CN112836130A - Context-aware recommendation system and method based on federated learning - Google Patents
Context-aware recommendation system and method based on federated learning Download PDFInfo
- Publication number
- CN112836130A CN112836130A CN202110192913.8A CN202110192913A CN112836130A CN 112836130 A CN112836130 A CN 112836130A CN 202110192913 A CN202110192913 A CN 202110192913A CN 112836130 A CN112836130 A CN 112836130A
- Authority
- CN
- China
- Prior art keywords
- local
- model
- user
- recommendation
- recommendation model
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000012545 processing Methods 0.000 claims abstract description 12
- 230000008569 process Effects 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000002776 aggregation Effects 0.000 claims description 4
- 238000004220 aggregation Methods 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000004931 aggregating effect Effects 0.000 claims description 3
- 230000009191 jumping Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 8
- 238000002474 experimental method Methods 0.000 description 8
- 238000011156 evaluation Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010348 incorporation Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003997 social interaction Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- 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
- G06N20/00—Machine learning
-
- 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]
- G06Q30/0623—Item investigation
- G06Q30/0625—Directed, with specific intent or strategy
- G06Q30/0629—Directed, with specific intent or strategy for generating comparisons
-
- 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)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Accounting & Taxation (AREA)
- Finance (AREA)
- Software Systems (AREA)
- Development Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Health & Medical Sciences (AREA)
- Bioethics (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Economics (AREA)
- Computer Hardware Design (AREA)
- Computer Security & Cryptography (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a context awareness recommendation system and method based on federal learning, which provides a user-defined data cooperation protocol module, splices user scoring data and context information through the module to obtain user data containing the context information and sends the user data to a client. The client trains a local recommendation model according to the user data and the server weight parameters, and sends the local model weight parameters to the central server. And the central server aggregates the local model weight parameters of all the clients to obtain new server weight parameters, and completes a round of training. And training the local recommendation model for multiple times until the local recommendation model converges. And the trained local recommendation model is used for processing the new user data to obtain the recommendation content of the user. According to the invention, the federal learning technology is applied to the recommendation system, and the context information is combined in the recommendation system, so that the higher recommendation accuracy is realized while the user privacy is protected.
Description
Technical Field
The invention relates to the field of recommendation systems, in particular to a context-aware recommendation system and method based on federal learning.
Background
In the field of recommendation systems, the traditional privacy protection methods mainly include data anonymization, encryption technology and differential privacy. The data anonymization technology is a method for carrying out anonymization processing on data in a data set, firstly, attributes of each piece of data are divided into 3 classes, identification attributes of an individual are uniquely identified, and standard identification attributes and sensitive attributes needing to be protected of the individual can be determined by combining the identification attributes. The identification attribute of the data set after anonymization processing is hidden, and meanwhile, the identification attribute is aligned to perform certain fuzzification (such as changing the age into a range, removing the last two digits of a zip code and the like) so that sensitive data is protected. The encryption technology requires a large additional calculation cost, complicated steps and a low overall resource utilization rate. The differential privacy technology is a privacy protection method which is popular in recent years, noise is added into a data set, so that even if the data set is changed in a small range, the probability distribution of result values obtained by inquiring the data set is basically unchanged, and information leakage caused by differential attack is effectively prevented. However, similar to data anonymization, differential privacy requires a large amount of randomization to be added to the query results, resulting in a drastic drop in data availability and a loss of information contained in the data.
The traditional privacy protection method plays a certain role in protecting user data, but when the methods are applied to a recommendation system, huge calculation overhead is caused, the accuracy of results is greatly influenced, and the training iterative upgrade process of artificial intelligence algorithms such as a recommendation algorithm cannot be well fitted. Federal learning, which is rapidly developed in recent years, is just a training framework suitable for privacy protection in deep learning. It was proposed by google researcher McMahan et al in 2016. Initially used to train a handset input prediction model without the user data leaving the home. Later, its central idea was refined and called federal learning. The Federal learning assumption comprises a plurality of training participants and a central server, the central server distributes the model to each participant, and the participants only need to train locally and return updated model parameters to the central server to complete a model updating process. Federal learning ensures that user data does not leave the user local storage device, and the privacy protection capability is greatly improved.
In addition, in the field of recommendation systems, context-aware recommendation systems have gained widespread attention. It is noted that the user's selection is not only dependent on people with similar interests, but also closely related to the context information when selecting a particular item, such as time, location, social interaction, etc. when selecting. Working as an early proxy, 'Adomavicius G, Sankaranarayana R, Sen S, et al, incorporation textual Information in recommendation Systems using a multidimensional analysis of Information J. ACM Transformations On Information Systems (TOIS),2005,23(1):103-145,' introduces the importance of contextual Information and a multidimensional method to incorporate contextual Information into the recommendation process. Meanwhile, the importance of the time dynamics of the data set to the recommendation result is gradually noticed, and some work of integrating the time features into the recommendation model also achieves good results. Recently, many researchers have taken advantage of deep learning to make context-based recommendations. In addition to the importance of contextual information in deep learning models, privacy disclosure is a major drawback from the user's perspective. Serious consequences can arise if the recommender system has full access to all of the contextual attributes and the attackers can collectively obtain or infer such information.
Therefore, applying the federal learning framework in recommendation system algorithms to improve privacy protection is a natural idea. The early representatives worked on the federal collaborative filtering recommendation model, which was thought to be very simple, separating the item vector from the user vector, with the item vector being located on a central server and the user vector being stored locally on the user. Subsequently, there are also algorithms based on meta-learning to make joint recommendations. Recently, the paper 'Tan B, Liu B, Zheng V, et al.A Federated Recommender System for on-line Services [ C ]// Fourtenth ACM Conference on Recommender systems.2020:579 @ 581' proposes a Federated recommendation System for Online Services that trains recommendation models based on data from multiple parties without exposing personal information of each party. The model includes a data layer, an algorithm layer, a service layer, and an interface layer. Authors also discuss various online applications such as content recommendations, product recommendations, and online advertising through the proposed model. Meanwhile, the paper ' Muhammad K, Wang Q, O ' Reilly-Morgan D, et al.Fedfast: Going beyond average for failure training of fed recognized systems [ C ]// Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data mining.2020:1234-1242 ' proposes another technique to accelerate the distributed learning of recommended tasks. It completes the early training process by sampling from a group of different customers in each training run and applying active aggregation to propagate the updated model to other customers. While some existing approaches have utilized federal learning to improve the reliability of recommendations, they do not make good use of contextual information.
Disclosure of Invention
Aiming at the defects in the prior art, the context-aware recommendation system and method based on the federal learning provided by the invention solve the problem that the existing recommendation system using the federal learning framework does not well utilize the context information by applying the federal learning technology to the recommendation system and combining the context information in the recommendation system, and the invention realizes that the privacy of users is protected and simultaneously has higher recommendation accuracy.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that: a federal learning based context aware recommendation system comprising: a central server and a plurality of clients;
the central server is respectively connected with a plurality of clients; the central server comprises: a global recommendation model; each of the clients includes thereon: a local recommendation model and a user-defined data cooperation protocol module;
the local recommendation model is used for performing federal learning training by using user data output by a user defined protocol (UDCP) and model weight parameters sent by the global recommendation model to obtain a trained local recommendation model; and the trained local recommendation model is used for processing new user data to obtain the recommendation content of the user.
Further, the user-defined data cooperation protocol module is used for splicing the user context information, the article context information, the scoring history and the time records of the scoring history to obtain user data containing the context information, a user can freely select whether the spliced data contains the user context information or the article context information in the process, and then the module sends the user data to each client during each iteration; each client is used for training a local recommendation model according to the user data and the server model weight parameters sent by the global recommendation model of the central server, completing one-time training of the local recommendation model, and sending the obtained local model weight parameters to the central server; the global recommendation model of the central server is used for aggregating the local model weight parameters sent by all the clients to obtain new server model weight parameters, sending the new server model weight parameters to the local recommendation model, performing secondary training on the local recommendation model, and training the local recommendation model for multiple times until the local recommendation model converges to obtain a trained local recommendation model; and the trained local recommendation model is used for processing new user data to obtain the recommendation content of the user.
The beneficial effects of the above further scheme are: the user-defined data cooperation protocol module effectively utilizes various context information, and associates the context information with the historical scoring behavior data of the user, so that the availability of the user data is greatly enhanced. Meanwhile, the most important point is that the user can control the specific degree of combination of personal data and the recommended model through the user-defined data cooperation protocol module, so that the personal privacy of the user is protected.
Further, the loss function of the local recommendation model is:
wherein L isi(θ) is the average loss function of the ith local recommendation model over all samples of the ith user, DiRepresenting items scored by a user
The total number of (a) and (b),
to input the local recommendation model during the training process,
j is the number of the local data,
the local model weight parameter for the t-th iteration,
is the MSE Loss function.
Further, the local model weight parameter
The calculation formula at each client is:
wherein,
is the directional weight parameter for the t-1 th iteration, alpha is the learning rate,
is a gradient factor, beta is a learning rate,
for a test set containing context-embedded information,
a training set containing context embedded information.
Further, the calculation formula of the server-side model weight parameter is as follows:
wherein,
a server model weight parameter for the t-th iteration, N is the number of clients, WiThe weight occupied by each client is large.
A context-aware recommendation method based on federated learning comprises the following steps:
s1, constructing a user-defined data cooperation protocol module between the local data and the local recommendation model of each client;
s2, splicing user context information, article context information, scoring history and time records of the scoring history in the local data by adopting a user-defined data cooperation protocol module to obtain user data;
s3, distributing the initial value of the server model weight parameter to each client through the central server;
s4, inputting the initial values of the user data and the server model weight parameters into a local recommendation model of the client, and training the local recommendation model;
s5, sending the local model weight parameters obtained by training on the local recommendation model back to the central server;
s6, receiving the local model weight parameters sent back by all the clients through the central server, carrying out aggregation processing on all the local model weight parameters through a weighting method based on the global recommendation model to obtain cached server model weight parameters, and taking the cached server model weight parameters as new server model weight parameter initial values;
s7, jumping to the step S3, and circularly executing the step S3 to the step S6 until the local recommendation model converges, and ending the federal learning process to obtain a trained local recommendation model;
and S8, inputting the new user data into the trained local recommendation model to obtain the recommendation content of the user.
In conclusion, the beneficial effects of the invention are as follows:
(1) the traditional recommendation system is combined with federal learning which has great effect on user privacy protection, and the recommendation system using context sensing is built under a federal learning framework, so that personal data of a user can be stored locally, a recommendation model provider does not need to upload the personal data to complete a training process, and a result with the same recommendation accuracy as that of the traditional method or even better can be obtained.
(2) The user-defined data collaboration protocol module can combine various context information of the user with data of the user, and usability of the user data is enhanced to the great extent. The user can control the specific degree of combination of personal data and the recommendation model through the user-defined data cooperation protocol module, so that the personal privacy of the user is protected, the initiative of the user on the use of the personal data is enhanced, and the use of the personal data in the recommendation process is clearer and more reasonable.
Drawings
FIG. 1 is a system block diagram of a context-aware recommendation system based on federated learning;
FIG. 2 is a connection diagram of a user-defined data collaboration protocol module;
FIG. 3 is a flow chart of a context-aware recommendation method based on federated learning;
FIG. 4(a) is a comparison graph of the mean absolute error MAE obtained by the experiment using the data set Ml-100K;
FIG. 4(b) is a comparison graph of the mean absolute error MAE obtained by the data set Ml-latest-small experiment;
FIG. 4(c) is a comparison of the root mean square error RMSE obtained using the Ml-100K data set experiment;
FIG. 4(d) is a graph comparing the root mean square error RMSE obtained from the data set Ml-latest-small experiment;
FIG. 4(e) is a comparison graph of nDCG indexes obtained by the experiment using the data set Ml-100K;
FIG. 4(f) is a comparison graph of nDCG indexes obtained by the data set Ml-latest-small experiment.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
1-2, a context-aware recommendation system based on federated learning includes: a central server and a plurality of clients;
the central server is respectively connected with a plurality of clients; the central server comprises: a global recommendation model; each of the clients includes thereon: a local recommendation model and a user-defined data cooperation protocol module;
the local recommendation model is used for performing federal learning training by using user data output by a user defined protocol (UDCP) and model weight parameters sent by the global recommendation model to obtain a trained local recommendation model; and the trained local recommendation model is used for processing new user data to obtain the recommendation content of the user.
The user-defined data cooperation protocol module is used for splicing the user context information, the article context information, the grading history and the time records of the grading history to obtain user data containing the context information, a user can freely select whether the spliced data contains the user context information or the article context information in the process, and then the module sends the user data to each client side during each iteration; each client is used for training a local recommendation model according to the user data and the server model weight parameters sent by the global recommendation model of the central server, completing one-time training of the local recommendation model, and sending the obtained local model weight parameters to the central server; the global recommendation model of the central server is used for aggregating the local model weight parameters sent by all the clients to obtain new server model weight parameters, sending the new server model weight parameters to the local recommendation model, performing secondary training on the local recommendation model, and training the local recommendation model for multiple times until the local recommendation model converges to obtain a trained local recommendation model; and the trained local recommendation model is used for processing new user data to obtain the recommendation content of the user.
The user-defined data cooperation protocol module effectively utilizes various context information, and associates the context information with the historical scoring behavior data of the user, so that the availability of the user data is greatly enhanced. Meanwhile, the most important point is that the user can control the specific degree of combination of personal data and the recommendation model through the user-defined data cooperation protocol module, so that the personal privacy of the user is protected.
The penalty function for the local recommendation model is:
wherein L isi(θ) is the average loss function of the ith local recommendation model over all samples of the ith user, DiRepresenting items scored by a user
The total number of (a) and (b),
to input the local recommendation model during the training process,
j is the number of the local data,
the local model weight parameter for the t-th iteration,
is the MSE Loss function.
Local model weight parameters
The calculation formula at each client is:
wherein,
is the directional weight parameter for the t-1 th iteration, alpha is the learning rate,
is a gradient factor, beta is a learning rate,
for a test set containing context-embedded information,
a training set containing context embedded information.
The calculation formula of the server-side model weight parameter is as follows:
wherein,
a server model weight parameter for the t-th iteration, N is the number of clients, WiThe weight occupied by each client is large.
As shown in fig. 3, a context-aware recommendation method based on federal learning includes the following steps:
s1, constructing a user-defined data cooperation protocol module between the local data and the local recommendation model of each client;
s2, splicing user context information, article context information, scoring history and time records of the scoring history in the local data by adopting a user-defined data cooperation protocol module to obtain user data;
s3, distributing the initial value of the server model weight parameter to each client through the central server;
s4, inputting the initial values of the user data and the server model weight parameters into a local recommendation model of the client, and training the local recommendation model;
s5, sending the local model weight parameters obtained by training on the local recommendation model back to the central server;
s6, receiving the local model weight parameters sent back by all the clients through the central server, carrying out aggregation processing on all the local model weight parameters through a weighting method based on the global recommendation model to obtain cached server model weight parameters, and taking the cached server model weight parameters as new server model weight parameter initial values;
s7, jumping to the step S3, and circularly executing the step S3 to the step S6 until the local recommendation model converges, and ending the federal learning process to obtain a trained local recommendation model;
and S8, inputting the new user data into the trained local recommendation model to obtain the recommendation content of the user.
Experiment:
the effect of the invention is verified on two public data sets Ml-100K and Ml-latest-small.
Ml-latest-small dataset: the data set contains 670 users' scores for 9742 movies, ranging from 0.5 to 5, with a score interval of 0.5.
The Ml-100K data set contained 100000 scores provided by 943 users for 1682 different movies. Each user rated at least 20 movies on a scale of 1 to 5. In addition, user information (such as age, gender, occupation) and item information (such as genre, release date, and IMDb URL) for the movie are provided.
In order to illustrate the superiority of the method (Feded learning based Context-Aware Recommendar System, hereinafter referred to as Fed-CARS) provided by the invention in comparison with the prediction accuracy and ranking performance of other common recommendation System technologies, the method is evaluated by using open source software CaseRecommendar. The application mainly adopts the following recommendation system technologies as comparison baselines:
BPR: bayes personalized ranking is an algorithm designed for implicit feedback, and a potential factor model is optimized by adopting pairwise ranking loss.
POP: the most popular recommendation algorithms are recommendation techniques that are very common in the e-commerce field and medium-scale recommendation engines. It predicts the ranking of users based on their popularity and the items.
UserKNN: a well-known collaborative filtering technique based on user similarity is used to find nearest neighbor users whose favorite items are recommended to the users.
ItemKNN: a well-known collaborative filtering technique based on item similarity is used to find nearest neighbor items to recommend to a user.
PaCo: a recommendation algorithm based on joint clustering has high expandability and anti-noise capability. It is an extension of k-means and agglomerative hierarchical clustering methods.
For the evaluation index of the algorithm, the application decides to adopt the average absolute error (MAE) and the Root Mean Square Error (RMSE) which are widely applied. In addition, in order to more accurately measure the performance of the algorithm in the aspect of ranking evaluation, normalized discounted cumulative gain (nDCG), a performance evaluation method widely used in a recommendation system, is used. The equations for Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are as follows:
MAE estimates the distance between the predicted ranking PR and the actual ranking AR of each user item pairAverage absolute deviation of (d). Here, PRuiRepresenting a predicted ranking, AR, of user u for item iuiThe actual ranking is represented. RMSE reflects the degree of deviation between the estimated and actual rankings. The error is squared before it is summed, thus penalizing more for large deviations. The lower these two metrics reflect the higher the prediction accuracy of the recommendation algorithm.
The nDCG index is obtained by calculating the ranking of the articles i on n positions and then adding the ranking to the reciprocal of the articles i. Here, N is the number of instances of the ranking that each user in the query set considers as top-N, and N is the number of users in the test data set. The factor mi is 1 if the item is present in the appropriate position in the predicted top n list, and 0 otherwise. R is based on a truly ordered list of recommended items, which means that all truly recommended or purchased items have a smaller index value than non-purchased items. The divided IDCG is ideally the maximum DCG value, which is calculated as follows:
the final experimental results are shown in table 1 below:
TABLE 1
The bold type indicates that each item works best, and the underlined value indicates that each item works second best. Compared with some traditional or widely-used methods at present, the method provided by the invention has good performance on the precision of the recommendation result. Furthermore, it is also emphasized that the present invention is implemented in a manner that preserves user privacy.
Meanwhile, the utility of the user-defined data cooperation protocol module is verified. This experiment included 3 cases: the first is the invention (symbol: Fed-CARS), the second is the invention which adopts user context information, article context information, time record of scoring history and scoring history, the third is the invention which adopts single user context information (symbol: Fed-CARS-user), and the third is the invention which adopts single article context information (symbol: Fed-CARS-user). The curve trends in fig. 4(a) -4 (f) show that the Fed-CARS with both user context and commodity context achieves lower error rates on both MAE and RMSE indices in all training rounds of model update. In terms of ranking performance, Fed-CARS has slightly improved nDCG values compared to the version without context information.
Claims (6)
1. A federated learning-based context-aware recommendation system, comprising: a central server and a plurality of clients;
the central server is respectively connected with a plurality of clients; the central server comprises: a global recommendation model; each of the clients includes thereon: a local recommendation model and a user-defined data cooperation protocol module;
the local recommendation model is used for performing federal learning training by using user data output by the user-defined data cooperation protocol module and the server model weight parameters sent by the global recommendation model to obtain a trained local recommendation model; and the trained local recommendation model is used for processing new user data to obtain the recommendation content of the user.
2. The context-aware recommendation system based on federated learning of claim 1, wherein the user-defined data collaboration protocol module is configured to splice user context information, item context information, score history and time records of the score history to obtain user data containing the context information, a user can freely select whether the spliced data contains the user context information or the item context information during the process, and then the module sends the user data to each client during each iteration; each client is used for training a local recommendation model according to the user data and the server model weight parameters sent by the global recommendation model of the central server, completing one-time training of the local recommendation model, and sending the obtained local model weight parameters to the central server; the global recommendation model of the central server is used for aggregating the local model weight parameters sent by all the clients to obtain new server model weight parameters, sending the new server model weight parameters to the local recommendation model, performing secondary training on the local recommendation model, and training the local recommendation model for multiple times until the local recommendation model converges to obtain a trained local recommendation model; and the trained local recommendation model is used for processing new user data to obtain the recommendation content of the user.
3. The federated learning-based context-aware recommendation system of claim 2, wherein the loss function of the local recommendation model is:
wherein L isi(θ) is the average loss function of the ith local recommendation model over all samples of the ith user, DiRepresenting items scored by a user
The total number of (a) and (b),
to input the local recommendation model during the training process,
j is the number of the local data,
the local model weight parameter for the t-th iteration,
is the MSE Loss function.
4. The federated learning-based context-aware recommendation system of claim 3, wherein the local model weight parameter
The calculation formula at each client is:
wherein,
is the directional weight parameter for the t-1 th iteration, alpha is the learning rate,
is a gradient factor, beta is a learning rate,
for a test set containing context-embedded information,
a training set containing context embedded information.
5. The context-aware recommendation system based on federated learning according to claim 4, wherein the calculation formula of the server-side model weight parameter is:
wherein,
a server model weight parameter for the t-th iteration, N is the number of clients, WiThe weight occupied by each client is large.
6. A context-aware recommendation method based on federated learning is characterized by comprising the following steps:
s1, constructing a user-defined data cooperation protocol module between the local data and the local recommendation model of each client;
s2, splicing user context information, article context information, scoring history and time records of the scoring history in the local data by adopting a user-defined data cooperation protocol module to obtain user data;
s3, distributing the initial value of the server model weight parameter to each client through the central server;
s4, inputting the initial values of the user data and the server model weight parameters into a local recommendation model of the client, and training the local recommendation model;
s5, sending the local model weight parameters obtained by training on the local recommendation model back to the central server;
s6, receiving the local model weight parameters sent back by all the clients through the central server, carrying out aggregation processing on all the local model weight parameters through a weighting method based on the global recommendation model to obtain cached server model weight parameters, and taking the cached server model weight parameters as new server model weight parameter initial values;
s7, jumping to the step S3, and circularly executing the step S3 to the step S6 until the local recommendation model converges, and ending the federal learning process to obtain a trained local recommendation model;
and S8, inputting the new user data into the trained local recommendation model to obtain the recommendation content of the user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110192913.8A CN112836130B (en) | 2021-02-20 | 2021-02-20 | Context-aware recommendation system and method based on federated learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110192913.8A CN112836130B (en) | 2021-02-20 | 2021-02-20 | Context-aware recommendation system and method based on federated learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112836130A true CN112836130A (en) | 2021-05-25 |
| CN112836130B CN112836130B (en) | 2023-02-03 |
Family
ID=75933955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110192913.8A Active CN112836130B (en) | 2021-02-20 | 2021-02-20 | Context-aware recommendation system and method based on federated learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112836130B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113420232A (en) * | 2021-06-02 | 2021-09-21 | 杭州电子科技大学 | Privacy protection-oriented graph neural network federal recommendation method |
| CN113570071A (en) * | 2021-08-09 | 2021-10-29 | 山东产业技术研究院智能计算研究院 | Method and system for publishing federated learning model service |
| CN113626687A (en) * | 2021-07-19 | 2021-11-09 | 浙江师范大学 | Online course recommendation method and system taking federal learning as core |
| CN113761371A (en) * | 2021-09-08 | 2021-12-07 | 东北大学 | Recommendation method based on anonymization privacy protection |
| CN114202397A (en) * | 2022-02-17 | 2022-03-18 | 浙江君同智能科技有限责任公司 | Longitudinal federal learning backdoor defense method based on neuron activation value clustering |
| CN114510646A (en) * | 2022-01-28 | 2022-05-17 | 南京邮电大学 | Neural network collaborative filtering recommendation method based on federal learning |
| CN114662006A (en) * | 2022-05-23 | 2022-06-24 | 阿里巴巴达摩院(杭州)科技有限公司 | End cloud collaborative recommendation system and method and electronic equipment |
| CN114860927A (en) * | 2022-04-18 | 2022-08-05 | 广东工业大学 | Text matching method for multi-task federal learning |
| CN115022316A (en) * | 2022-05-20 | 2022-09-06 | 阿里巴巴（中国）有限公司 | End cloud cooperative data processing system, method, equipment and computer storage medium |
| CN115238065A (en) * | 2022-09-22 | 2022-10-25 | 太极计算机股份有限公司 | Intelligent document recommendation method based on federal learning |
| WO2023066205A1 (en) * | 2021-10-19 | 2023-04-27 | 维沃移动通信有限公司 | Model data sending method and device, and model data integration method and device |
| CN116226540A (en) * | 2023-05-09 | 2023-06-06 | 浙江大学 | End-to-end federation personalized recommendation method and system based on user interest domain |
| CN117494191A (en) * | 2023-10-17 | 2024-02-02 | 南昌大学 | Point-of-interest micro-service system and method for information physical security |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170178080A1 (en) * | 2015-12-17 | 2017-06-22 | International Business Machines Corporation | Machine learning system for intelligently identifying suitable time slots in a user's electronic calendar |
| US20180052898A1 (en) * | 2016-08-22 | 2018-02-22 | Oracle International Corporation | System and method for dynamic, incremental recommendations within real-time visual simulation |
| CN108287904A (en) * | 2018-05-09 | 2018-07-17 | 重庆邮电大学 | A kind of document context perception recommendation method decomposed based on socialization convolution matrix |
| CN111177569A (en) * | 2020-01-07 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Recommendation processing method, device and equipment based on artificial intelligence |
| CN111967615A (en) * | 2020-09-25 | 2020-11-20 | 北京百度网讯科技有限公司 | Multi-model training method and system based on feature extraction, electronic device and medium |
| CN112115321A (en) * | 2019-06-03 | 2020-12-22 | 阿里巴巴集团控股有限公司 | Training method and device of content recommendation model, electronic equipment and storage medium |
| CN112367109A (en) * | 2020-09-28 | 2021-02-12 | 西北工业大学 | Incentive method for digital twin-driven federal learning in air-ground network |
-
2021
- 2021-02-20 CN CN202110192913.8A patent/CN112836130B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170178080A1 (en) * | 2015-12-17 | 2017-06-22 | International Business Machines Corporation | Machine learning system for intelligently identifying suitable time slots in a user's electronic calendar |
| US20180052898A1 (en) * | 2016-08-22 | 2018-02-22 | Oracle International Corporation | System and method for dynamic, incremental recommendations within real-time visual simulation |
| CN108287904A (en) * | 2018-05-09 | 2018-07-17 | 重庆邮电大学 | A kind of document context perception recommendation method decomposed based on socialization convolution matrix |
| CN112115321A (en) * | 2019-06-03 | 2020-12-22 | 阿里巴巴集团控股有限公司 | Training method and device of content recommendation model, electronic equipment and storage medium |
| CN111177569A (en) * | 2020-01-07 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Recommendation processing method, device and equipment based on artificial intelligence |
| CN111967615A (en) * | 2020-09-25 | 2020-11-20 | 北京百度网讯科技有限公司 | Multi-model training method and system based on feature extraction, electronic device and medium |
| CN112367109A (en) * | 2020-09-28 | 2021-02-12 | 西北工业大学 | Incentive method for digital twin-driven federal learning in air-ground network |
Non-Patent Citations (3)
| Title |
|---|
| LIU YANG等: "《Federated Learning》", 26 November 2020 * |
| WAQAR ALI等: "A Federated Learning Approach for Privacy Protection in Context-Aware Recommender Systems", 《THE COMPUTER JOURNAL》 * |
| 陈涛等: "面向大数据隐私保护的联邦学习算法航空应用模型研究", 《信息安全与通信保密》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113420232A (en) * | 2021-06-02 | 2021-09-21 | 杭州电子科技大学 | Privacy protection-oriented graph neural network federal recommendation method |
| CN113420232B (en) * | 2021-06-02 | 2022-05-10 | 杭州电子科技大学 | Privacy protection-oriented federated recommendation method for neural network of graph |
| CN113626687A (en) * | 2021-07-19 | 2021-11-09 | 浙江师范大学 | Online course recommendation method and system taking federal learning as core |
| CN113570071A (en) * | 2021-08-09 | 2021-10-29 | 山东产业技术研究院智能计算研究院 | Method and system for publishing federated learning model service |
| CN113761371A (en) * | 2021-09-08 | 2021-12-07 | 东北大学 | Recommendation method based on anonymization privacy protection |
| CN113761371B (en) * | 2021-09-08 | 2023-08-01 | 东北大学 | Recommendation method based on anonymization privacy protection |
| WO2023066205A1 (en) * | 2021-10-19 | 2023-04-27 | 维沃移动通信有限公司 | Model data sending method and device, and model data integration method and device |
| CN114510646A (en) * | 2022-01-28 | 2022-05-17 | 南京邮电大学 | Neural network collaborative filtering recommendation method based on federal learning |
| CN114202397A (en) * | 2022-02-17 | 2022-03-18 | 浙江君同智能科技有限责任公司 | Longitudinal federal learning backdoor defense method based on neuron activation value clustering |
| CN114202397B (en) * | 2022-02-17 | 2022-05-10 | 浙江君同智能科技有限责任公司 | Longitudinal federal learning backdoor defense method based on neuron activation value clustering |
| CN114860927A (en) * | 2022-04-18 | 2022-08-05 | 广东工业大学 | Text matching method for multi-task federal learning |
| CN114860927B (en) * | 2022-04-18 | 2023-04-25 | 广东工业大学 | Text matching method for multitasking federal learning |
| CN115022316A (en) * | 2022-05-20 | 2022-09-06 | 阿里巴巴（中国）有限公司 | End cloud cooperative data processing system, method, equipment and computer storage medium |
| CN115022316B (en) * | 2022-05-20 | 2023-08-11 | 阿里巴巴（中国）有限公司 | End cloud collaborative data processing system, method, equipment and computer storage medium |
| CN114662006A (en) * | 2022-05-23 | 2022-06-24 | 阿里巴巴达摩院(杭州)科技有限公司 | End cloud collaborative recommendation system and method and electronic equipment |
| CN114662006B (en) * | 2022-05-23 | 2022-09-02 | 阿里巴巴达摩院(杭州)科技有限公司 | End cloud collaborative recommendation system and method and electronic equipment |
| CN115238065B (en) * | 2022-09-22 | 2022-12-20 | 太极计算机股份有限公司 | Intelligent document recommendation method based on federal learning |
| CN115238065A (en) * | 2022-09-22 | 2022-10-25 | 太极计算机股份有限公司 | Intelligent document recommendation method based on federal learning |
| CN116226540A (en) * | 2023-05-09 | 2023-06-06 | 浙江大学 | End-to-end federation personalized recommendation method and system based on user interest domain |
| CN116226540B (en) * | 2023-05-09 | 2023-09-26 | 浙江大学 | End-to-end federation personalized recommendation method and system based on user interest domain |
| CN117494191A (en) * | 2023-10-17 | 2024-02-02 | 南昌大学 | Point-of-interest micro-service system and method for information physical security |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112836130B (en) | 2023-02-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112836130B (en) | Context-aware recommendation system and method based on federated learning | |
| KR102591421B1 (en) | Intent recommendation method, apparatus, device and storage medium | |
| US11593894B2 (en) | Interest recommendation method, computer device, and storage medium | |
| Huang et al. | Social friend recommendation based on multiple network correlation | |
| US20170250930A1 (en) | Interactive content recommendation personalization assistant | |
| US9767417B1 (en) | Category predictions for user behavior | |
| US20210263978A1 (en) | Intelligent interface accelerating | |
| US11423018B1 (en) | Multivariate analysis replica intelligent ambience evolving system | |
| TW201447797A (en) | Method and system for multi-phase ranking for content personalization | |
| KR20100094021A (en) | Customized and intellectual symbol, icon internet information searching system utilizing a mobile communication terminal and ip-based information terminal | |
| Yang et al. | A Time‐Aware CNN‐Based Personalized Recommender System | |
| US10474670B1 (en) | Category predictions with browse node probabilities | |
| Mu et al. | [Retracted] Collaborative Filtering Recommendation Algorithm Based on Knowledge Graph | |
| US11809506B1 (en) | Multivariant analyzing replicating intelligent ambience evolving system | |
| Cong | Personalized recommendation of film and television culture based on an intelligent classification algorithm | |
| CN115689672A (en) | Chat type commodity shopping guide method and device, equipment and medium thereof | |
| CN112818195B (en) | Data acquisition method, device and system and computer storage medium | |
| Cao et al. | Web API recommendation via combining graph attention representation and deep factorization machines quality prediction | |
| Arora et al. | Cross-domain based event recommendation using tensor factorization | |
| Ali et al. | Dynamic context management in context-aware recommender systems | |
| Addagarla et al. | A survey on comprehensive trends in recommendation systems & applications | |
| US10387934B1 (en) | Method medium and system for category prediction for a changed shopping mission | |
| Sun | Music Individualization Recommendation System Based on Big Data Analysis | |
| Liu | POI recommendation model using multi-head attention in location-based social network big data | |
| CN107798045A (en) | User towards middle-size and small-size website, which accesses, is intended to acquisition methods and system |
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 |