on the interest row of each user and the item clustering matrix to generate an item scoring matrix of each user;
and determining the grade of each user on each system item according to the item grade matrix of each user.
Optionally, before the step of invoking the local interest matrix in the recommendation system, the method includes:
receiving a global interest feature matrix sent by a preset coordinator, and calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained to obtain a calculation result;
and judging whether the calculation result meets a calculation ending condition, and if so, generating the local interest matrix to be trained into the local interest matrix in the recommendation system.
Optionally, after the step of determining whether the calculation result satisfies the calculation end condition, the method further includes:
if the calculation result does not meet the condition for finishing calculation, generating a first gradient of the local interest matrix to be trained, a second gradient of the article clustering matrix to be trained and a third gradient of the global interest feature matrix according to the preset function;
and iteratively calculating the preset function according to the first gradient, the second gradient and the third gradient to update the calculation result until the calculation result meets the calculation ending condition.
Optionally, the step of iteratively calculating the preset function according to the first gradient, the second gradient and the third gradient to update the calculation result includes:
according to the first gradient and the second gradient, respectively updating the local interest matrix to be trained and the clustering matrix of the articles to be trained;
transmitting the third gradient to the preset coordinator, so that the preset coordinator can calculate a fourth gradient aggregate generated by the preset function through the third gradient and at least one other system based on second behavior data of each user, and generate a return gradient to update the global interest feature matrix;
and receiving the global interest feature matrix updated by the preset coordinator, and performing a step of calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained so as to perform iterative calculation on the preset function and update the calculation result.
Optionally, the step of receiving the global interest feature matrix sent by the preset coordinator includes:
and receiving a global interest feature matrix sent by the preset coordinator based on the encryption of a preset public key, and decrypting the global interest feature matrix based on a private key matched with the preset public key, wherein the preset public key and the private key are generated based on the recommendation system or other systems.
Optionally, after the step of generating the local interest matrix to be trained as the local interest matrix in the recommendation system, the method includes:
and generating the article clustering matrix to be trained into an article clustering matrix in the recommendation system.
Further, to achieve the above object, the present invention also provides an article recommendation apparatus, which includes a memory, a processor, and an article recommendation program stored on the memory and executable on the processor, and when executed by the processor, the article recommendation program implements the steps of the article recommendation method as described above.
Further, to achieve the above object, the present invention also provides a computer-readable storage medium having an item recommendation program stored thereon, which when executed by a processor implements the steps of the item recommendation method as described above.
Compared with the prior art that recommendation is performed according to user behavior data collected by a single system or according to a mode of transmission combination of user behavior data collected by a plurality of systems, the item recommendation method, the item recommendation equipment and the computer-readable storage medium calculate the scores of each user in the recommendation system on the items of each system by calling a local interest matrix in the recommendation system according to an item clustering matrix and a local interest matrix in the recommendation system; and recommending the system articles to each user according to the scores of each user on each system article. The local interest matrix and the article clustering matrix are generated by performing federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system; during federal training, the first behavior data and the second behavior data are in respective systems, and safety of user data is facilitated. In addition, the local interest matrix reflects the interest points of the users in the recommendation system through the first behavior data of the users, and the scores calculated through the local interest matrix and the item clustering matrix reflect the interest concentration degree of the users to each system item in the recommendation system, so that the items recommended according to the scores are directly related to the interests of the users, and the recommendation is more accurate. The method overcomes the defects of inaccuracy of recommendation according to the user behavior data collected by a single system and insecurity of transmission joint recommendation according to the user behavior data collected by a plurality of systems in the prior art. The method and the device ensure the safety of user data and realize the accurate recommendation of the articles.
Drawings
FIG. 1 is a schematic structural diagram of an apparatus hardware operating environment according to an embodiment of the item recommendation apparatus of the present invention;
fig. 2 is a flowchart illustrating a first embodiment of an item recommendation method according to the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides an article recommendation device, and referring to fig. 1, fig. 1 is a schematic structural diagram of a device hardware operating environment according to an embodiment of the article recommendation device of the invention.
As shown in fig. 1, the item recommendation apparatus may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a memory device separate from the processor 1001 described above.
Those skilled in the art will appreciate that the hardware configuration of the item recommendation device illustrated in FIG. 1 does not constitute a limitation of the item recommendation device, and may include more or less components than those illustrated, or some components in combination, or a different arrangement of components.
As shown in fig. 1, a memory 1005, which is a kind of computer-readable storage medium, may include therein an operating system, a network communication module, a user interface module, and an item recommendation program. The operating system is a program for managing and controlling the item recommendation equipment and software resources and supports the operation of the network communication module, the user interface module, the item recommendation program and other programs or software; the network communication module is used to manage and control the network interface 1004; the user interface module is used to manage and control the user interface 1003.
In the hardware structure of the item recommendation device shown in fig. 1, the network interface 1004 is mainly used for connecting to a background server and performing data communication with the background server; the user interface 1003 is mainly used for connecting a client (user side) and performing data communication with the client; the processor 1001 may call the item recommendation program stored in the memory 1005 and perform the following operations:
calling a local interest matrix in a recommendation system, wherein the local interest matrix is formed by carrying out federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system;
calculating the grade of each user on each system article in the recommendation system according to the article clustering matrix and the local interest matrix in the recommendation system;
and recommending system articles to each user according to the scores of the users on the system articles.
Further, the step of recommending system items to each user according to the rating of each user on each system item includes:
ranking the scores of the users on the system articles, determining the target scores arranged in the front preset position corresponding to the users, and respectively executing the following steps on the target scores arranged in the front preset position corresponding to the users:
determining target objects corresponding to the target scores, and searching for objects to be recommended in the target objects, wherein the objects to be recommended do not carry comment information corresponding to the user;
and recommending the item to be recommended to the user.
Further, the step of calculating the score of each user on each system item in the recommendation system according to the item clustering matrix and the local interest matrix in the recommendation system comprises:
determining an interest row of each user in the local interest matrix, and performing product operation on the interest row of each user and the item clustering matrix to generate an item scoring matrix of each user;
and determining the grade of each user on each system item according to the item grade matrix of each user.
Further, prior to the step of invoking a local interest matrix within the recommendation system, the processor 1001 may invoke an item recommendation program stored in the memory 1005 and perform the following operations:
receiving a global interest feature matrix sent by a preset coordinator, and calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained to obtain a calculation result;
and judging whether the calculation result meets a calculation ending condition, and if so, generating the local interest matrix to be trained into the local interest matrix in the recommendation system.
Further, after the step of determining whether the calculation result satisfies the end calculation condition, the processor 1001 may call an item recommendation program stored in the memory 1005, and perform the following operations:
if the calculation result does not meet the condition for finishing calculation, generating a first gradient of the local interest matrix to be trained, a second gradient of the article clustering matrix to be trained and a third gradient of the global interest feature matrix according to the preset function;
and iteratively calculating the preset function according to the first gradient, the second gradient and the third gradient to update the calculation result until the calculation result meets the calculation ending condition.
Further, the step of iteratively calculating the preset function according to the first gradient, the second gradient and the third gradient to update the calculation result includes:
according to the first gradient and the second gradient, respectively updating the local interest matrix to be trained and the clustering matrix of the articles to be trained;
transmitting the third gradient to the preset coordinator, so that the preset coordinator can calculate a fourth gradient aggregate generated by the preset function through the third gradient and at least one other system based on second behavior data of each user, and generate a return gradient to update the global interest feature matrix;
and receiving the global interest feature matrix updated by the preset coordinator, and performing a step of calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained so as to perform iterative calculation on the preset function and update the calculation result.
Further, the step of receiving the global interest feature matrix sent by the preset coordinator includes:
and receiving a global interest feature matrix sent by the preset coordinator based on the encryption of a preset public key, and decrypting the global interest feature matrix based on a private key matched with the preset public key, wherein the preset public key and the private key are generated based on the recommendation system or other systems.
Further, after the step of generating the local interest matrix to be trained as the local interest matrix in the recommendation system, the method includes that the processor 1001 may call an item recommendation program stored in the memory 1005, and perform the following operations:
and generating the article clustering matrix to be trained into an article clustering matrix in the recommendation system.
The specific implementation of the item recommendation device of the present invention is substantially the same as the following embodiments of the item recommendation method, and is not described herein again.
The invention further provides an article recommendation method.
Referring to fig. 2, fig. 2 is a flowchart illustrating a first embodiment of an item recommendation method according to the present invention.
While a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in an order different than presented herein. Specifically, the item recommendation method in this embodiment includes:
step S10, a local interest matrix in a recommendation system is called, wherein the local interest matrix is formed by carrying out federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system;
the item recommendation method in the embodiment is applied to a recommendation system participating in federal learning, and the recommendation system comprises a local interest matrix and an item clustering matrix trained by the federal and is suitable for performing systematic item recommendation to users in the recommendation system according to scores obtained by calculation of the local interest matrix and the item clustering matrix. Specifically, a federal training mechanism is provided between the recommendation system and other systems, and the other systems are additional systems different from the recommendation system, and may be the same as or different from the recommendation system in type, such as a system for implementing shopping or a system for implementing web browsing. The same user performs different operations in different systems to generate different behavior data, and the recommendation system and other systems collect behavior data of multiple users and screen out the same user from the multiple users.
Further, the recommendation system takes the behavior data collected for the same user as the first behavior data of the user in the recommendation system; and the other systems take the collected behavior data of the same user as second behavior data of each user in the other systems. And performing federal training according to the first behavior data and the second behavior data to generate a local interest matrix and an article clustering matrix in the recommendation system. The local interest matrix represents interest points of each user in the recommendation system, and the item clustering matrix represents items corresponding to the interest points in the recommendation system. Moreover, the local interest Matrix and the article clustering Matrix may be formed according to a Matrix Factorization (MF for short), and the behavior data Matrix of the article by the user may be decomposed into a product of two sub-matrices, such as the local interest Matrix and the article clustering Matrix in this embodiment. In the embodiment, the item recommendation is performed according to the local interest matrix and the local interest matrix by calling the local interest matrix obtained through federal training in the recommendation system.
Step S20, calculating the grade of each user on each system item in the recommendation system according to the item clustering matrix and the local interest matrix in the recommendation system;
further, after an article matrix and a local interest matrix of the recommendation system are obtained through federal training and the local interest matrix is called, the article clustering matrix and the local interest matrix are calculated to obtain the scores of each user in the recommendation system on each system article. And each system item is an item possessed by the recommendation system, and the interest level of each user in each item is reflected through each score. Specifically, the step of calculating the score of each user on each system item in the recommendation system according to the item clustering matrix and the local interest matrix in the recommendation system comprises:
step S21, determining the interest row of each user in the local interest matrix, and performing product operation on the interest row of each user and the item clustering matrix to generate an item scoring matrix of each user;
step S22, determining the rating of each user on each system item according to the item rating matrix of each user.
Understandably, in the local interest matrix, the matrix row is marked as a user and the interest is marked as an array; a matrix row characterizes the interest data of a user, and a data array characterizes all users having the interest. In the article clustering matrix, the matrix row titles are interests, and articles are listed according to the array titles; a matrix row characterizes items that represent an interest, and a matrix row characterizes the interest corresponding to the item according to an array. In the process of generating the scores of the users on the system articles according to the article clustering matrix and the local interest matrix, the interest rows of each user in the local interest matrix are determined, and then the product operation is carried out on the interest rows and the article clustering matrix respectively to generate the article scoring matrix of each user. A user corresponds to an item scoring matrix, the line titles of the matrix are all the interests, the column titles of the matrix are all the items, and all the numerical values in the item scoring matrix represent the interest of the user in all the items in all the interests.
Furthermore, due to the numerical values in the item scoring matrix, the interest level of the user in each item in each interest is reflected, so that the score of each user on each system item can be determined according to the item scoring matrix of each user. Specifically, each numerical value is used as a score of each system article, and the interest level of the user in each system article in the recommendation system is reflected.
And step S30, recommending system items to each user according to the scores of the users on the system items.
Furthermore, after the scores of each user on the system items are determined, the level of each score represents the interest level of each user in each system item; system items may be recommended to each user based on each user's respective scores. Specifically, the step of recommending system items to each user according to the scores of each user on the system items comprises:
step S31, ranking the scores of each user on each system item, determining the target score arranged in the front preset position corresponding to each user, and performing the following steps for the target scores arranged in the front preset position corresponding to each user:
step S32, determining target articles corresponding to each target score, and searching for articles to be recommended in each target article, wherein the articles to be recommended do not carry comment information corresponding to the user;
and step S33, recommending the item to be recommended to the user.
Understandably, the demands of each user on different items are different, and the obtained scores are also different, so that the scores of each user on the system items are arranged from large to small according to the user, and the score sequence of each user is obtained. And the front preset position for recommendation is set in advance according to the requirement, and if the front 3 positions are set, the front three items arranged in the scoring sequence are recommended. Therefore, the scores arranged at the front preset positions in each score sequence are determined as target scores, and the articles generating each target score in the article clustering matrix are searched as target articles. And then, for each target object of each user, judging whether each target object carries comment information corresponding to the user. The comment information is information generated by evaluating the target object after the user uses the target object. If some object has the comment information of the user, the object is indicated to be used by the user, and the object does not have the reuse requirement at present, so that the object is removed from the object. And forming the target object without the object into an object to be recommended, and recommending the object to the user.
In the method, a time detection mechanism can be set for judging whether comment information exists or not. Presetting preset time for representing recent use, continuously searching for the generation time of comment information after determining that some object has the comment information of a user in a target object, further determining the time length of the generation time from the current time, and judging whether the time length is within a preset time range. If the time is within the preset time range, the user is indicated to use the article in a short time, and the article is removed from the target article. If the time is not within the preset time range, the fact that the time when the user uses the object is longer than the current time is indicated, the object possibly has a reuse requirement, and therefore the object is not removed from the target object and is used as an object to be recommended.
Compared with the prior art that recommendation is performed according to user behavior data collected by a single system or according to a transmission combination of user behavior data collected by a plurality of systems, the item recommendation method provided by the invention calls a local interest matrix in the recommendation system for recommendation of items, and calculates the score of each user in the recommendation system on each system item according to an item clustering matrix and the local interest matrix in the recommendation system; and recommending the system articles to each user according to the scores of each user on each system article. The local interest matrix and the article clustering matrix are generated by performing federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system; during federal training, the first behavior data and the second behavior data are in respective systems, and safety of user data is facilitated. In addition, the local interest matrix reflects the interest points of the users in the recommendation system through the first behavior data of the users, and the scores calculated through the local interest matrix and the item clustering matrix reflect the interest concentration degree of the users to each system item in the recommendation system, so that the items recommended according to the scores are directly related to the interests of the users, and the recommendation is more accurate. The method overcomes the defects of inaccuracy of recommendation according to the user behavior data collected by a single system and insecurity of transmission joint recommendation according to the user behavior data collected by a plurality of systems in the prior art. The method and the device ensure the safety of user data and realize the accurate recommendation of the articles.
Further, based on the first embodiment of the item recommendation method of the present invention, a second embodiment of the item recommendation method of the present invention is proposed.
The second embodiment of the item recommendation method differs from the first embodiment of the item recommendation method in that the step of invoking a local interest matrix within the recommendation system is preceded by the method comprising:
step S11, receiving a global interest feature matrix sent by a preset coordinator, and calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained to obtain a calculation result;
in the embodiment, the local interest matrix and the item clustering matrix are generated through the first behavior data of each user in the recommendation system and the federal training between the second data of each user in other systems. Specifically, a preset coordinator of federal training is set, and the preset coordinator can be a third-party server in communication connection with the recommendation system and other systems, and can also be any system of the recommendation system and other systems. And forming a whole local interest characteristic matrix by a preset coordination square, sending the whole local interest characteristic matrix to a recommendation system, and carrying out federal training by combining the first behavior data and the second behavior data.
Further, in order to ensure the security of the communication data between the predetermined coordinator and the recommendation system, as well as between other systems, a mechanism for encrypted transmission is set for the transmitted data, i.e. the global interest feature matrix. Specifically, the step of receiving the global interest feature matrix sent by the preset coordinator includes:
step a, receiving a global interest feature matrix sent by the preset coordinator based on the encryption of a preset public key, and decrypting the global interest feature matrix based on a private key matched with the preset public key, wherein the preset public key and the private key are generated based on the recommendation system or other systems.
Further, a public key pk and a private key sk are generated by any one of the recommendation system and the other systems, the public key pk is used as a preset public key, and the private key sk is used as a private key matched with the preset public key. And sending the public key pk to a preset coordinator, and sending the private key sk to a recommendation system and other systems. And then, a preset coordination square forms a global interest characteristic matrix, the global interest characteristic matrix is initialized by using a random value, and the public key sk is used for encrypting and sending the global interest characteristic matrix to a recommendation system and other systems. After receiving the global interest feature matrix sent by the preset coordinator after being encrypted by the preset public key, the recommendation system calls the received private key sk to decrypt the global interest feature matrix to obtain the global interest feature matrix of the plaintext.
Further, a local interest matrix to be trained and an article clustering matrix to be trained for training are preset in the recommendation system, the recommendation system initializes the local interest matrix to be trained and the article clustering matrix to be trained by random numbers, and after receiving the global interest feature matrix, combines the global interest feature matrix with first behavior data of each user in the recommendation system, the local interest matrix to be trained and the article clustering matrix to be trained, calculates a preset function, and calculates a result. And training the local interest matrix to be trained and the article clustering matrix to be trained through the calculation result to obtain the final local interest matrix and article clustering matrix. The preset function is an optimization function of matrix decomposition, and the expression of the preset function is shown as formula (1):
Figure BDA0002617857180000111
wherein N represents the number of the recommendation system and other systems, Xi represents a behavior data matrix formed by behavior data of a user in the system, and when i represents the recommendation system, Xi is a behavior data matrix formed by first behavior data; u represents a global interest characteristic matrix, Ui represents a local interest matrix to be trained, Vi represents an article clustering matrix to be trained, lambda represents a local interest hyper-parameter, and gamma represents the complexity of a model corresponding to a preset function. And, | | Xi-UViI is used to mine the global interest of the user, Xi-UiViI is used for mining local interest of users, Ui-UI represents the difference of local interest from global interest,
Figure BDA0002617857180000121
a regularization term is represented. When λ is set to a small value, mining of global interest is focused, and when λ is set to a large value, mining of local interest is focused. The global interest characteristic matrix reflects the global property of the user interest and represents the interest of each user in the recommendation system and other systems; the local interest characteristic matrix reflects the locality of the user interests and represents the interests of each user on a recommendation system or other systems. The fine granularity of the local interests is higher relative to the global interests, and if the global interests are sports, the local behaviors are basketball, football, and the like.
And step S12, judging whether the calculation result meets the calculation ending condition, and if the calculation ending condition is met, generating the local interest matrix to be trained into the local interest matrix in the recommendation system.
Step S13, if the calculation result does not meet the end calculation condition, generating a first gradient of the local interest matrix to be trained, a second gradient of the item cluster matrix to be trained and a third gradient of the global interest feature matrix according to the preset function;
step S14, iteratively calculating the preset function according to the first gradient, the second gradient, and the third gradient to update the calculation result until the calculation result satisfies the end calculation condition.
Further, an end calculation condition for judging whether the calculation is ended is set in advance, and after the calculation result is generated, the calculation result is compared with the end calculation condition to judge whether the calculation result satisfies the end calculation condition. If the local interest matrix and the object clustering matrix are satisfied, the local interest matrix to be trained and the object clustering matrix to be trained for training achieve a better effect through training, so that the calculation is finished, and the local interest matrix to be trained and the object clustering matrix to be trained are respectively generated into the local interest matrix and the object clustering matrix.
It should be noted that the ending calculation condition may be in the form of a numerical value, that is, the ending calculation condition is set to a smaller numerical value representing that the training effect is achieved. If the calculated result is smaller than the numerical value, the representation calculation result meets the condition of finishing the calculation, otherwise, the finish calculation condition is not met.
Further, if the calculation result does not meet the end calculation condition after determination, a preset function is called to respectively derive the local interest matrix to be trained, the item clustering matrix to be trained and the global interest feature matrix, and a first gradient of the local interest matrix to be trained, a second gradient of the item clustering matrix to be trained and a third gradient of the global interest feature matrix are generated. And updating the interest matrix to be trained, the clustering matrix of the articles to be trained and the global interest characteristic matrix according to the first gradient, the second gradient and the second gradient, calculating a preset function after updating to obtain a new calculation result, and judging whether a calculation finishing condition is met. And (4) performing iterative calculation in such a way until the calculated result meets the calculation ending condition, stopping calculation, and obtaining a local interest matrix and an article clustering matrix.
Further, the step of iteratively calculating the preset function according to the first gradient, the second gradient and the third gradient to update the calculation result includes:
step S141, respectively updating the local interest matrix to be trained and the clustering matrix of the articles to be trained according to the first gradient and the second gradient;
step S142, transmitting the third gradient to the preset coordinator, so that the preset coordinator can calculate a fourth gradient aggregation generated by the preset function with the third gradient and at least one other system based on the second behavior data of each user, and generate a return gradient to update the global interest feature matrix;
step S143, receiving the global interest feature matrix updated by the preset coordinator, and performing a step of calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system, and the clustering matrix of the articles to be trained, so as to perform iterative calculation on the preset function and update the calculation result.
Understandably, the local interest matrix to be trained and the clustering matrix of the articles to be trained are related to the propulsion system and are not related to other systems, so that when the local interest matrix to be trained and the clustering matrix of the articles to be trained are updated, the first gradient is added to the local interest matrix to be trained, the local interest matrix to be trained is updated, and the second gradient is added to the clustering matrix of the articles to be trained, and the clustering matrix of the articles to be trained is updated.
And for the global interest characteristic matrix, because the global interest characteristic matrix is related to the recommendation system and other systems, when the global interest characteristic matrix is updated, the third gradient is encrypted and then transmitted to the preset coordinator. And the preset coordinator receives at least a fourth gradient transmitted by other systems besides the third gradient transmitted by the recommendation system. The fourth gradient is generated by the other systems through calculation of the preset function based on the second behavior data of each user, and the calculation mode is the same as that of the third gradient, which is not described herein again.
Further, the preset coordinator decrypts and aggregates the received third gradient and the at least one fourth gradient to obtain a returned gradient, updates the global interest feature matrix, and transmits the updated global interest feature matrix to the recommendation system and other systems. After receiving the global interest feature matrix updated by the preset coordinator, the recommendation system calculates the preset function according to the updated global interest feature matrix, the first behavior data of each item, the local interest matrix to be trained and the clustering matrix of the items to be trained so as to update the preset function and realize iterative calculation of the preset function until the calculation result after iteration meets the calculation ending condition.
The preset function is implemented to obtain a local interest matrix, an article clustering matrix and a global interest characteristic matrix through federal training, and the local interest matrix and the article clustering matrix of the recommendation system are obtained by combining the global interest characteristic matrices embodied by user data in a plurality of systems on the basis of ensuring user privacy and data safety so as to embody the interest of users in each system article in the recommendation system, so that article recommendation performed through the local interest matrix and the article clustering matrix has higher accuracy and safety.
In addition, the embodiment of the invention also provides a computer readable storage medium.
The computer readable storage medium has stored thereon an item recommendation program which, when executed by a processor, implements the steps of the item recommendation method as described above. The specific implementation of the computer-readable storage medium of the present invention is substantially the same as the embodiments of the item recommendation method described above, and is not described herein again.
The present invention is described in connection with the accompanying drawings, but the present invention is not limited to the above embodiments, which are only illustrative and not restrictive, and those skilled in the art can make various changes without departing from the spirit and scope of the invention as defined by the appended claims, and all changes that come within the meaning and range of equivalency of the specification and drawings that are obvious from the description and the attached claims are intended to be embraced therein.

Claims (10)

1. An item recommendation method, characterized in that the item recommendation method comprises the steps of:
calling a local interest matrix in a recommendation system, wherein the local interest matrix is formed by carrying out federal training on first behavior data of each user in the recommendation system and second behavior data of each user in at least one other system;
calculating the grade of each user on each system article in the recommendation system according to the article clustering matrix and the local interest matrix in the recommendation system;
and recommending system articles to each user according to the scores of the users on the system articles.
2. The item recommendation method of claim 1, wherein said step of making system item recommendations to each of said users based on each of said users' scores on each of said system items comprises:
ranking the scores of the users on the system articles, determining the target scores arranged in the front preset position corresponding to the users, and respectively executing the following steps on the target scores arranged in the front preset position corresponding to the users:
determining target objects corresponding to the target scores, and searching for objects to be recommended in the target objects, wherein the objects to be recommended do not carry comment information corresponding to the user;
and recommending the item to be recommended to the user.
3. The item recommendation method of claim 1, wherein said step of calculating a rating for each of said users on system items in said recommendation system based on an item clustering matrix and said local interest matrix in said recommendation system comprises:
determining an interest row of each user in the local interest matrix, and performing product operation on the interest row of each user and the item clustering matrix to generate an item scoring matrix of each user;
and determining the grade of each user on each system item according to the item grade matrix of each user.
4. The item recommendation method of claim 1, wherein said step of invoking a local interest matrix within a recommendation system is preceded by the method comprising:
receiving a global interest feature matrix sent by a preset coordinator, and calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained to obtain a calculation result;
and judging whether the calculation result meets a calculation ending condition, and if so, generating the local interest matrix to be trained into the local interest matrix in the recommendation system.
5. The item recommendation method according to claim 4, wherein after the step of determining whether the calculation result satisfies the end calculation condition, the method further comprises:
if the calculation result does not meet the condition for finishing calculation, generating a first gradient of the local interest matrix to be trained, a second gradient of the article clustering matrix to be trained and a third gradient of the global interest feature matrix according to the preset function;
and iteratively calculating the preset function according to the first gradient, the second gradient and the third gradient to update the calculation result until the calculation result meets the calculation ending condition.
6. The item recommendation method of claim 5, wherein the step of iteratively calculating the preset function based on the first gradient, the second gradient, and the third gradient to update the calculation result comprises:
according to the first gradient and the second gradient, respectively updating the local interest matrix to be trained and the clustering matrix of the articles to be trained;
transmitting the third gradient to the preset coordinator, so that the preset coordinator can calculate a fourth gradient aggregate generated by the preset function through the third gradient and at least one other system based on second behavior data of each user, and generate a return gradient to update the global interest feature matrix;
and receiving the global interest feature matrix updated by the preset coordinator, and performing a step of calculating a preset function according to the global interest feature matrix, the first behavior data, the local interest matrix to be trained in the recommendation system and the clustering matrix of the articles to be trained so as to perform iterative calculation on the preset function and update the calculation result.
7. The item recommendation method according to claim 4, wherein the step of receiving the global interest feature matrix sent by the preset coordinator comprises:
and receiving a global interest feature matrix sent by the preset coordinator based on the encryption of a preset public key, and decrypting the global interest feature matrix based on a private key matched with the preset public key, wherein the preset public key and the private key are generated based on the recommendation system or other systems.
8. The item recommendation method of claim 4, wherein after the step of generating the local interest matrix to be trained as a local interest matrix within the recommendation system, the method comprises:
and generating the article clustering matrix to be trained into an article clustering matrix in the recommendation system.
9. An item recommendation device, characterized in that the item recommendation device comprises a memory, a processor and an item recommendation program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the item recommendation method according to any one of claims 1-8.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon an item recommendation program which, when executed by a processor, implements the steps of the item recommendation method according to any one of claims 1-8.
CN202010774401.8A 2020-08-04 2020-08-04 Item recommendation method, device and computer readable storage medium Active CN111859155B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN202010774401.8A CN111859155B (en) 2020-08-04 2020-08-04 Item recommendation method, device and computer readable storage medium
PCT/CN2021/101999 WO2022028140A1 (en) 2020-08-04 2021-06-24 Item recommendation method and device, and computer readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010774401.8A CN111859155B (en) 2020-08-04 2020-08-04 Item recommendation method, device and computer readable storage medium

Publications (2)

Publication Number Publication Date
CN111859155A true CN111859155A (en) 2020-10-30
CN111859155B CN111859155B (en) 2024-08-13

Family

ID=72953567

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010774401.8A Active CN111859155B (en) 2020-08-04 2020-08-04 Item recommendation method, device and computer readable storage medium

Country Status (2)

Country Link
CN (1) CN111859155B (en)
WO (1) WO2022028140A1 (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112380439A (en) * 2020-11-19 2021-02-19 平安科技（深圳）有限公司 Target object recommendation method and device, electronic equipment and computer-readable storage medium
CN112633356A (en) * 2020-12-18 2021-04-09 平安科技（深圳）有限公司 Recommendation model training method, recommendation device, recommendation equipment and storage medium
WO2022028140A1 (en) * 2020-08-04 2022-02-10 深圳前海微众银行股份有限公司 Item recommendation method and device, and computer readable storage medium

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117951388B (en) * 2024-03-19 2024-07-05 深圳市幺柒零信息科技有限公司 Big data-based product recommendation strategy generation method, system and medium

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
IL212502A0 (en) * 2011-04-26 2011-07-31 Friedmann Michael Method and system for recommending geo-tagged items
US20170206551A1 (en) * 2016-01-15 2017-07-20 Adobe Systems Incorporated Personalized Recommendation Computation in Real Time using Incremental Matrix Factorization and User Factor Clustering
CN108509573A (en) * 2018-03-27 2018-09-07 陕西科技大学 Book recommendation method based on matrix decomposition collaborative filtering and system
CN108776919A (en) * 2018-05-31 2018-11-09 西安电子科技大学 The item recommendation method of information core is built based on cluster and evolution algorithm
CN108984551A (en) * 2017-05-31 2018-12-11 广州智慧城市发展研究院 A kind of recommended method and system based on the multi-class soft cluster of joint
CN109271592A (en) * 2018-09-30 2019-01-25 东软集团股份有限公司 A kind of item recommendation method, device, storage medium and program product
CN109635291A (en) * 2018-12-04 2019-04-16 重庆理工大学 A kind of recommended method of fusion score information and item contents based on coorinated training
CN109902235A (en) * 2019-03-06 2019-06-18 太原理工大学 User preference based on bat optimization clusters Collaborative Filtering Recommendation Algorithm
CN110189192A (en) * 2019-05-10 2019-08-30 深圳前海微众银行股份有限公司 A kind of generation method and device of information recommendation model
CN110503506A (en) * 2019-07-05 2019-11-26 平安科技（深圳）有限公司 Item recommendation method, device and medium based on score data
WO2020098156A1 (en) * 2018-11-15 2020-05-22 平安科技（深圳）有限公司 Product data pushing method and apparatus, and non-volatile computer-readable storage medium
CN111292151A (en) * 2018-12-07 2020-06-16 中移互联网有限公司 Object recommendation method and device, storage medium and object recommendation equipment
CN111291273A (en) * 2020-02-20 2020-06-16 深圳前海微众银行股份有限公司 Recommendation system optimization method, device, equipment and readable storage medium
CN111324812A (en) * 2020-02-20 2020-06-23 深圳前海微众银行股份有限公司 Federal recommendation method, device, equipment and medium based on transfer learning

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7240055B2 (en) * 2003-12-11 2007-07-03 Xerox Corporation Method and system for expertise mapping based on user activity in recommender systems
CN104866540B (en) * 2015-05-04 2018-04-27 华中科技大学 A kind of personalized recommendation method based on group of subscribers behavioural analysis
CN106682114B (en) * 2016-12-07 2020-10-27 广东工业大学 Personalized recommendation method integrating user trust relationship and comment information
US11250347B2 (en) * 2018-06-27 2022-02-15 Microsoft Technology Licensing, Llc Personalization enhanced recommendation models
CN110188283B (en) * 2019-06-05 2021-11-23 中国人民解放军国防科技大学 Information recommendation method and system based on joint neural network collaborative filtering
CN110825957B (en) * 2019-09-17 2023-04-11 中国平安人寿保险股份有限公司 Deep learning-based information recommendation method, device, equipment and storage medium
CN110991601B (en) * 2019-11-29 2023-04-21 中山大学 Neural network recommendation method based on multi-user behavior
CN111339412A (en) * 2020-02-20 2020-06-26 深圳前海微众银行股份有限公司 Longitudinal federal recommendation recall method, device, equipment and readable storage medium
CN111859155B (en) * 2020-08-04 2024-08-13 深圳前海微众银行股份有限公司 Item recommendation method, device and computer readable storage medium

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
IL212502A0 (en) * 2011-04-26 2011-07-31 Friedmann Michael Method and system for recommending geo-tagged items
US20170206551A1 (en) * 2016-01-15 2017-07-20 Adobe Systems Incorporated Personalized Recommendation Computation in Real Time using Incremental Matrix Factorization and User Factor Clustering
CN108984551A (en) * 2017-05-31 2018-12-11 广州智慧城市发展研究院 A kind of recommended method and system based on the multi-class soft cluster of joint
CN108509573A (en) * 2018-03-27 2018-09-07 陕西科技大学 Book recommendation method based on matrix decomposition collaborative filtering and system
CN108776919A (en) * 2018-05-31 2018-11-09 西安电子科技大学 The item recommendation method of information core is built based on cluster and evolution algorithm
CN109271592A (en) * 2018-09-30 2019-01-25 东软集团股份有限公司 A kind of item recommendation method, device, storage medium and program product
WO2020098156A1 (en) * 2018-11-15 2020-05-22 平安科技（深圳）有限公司 Product data pushing method and apparatus, and non-volatile computer-readable storage medium
CN109635291A (en) * 2018-12-04 2019-04-16 重庆理工大学 A kind of recommended method of fusion score information and item contents based on coorinated training
CN111292151A (en) * 2018-12-07 2020-06-16 中移互联网有限公司 Object recommendation method and device, storage medium and object recommendation equipment
CN109902235A (en) * 2019-03-06 2019-06-18 太原理工大学 User preference based on bat optimization clusters Collaborative Filtering Recommendation Algorithm
CN110189192A (en) * 2019-05-10 2019-08-30 深圳前海微众银行股份有限公司 A kind of generation method and device of information recommendation model
CN110503506A (en) * 2019-07-05 2019-11-26 平安科技（深圳）有限公司 Item recommendation method, device and medium based on score data
CN111291273A (en) * 2020-02-20 2020-06-16 深圳前海微众银行股份有限公司 Recommendation system optimization method, device, equipment and readable storage medium
CN111324812A (en) * 2020-02-20 2020-06-23 深圳前海微众银行股份有限公司 Federal recommendation method, device, equipment and medium based on transfer learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李豪、张海英、张俊: "一种基于用户兴趣分析的协同过滤推荐算法", 《软件与算法》, vol. 36, no. 15, 16 August 2017 (2017-08-16), pages 25 - 28 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2022028140A1 (en) * 2020-08-04 2022-02-10 深圳前海微众银行股份有限公司 Item recommendation method and device, and computer readable storage medium
CN112380439A (en) * 2020-11-19 2021-02-19 平安科技（深圳）有限公司 Target object recommendation method and device, electronic equipment and computer-readable storage medium
CN112380439B (en) * 2020-11-19 2024-06-04 平安科技（深圳）有限公司 Target object recommendation method and device, electronic equipment and computer readable storage medium
CN112633356A (en) * 2020-12-18 2021-04-09 平安科技（深圳）有限公司 Recommendation model training method, recommendation device, recommendation equipment and storage medium
CN112633356B (en) * 2020-12-18 2024-09-10 平安科技（深圳）有限公司 Recommendation model training method, recommendation device, recommendation equipment and storage medium

Also Published As

Publication number Publication date
CN111859155B (en) 2024-08-13
WO2022028140A1 (en) 2022-02-10

Similar Documents

Publication Publication Date Title
CN111859155B (en) Item recommendation method, device and computer readable storage medium
KR102358604B1 (en) Convergence data processing method and information recommendation system
CN106202331B (en) The recommender system of secret protection and the operational method based on the recommender system by different level
US10102292B2 (en) Method and system of processing a search query
US20200356875A1 (en) Model training
US9177046B2 (en) Refining image relevance models
US8671093B2 (en) Click model for search rankings
US9384511B1 (en) Version control for resource allocation modeling
WO2019029046A1 (en) Video recommendation method and system
Zhao et al. Crowd-Selection Query Processing in Crowdsourcing Databases: A Task-Driven Approach.
CN111324812B (en) Federal recommendation method, device, equipment and medium based on transfer learning
US20170249388A1 (en) Expert Detection in Social Networks
CN107507073A (en) Based on the service recommendation method for trusting extension and the sequence study of list level
US10936601B2 (en) Combined predictions methodology
US20200084019A1 (en) Blockchain Ranking Engine
CN105760443A (en) Project recommending system, device and method
CN111553744A (en) Federal product recommendation method, device, equipment and computer storage medium
CN111339412A (en) Longitudinal federal recommendation recall method, device, equipment and readable storage medium
US9047319B2 (en) Tag association with image regions
EP3699840A1 (en) Method for analyzing supply chain risk of suppliers
Oosterhuis et al. Sensitive and scalable online evaluation with theoretical guarantees
US20240143607A1 (en) Facilitating efficient identification of relevant data
CN103744904A (en) Method and device for providing information
US12008033B2 (en) Determining user affinities for content generation applications
KR102698458B1 (en) Secure management of data distribution restrictions

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant