CN117493662A - Personalized recommendation method and system based on federal learning - Google Patents
Personalized recommendation method and system based on federal learning Download PDFInfo
- Publication number
- CN117493662A CN117493662A CN202311302679.5A CN202311302679A CN117493662A CN 117493662 A CN117493662 A CN 117493662A CN 202311302679 A CN202311302679 A CN 202311302679A CN 117493662 A CN117493662 A CN 117493662A
- Authority
- CN
- China
- Prior art keywords
- user
- model
- product
- recommendation
- federal learning
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000012549 training Methods 0.000 claims description 28
- 230000006399 behavior Effects 0.000 claims description 8
- 230000002776 aggregation Effects 0.000 claims description 6
- 238000004220 aggregation Methods 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000009826 distribution Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000003213 activating effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000013604 expression vector Substances 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
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
- 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
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/098—Distributed learning, e.g. federated learning
Abstract
The invention discloses a personalized recommendation method and a personalized recommendation system based on federal learning, which belong to the technical field of communication. In addition, the recommendation system also protects the privacy of the user to a certain extent because the browsing history of the user is always stored in the client.
Description
Technical Field
The invention relates to the field of data processing, in particular to a personalized recommendation method and system based on federal learning.
Background
With the rise of internet technology and the popularization of smart phones, the number of netizens is rapidly increased, the large data age comes gradually, and how to mine the value of data is becoming more and more important. Privacy protection is becoming increasingly important as the demand for personalized data recommendation systems increases. A good recommendation system must adequately keep track of the user's personalized information and user needs. Only then can highly personalized, accurate and effective recommendations be generated.
With the improvement of user privacy willingness and the implementation of government protection personal privacy policies, the contradiction between the personalized recommendation system and the personal privacy is gradually reflected, in this case, a designer of the recommendation system needs to know which types of data should be regarded as sensitive data, know the basic structure of the recommendation system and related privacy protection methods, and establish an effective privacy protection recommendation system.
Disclosure of Invention
The invention overcomes the defects of the prior art and makes the following improvements and optimizations aiming at the defects of the prior art.
The aim of the invention is achieved by the following technical scheme:
in one aspect, the invention provides a personalized recommendation method based on federal learning, comprising the following steps:
s1, confirming electronic identity information of a user;
s2, performing an offline training model according to the browsing record of the user;
s3, obtaining a new model after multiple iterations, and entering a model deployment stage;
s4, recommending online products according to the deployed model.
Preferably, in the step S1, the step of confirming the electronic identity information of the user includes the step of logging in the terminal by an account password, taking the user as first authentication information, and generating a biometric identification code by collecting biometric information of the user and based on the biometric information; comparing the biological characteristic identification code with a prestored biological characteristic identification code, and determining an authentication result of the user.
Preferably, in the step S1, after the electronic identity information of the user is confirmed, the relevant behavior information of the user' S attention product data and the intention product data is established according to the past attention product of the user and the behavior information generated when the user browses the product currently.
Preferably, in the step S2, if the user is newly registered, a batch of products is randomly pushed and recommended to the user for reading, and after a certain amount of data is accumulated in both the user side and the server, an offline training model is performed.
More preferably, the offline training model specifically includes:
initializing model parameters, then distributing the parameters of the model to a user terminal, and selecting proper super parameters during distribution; the user terminal starts local training after receiving the model, the data of the user are always kept in the local in the training process, after the loss gradient of the training is calculated, the user terminal uploads the loss gradient to the server, the server performs aggregation calculation according to the gradient in a safe aggregation mode, updates the global recommendation model, and then sends the model to the user terminal again, and a federal recommendation model is obtained after multiple iterations.
Preferably, after the federal recommendation model is obtained, the user terminal can select whether to retrain according to the local browsing history data, so as to obtain a new personalized federal recommendation model.
Preferably, in the step S4, when the user terminal sends a request for product recommendation to the central server, a part of products are extracted from the product database, the central server calculates the scores of the products by using the global model to generate a recall list, the recall list is returned to the user, and the local model orders and recommends the recall list to the user from high to low according to the scores by predicting the scores of the products of the recall list.
On the other hand, the invention also provides a personalized recommendation system based on federal learning, which comprises an information authentication module, an offline training module, a model deployment module and a product recommendation module which are connected in sequence;
the information authentication module is used for acquiring electronic identity information of a user; authenticating the electronic identity information;
the offline training module is used for performing an offline training model according to the browsing record of the user;
the model deployment module is used for preparing products according to the new model;
the product recommendation is used for online product recommendation according to the deployed model.
The invention provides a personalized recommendation method and a personalized recommendation system based on federal learning, which overcome the centralized storage of user data and protect the privacy of users by utilizing a federal learning framework. The federal recommendation model often brings huge communication load, and when the local model is too large, the client lacks resources required by calculation, the paper decomposes the product model into a large product model stored in the server and a small product representation shared by the client and the server, so that the problems of high communication cost and resource shortage of the client are solved, the effectiveness of a product recommendation algorithm designed in the text is verified through experiments, and the system designed by the invention is proved to be capable of realizing personalized recommendation for the user according to the behavior data of the user. In addition, the recommendation system also protects the privacy of the user to a certain extent because the browsing history of the user is always stored in the client.
Drawings
The invention will be further described with reference to the accompanying drawings, in which embodiments do not constitute any limitation of the invention, and other drawings can be obtained by one of ordinary skill in the art without inventive effort from the following drawings.
FIG. 1 is a schematic flow chart of a personalized recommendation method based on federal learning;
FIG. 2 is a block diagram of a personalized recommendation system based on federal learning;
FIG. 3 is a flow chart of the offline model training of the present invention;
FIG. 4 is a schematic diagram of a system login interface of the present invention;
FIG. 5 is a schematic diagram of an output recommended product of a preferred embodiment of the invention;
FIG. 6 is an overall block diagram of a product recommendation model of the present invention.
Detailed Description
A federal learning-based personalized recommendation method and system is described in further detail below in conjunction with specific embodiments, which are provided for comparison and explanation purposes only, and the present invention is not limited to these embodiments.
In the description of the present invention, it should be understood that the directions or positional relationships indicated by the terms "upper", "lower", "left", "right", "top", "bottom", etc. are based on the directions or positional relationships shown in the drawings, are merely for convenience of describing the present invention and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention.
As shown in fig. 1, in one aspect, the present invention provides a personalized recommendation method based on federal learning, including the following steps:
s1, confirming electronic identity information of a user;
s2, performing an offline training model according to the browsing record of the user;
s3, obtaining a new model after multiple iterations, and entering a model deployment stage;
s4, recommending online products according to the deployed model.
Preferably, in the step S1, the step of confirming the electronic identity information of the user includes the step of logging in the terminal by an account password, taking the user as first authentication information, and generating a biometric identification code by collecting biometric information of the user and based on the biometric information; comparing the biological characteristic identification code with a prestored biological characteristic identification code, and determining an authentication result of the user.
The user login interface is shown in fig. 4, and the user can check the product list after logging in, and the login information comprises a user name and a password. In addition, the login interface also has the function of remembering my, and after the login is successful, the user accesses the website again within a certain period without re-login.
Preferably, in the step S1, after the electronic identity information of the user is confirmed, the relevant behavior information of the user' S attention product data and the intention product data is established according to the past attention product of the user and the behavior information generated when the user browses the product currently.
Personal information and log information of the user are stored in a MySQL database, and a system administrator can manage the information of the user. The central server stores the basic information of the product in its own database. In addition, the user terminal and the server need to communicate to realize the transmission of the loss gradient and the model parameters, the global model of the system transmits the updated model parameters to the local of the client, the user terminal uploads the locally trained loss gradient to the server, and when a recommendation list is obtained, the federal recommendation module is responsible for coordinating the interactive recommendation of the client and the central server.
Preferably, in the step S2, if the user is newly registered, a batch of products is randomly pushed and recommended to the user for reading, and after a certain amount of data is accumulated in both the user side and the server, an offline training model is performed.
When registering, the user needs to set the user name and password of the user account and add the mailbox for activating the service. After the user confirms the own password, submitting the registration information form, verifying whether the account number exists or not by a system administrator, if not, sending an activation mail, and clicking an activation link in the mail by the user to activate the account. After the account is activated, the user can enter the login interface to log in the system by means of the information filled during registration.
After entering the product recommendation system, the user can obtain a list of product recommendations and view own browsing history. For the first logged-in user, the system does not have any record about the user, and the interest of the user is difficult to judge, so that the system can randomly select a batch of different types of products to push to the user, and the cold start problem is relieved.
More preferably, the offline training model specifically includes:
as shown in fig. 3, firstly, initializing model parameters, then distributing the model parameters to a user terminal, and selecting proper super parameters during distribution; the user terminal starts local training after receiving the model, the data of the user are always kept in the local in the training process, after the loss gradient of the training is calculated, the user terminal uploads the loss gradient to the server, the server performs aggregation calculation according to the gradient in a safe aggregation mode, updates the global recommendation model, and then sends the model to the user terminal again, and a federal recommendation model is obtained after multiple iterations.
As shown in fig. 6, a product model is designed first, a product is modeled, and then interests of a user are modeled according to the product model. The product Model (New Model) takes candidate products and click history products of users as input to obtain representations of the products, and then the user Model (UserModel) takes a click product representation sequence as input to obtain representations of user interests. Finally, a Click score Predictor (Click Predictor) takes the candidate product representation and the user interest representation as inputs, and outputs a prediction score of the user for the candidate product.
Considering the sensitivity of user data, a federal learning framework is introduced, and a federal learning-based product recommendation model (Federated News Recommendation Model, fedNTM) is designed. The model realizes the decentralization of user data, saves the user data on a local client, trains the client locally, transmits the loss gradient to a server, aggregates the gradient and updates the global model by the server, and keeps a large-scale product model at the server, only keeps part of product representation models at the client, thereby reducing the communication cost.
Preferably, after the federal recommendation model is obtained, the user terminal can select whether to retrain according to the local browsing history data, so as to obtain a new personalized federal recommendation model.
Preferably, in the step S4, when the user terminal sends a request for product recommendation to the central server, a part of products are extracted from the product database, the central server calculates the scores of the products by using the global model to generate a recall list, the recall list is returned to the user, and the local model orders and recommends the recall list to the user from high to low according to the scores by predicting the scores of the products of the recall list.
And calculating a loss function and a gradient according to the predicted score and the actual click condition of the user, and then correspondingly adjusting parameters of the model according to an optimization algorithm such as gradient descent. If the user clicks on a product, the user product pair will get a higher score. If the user does not click on a product, the score of the product by the user is correspondingly reduced after the parameter of the model is adjusted.
Predicting the scores of the users on the candidate products through a click score prediction model, pushing the products with the scores ranked at the front to the users according to the predicted scores, and assuming that the expression vector of the candidate product N isUser u has a representation vector of +.>Then the score S for candidate product N for user u is:
where T represents the matrix.
The personalized recommendation list of the user is shown in fig. 5, and after the user clicks and reads the full text, the user can check the complete information of the product. In this embodiment, it can be seen from the recommendation list in the figure that most of the recommended products for the user are food products, telephone fee products and travel products, the recommendation system uses the behavior data of the user to characterize the interests of the user, then makes product recommendation, and the user model captures the interests of the user to concentrate on the products of the types.
As shown in fig. 2, in another aspect, the invention further provides a personalized recommendation system based on federal learning, which comprises an information authentication module, an offline training module, a model deployment module and a product recommendation module which are sequentially connected;
the information authentication module is used for acquiring electronic identity information of a user; authenticating the electronic identity information;
the offline training module is used for performing an offline training model according to the browsing record of the user;
the model deployment module is used for preparing products according to the new model;
the product recommendation is used for online product recommendation according to the deployed model.
Finally, it should be noted that the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the scope of the present invention, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made to the technical solution of the present invention without departing from the spirit and scope of the technical solution of the present invention.
Claims (8)
1. A personalized recommendation method based on federal learning is characterized by comprising the following steps:
s1, confirming electronic identity information of a user;
s2, performing an offline training model according to the browsing record of the user;
s3, obtaining a new model after multiple iterations, and entering a model deployment stage;
s4, recommending online products according to the deployed model.
2. The personalized recommendation method based on federal learning according to claim 1, wherein in S1, the step of confirming the electronic identity information of the user includes the step of logging in the terminal through an account password, taking the account password as first authentication information, and generating a biometric identification code by collecting biometric information of the user and based on the biometric information; comparing the biological characteristic identification code with a prestored biological characteristic identification code, and determining an authentication result of the user.
3. The personalized recommendation method based on federal learning according to claim 1, wherein in S1, after confirming the electronic identity information of the user, the relevant behavior information of the user' S attention product data and the intention product data is established according to the past attention product of the user and the behavior information generated when the user browses the product currently.
4. The personalized recommendation method based on federal learning according to claim 1, wherein in S2, if the user is newly registered, a batch of products is randomly pushed to be recommended to the user for reading, and after a certain amount of data is accumulated in both the user side and the server, an offline training model is performed.
5. The personalized recommendation method based on federal learning according to claim 4, wherein the offline training model specifically comprises:
initializing model parameters, then distributing the parameters of the model to a user terminal, and selecting proper super parameters during distribution; the user terminal starts local training after receiving the model, the data of the user are always kept in the local in the training process, after the loss gradient of the training is calculated, the user terminal uploads the loss gradient to the server, the server performs aggregation calculation according to the gradient in a safe aggregation mode, updates the global recommendation model, and then sends the model to the user terminal again, and a federal recommendation model is obtained after multiple iterations.
6. The personalized recommendation method based on federal learning according to claim 4, wherein after obtaining a federal recommendation model, the user terminal can choose whether to retrain according to local browsing history data to obtain a new personalized federal recommendation model.
7. The personalized recommendation method based on federal learning according to claim 1, wherein in S4, when the user terminal sends a request for product recommendation to the central server, a part of products are extracted from the product database, the central server calculates the scores of the products by using the global model to generate a recall list, and returns the recall list to the user, and the local model predicts the scores of the products of the recall list and ranks and recommends the recall list to the user according to the score from high to low.
8. The personalized recommendation system based on federal learning is characterized by comprising an information authentication module, an offline training module, a model deployment module and a product recommendation module which are connected in sequence;
the information authentication module is used for acquiring electronic identity information of a user; authenticating the electronic identity information;
the offline training module is used for performing an offline training model according to the browsing record of the user;
the model deployment module is used for preparing products according to the new model;
the product recommendation is used for online product recommendation according to the deployed model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311302679.5A CN117493662A (en) | 2023-10-09 | 2023-10-09 | Personalized recommendation method and system based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311302679.5A CN117493662A (en) | 2023-10-09 | 2023-10-09 | Personalized recommendation method and system based on federal learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117493662A true CN117493662A (en) | 2024-02-02 |
Family
ID=89667976
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311302679.5A Pending CN117493662A (en) | 2023-10-09 | 2023-10-09 | Personalized recommendation method and system based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117493662A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079022A (en) * | 2019-12-20 | 2020-04-28 | 深圳前海微众银行股份有限公司 | Personalized recommendation method, device, equipment and medium based on federal learning |
| 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 |
| CN114625976A (en) * | 2022-05-16 | 2022-06-14 | 深圳市宏博信息科技有限公司 | Data recommendation method, device, equipment and medium based on federal learning |
| CN114741611A (en) * | 2022-06-08 | 2022-07-12 | 杭州金智塔科技有限公司 | Federal recommendation model training method and system |
-
2023
- 2023-10-09 CN CN202311302679.5A patent/CN117493662A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111079022A (en) * | 2019-12-20 | 2020-04-28 | 深圳前海微众银行股份有限公司 | Personalized recommendation method, device, equipment and medium based on federal learning |
| 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 |
| CN114625976A (en) * | 2022-05-16 | 2022-06-14 | 深圳市宏博信息科技有限公司 | Data recommendation method, device, equipment and medium based on federal learning |
| CN114741611A (en) * | 2022-06-08 | 2022-07-12 | 杭州金智塔科技有限公司 | Federal recommendation model training method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110363305B (en) | Federal learning method, system, terminal device and storage medium | |
| CN102855535B (en) | Social influencers discovery | |
| US8543675B1 (en) | Consistent link sharing | |
| CN108520470B (en) | Method and apparatus for generating user attribute information | |
| CN107220386A (en) | Information-pushing method and device | |
| CN108363811A (en) | Device identification method and device, electronic equipment, storage medium | |
| CN107908789A (en) | Method and apparatus for generating information | |
| CN101836185A (en) | Real-time interactive authorization for enterprise search | |
| CN103562925B (en) | Information processing system, information processing method and information processor | |
| CN102081617A (en) | Online consultation method and system | |
| EP2850576A1 (en) | Multi-channel customer identification | |
| CN110020022B (en) | Data processing method, device, equipment and readable storage medium | |
| EP3830723B1 (en) | Increasing security of a password-protected resource based on publicly available data | |
| CN111026858A (en) | Project information processing method and device based on project recommendation model | |
| CN114722281B (en) | Training course configuration method and device based on user portrait and user course selection behavior | |
| CN112989186B (en) | Information recommendation model training method and device, electronic equipment and storage medium | |
| CN109871673A (en) | Based on the lasting identity identifying method and system in different context environmentals | |
| CN113326991A (en) | Automatic authorization method, device, computer equipment and storage medium | |
| CN112507230A (en) | Webpage recommendation method and device based on browser, electronic equipment and storage medium | |
| CN107735785A (en) | Automated information retrieval | |
| CN109949175B (en) | User attribute inference method based on collaborative filtering and similarity measurement | |
| CN117493662A (en) | Personalized recommendation method and system based on federal learning | |
| CN110929526A (en) | Sample generation method and device and electronic equipment | |
| EP3827362A1 (en) | Web browser incorporating social and community features | |
| CN111523034B (en) | Application processing method, device, equipment and medium |
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 |