CN116502271B - Privacy protection cross-domain recommendation method based on generation model - Google Patents
Privacy protection cross-domain recommendation method based on generation model Download PDFInfo
- Publication number
- CN116502271B CN116502271B CN202310742959.1A CN202310742959A CN116502271B CN 116502271 B CN116502271 B CN 116502271B CN 202310742959 A CN202310742959 A CN 202310742959A CN 116502271 B CN116502271 B CN 116502271B
- Authority
- CN
- China
- Prior art keywords
- domain
- data
- target
- information vector
- vector
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 72
- 239000013598 vector Substances 0.000 claims abstract description 245
- 238000012549 training Methods 0.000 claims abstract description 76
- 238000012545 processing Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 9
- 238000011161 development Methods 0.000 description 4
- 238000013508 migration Methods 0.000 description 4
- 230000005012 migration Effects 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000007774 longterm Effects 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005755 formation reaction Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000004984 smart glass Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- 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
- 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
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Bioethics (AREA)
- General Health & Medical Sciences (AREA)
- Computer Security & Cryptography (AREA)
- Computer Hardware Design (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a privacy protection cross-domain recommendation method based on a generation model, wherein the privacy protection cross-domain recommendation method based on the generation model comprises the following steps: acquiring source domain prediction data and target domain training data, wherein the source domain prediction data is generated by inputting noise data into a generation model; extracting the characteristic vector of the source domain prediction data to obtain a source domain user information vector; extracting feature vectors of the target domain training data to obtain a target domain user information vector and a target domain project information vector; training a recommendation model based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector; the source domain prediction data obtained by the generation model does not directly touch the original information of the user in the source domain, so that the privacy protection of the source domain user data is realized, and the user knowledge of the source domain is transferred to the target domain, so that the recommendation performance of the recommendation model in the target domain is improved.
Description
Technical Field
The application relates to the technical field of computers, in particular to a privacy protection cross-domain recommendation method based on a generation model. The application also relates to an item recommendation method, a privacy protection cross-domain recommendation device based on the generation model, a computing device and a computer readable storage medium.
Background
With the rapid development of the internet, data information in the internet is explosive, and a user is difficult to select interesting items from a plurality of choices in the face of massive data information. In order to improve the user experience, recommendation systems are widely used in different scenarios, such as online shopping, music recommendation, movie recommendation, etc.
The task of cross-domain recommendation generally refers to the overlapping of user formations in two data domains such that there is a certain amount of user data between the different data domains. The current cross-domain recommendation is generally realized by using a cross-domain recommendation model, a target domain can acquire user data of a source domain, and then the cross-domain recommendation model is trained by using the acquired user data of the source domain and user data of a local target domain, so that the trained model can realize the cross-domain recommendation. However, in the current cross-domain recommendation model training, user data are all display interactive, and privacy of the user data is seriously affected.
Disclosure of Invention
In view of this, the embodiment of the application provides a privacy protection cross-domain recommendation method based on a generation model. The application also relates to an item recommending method, a privacy protection cross-domain recommending device based on a generating model, a computing device and a computer readable storage medium, so as to solve the problem of user privacy data leakage caused by the user cross-domain recommending item in the prior art.
According to a first aspect of an embodiment of the present application, there is provided a privacy protection cross-domain recommendation method based on a generation model, including:
acquiring source domain prediction data and target domain training data, wherein the source domain prediction data is generated by inputting noise data into a generation model;
extracting the characteristic vector of the source domain prediction data to obtain a source domain user information vector;
extracting feature vectors of the target domain training data to obtain a target domain user information vector and a target domain project information vector;
a recommendation model is trained based on the cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
According to a second aspect of the embodiment of the present application, there is provided an item recommendation method, including:
acquiring user information of a target user;
and inputting the user information into a recommendation model to obtain project recommendation information for the target user, wherein the recommendation model is obtained by training by using the recommendation model training method.
According to a third aspect of an embodiment of the present application, there is provided a privacy-preserving cross-domain recommendation apparatus based on a generation model, including:
The training data acquisition module is configured to acquire source domain prediction data and target domain training data, wherein the source domain prediction data is acquired by generating a model based on noise data;
the first feature vector extraction module is configured to extract the feature vector of the source domain prediction data to obtain a source domain user information vector;
the second feature vector extraction module is configured to extract feature vectors of the target domain training data to obtain target domain user information vectors and target domain project information vectors;
a model training module configured to train a recommendation model based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
According to a third aspect of embodiments of the present application, there is provided a computing device comprising a memory, a processor and computer instructions stored on the memory and executable on the processor, the processor implementing the steps of the model-based privacy preserving cross-domain recommendation method when executing the computer instructions.
According to a fourth aspect of embodiments of the present application, there is provided a computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the generation model based privacy preserving cross-domain recommendation method.
The privacy protection cross-domain recommendation method based on the generation model provided by the application is used for acquiring source domain prediction data and target domain training data, wherein the source domain prediction data is generated by inputting noise data into the generation model; extracting the characteristic vector of the source domain prediction data to obtain a source domain user information vector; extracting feature vectors of the target domain training data to obtain a target domain user information vector and a target domain project information vector; a recommendation model is trained based on the cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
According to the embodiment of the application, the source domain prediction data generated by the generation model according to the noise data is obtained, the target domain training data is obtained, the source domain user information vector in the source domain prediction data is extracted, and the target domain user information vector and the target domain project information vector in the target domain training data are extracted, so that the recommendation model is trained in the target domain according to the cross-domain information vector between the source domain user information vector and the target domain project information vector; the source domain prediction data obtained by the generation model can not directly touch the original information of the user in the source domain, so that the privacy protection of the source domain user data is realized; in addition, the input data for training the recommendation model also comprises a cross-domain information vector, and the user knowledge of the source domain is transferred to the target domain, so that the recommendation performance of the recommendation model in the target domain is improved.
Drawings
Fig. 1 is a schematic structural diagram of a privacy protection cross-domain recommendation method based on a generation model according to an embodiment of the present application;
FIG. 2 is a flowchart of a method for generating model-based privacy preserving cross-domain recommendation according to an embodiment of the present application;
FIG. 3 is a process flow diagram of a privacy preserving cross-domain recommendation method based on a generative model according to an embodiment of the present application;
FIG. 4 is a flowchart of a method for recommending items according to an embodiment of the present application;
FIG. 5 is a schematic structural diagram of a privacy preserving cross-domain recommendation apparatus based on a generation model according to an embodiment of the present application;
FIG. 6 is a block diagram of a computing device according to one embodiment of the application.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present application may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present application is not limited to the specific embodiments disclosed below.
The terminology used in the one or more embodiments of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of the application. As used in one or more embodiments of the application and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used in one or more embodiments of the present application refers to and encompasses any or all possible combinations of one or more of the associated listed items.
It should be understood that, although the terms first, second, etc. may be used in one or more embodiments of the application to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, a first may also be referred to as a second, and similarly, a second may also be referred to as a first, without departing from the scope of one or more embodiments of the application. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.
Furthermore, it should be noted that, user information (including, but not limited to, user equipment information, user personal information, etc.) and data (including, but not limited to, data for analysis, stored data, presented data, etc.) according to one or more embodiments of the present disclosure are information and data authorized by a user or sufficiently authorized by each party, and the collection, use, and processing of relevant data is required to comply with relevant laws and regulations and standards of relevant countries and regions, and is provided with corresponding operation entries for the user to select authorization or denial.
First, terms related to one or more embodiments of the present application will be explained.
Differential privacy: differential privacy may provide a strong privacy protection for individual data in a data set.
Privacy-preserving cross-domain recommendations aim at privacy-preserving source domain data, and then use knowledge of source domain encryption for the task of assisting the target domain to promote recommendation performance. In society with increasingly strong awareness of privacy protection, solving this problem is particularly important for the long-term development of recommendation systems.
At present, a cross-domain recommending task, namely a recommending task of assisting a target domain by utilizing knowledge in source domain data, is widely studied in the aspect of improving recommending performance, but the cross-domain recommending methods cannot be directly used in a real scene for protecting user data privacy, and the recommending performance is reduced along with the reduction of the utilization rate of data by directly utilizing a privacy protection technology in the target domain. Therefore, the problem of privacy protection cross-domain recommendation is solved, and the problem of how to transfer knowledge from a source domain to a target domain to obtain better performance on the premise of protecting the privacy of user data can be understood, which is important for the long-term development of a recommendation system.
Based on the privacy protection cross-domain recommendation task, the embodiment of the application provides a solution based on a generation model, the generation model is used for carrying out privacy protection modeling on source domain data, then a target domain acquires source domain knowledge from the generation model, and further, direct access of the target domain to original data is isolated, so that privacy leakage is prevented; in addition, on the premise that the source domain data is protected by privacy, an economic privacy protection modeling strategy and a robust recommendation scoring strategy are adopted to maintain good recommendation performance of the target domain.
In the present application, a privacy protection cross-domain recommendation method based on a generation model is provided, and the present application relates to an item recommendation method, a privacy protection cross-domain recommendation device based on a generation model, a computing device, and a computer-readable storage medium, which are described in detail in the following embodiments one by one.
Fig. 1 shows a schematic structural diagram of a privacy protection cross-domain recommendation method based on a generative model according to an embodiment of the present application.
The privacy protection cross-domain recommendation method based on the generated model can be applied to the training process of the recommendation model in the target domain, and in order to train the recommendation model with good recommendation performance in the target domain, knowledge is migrated to the target domain after privacy protection is carried out on the original data in the source domain, so that the training process of the recommendation model is completed.
It should be noted that, the source domain generally refers to a data domain with dense user data, and the target domain generally refers to a data domain with sparse user data, and it is generally desired that knowledge obtained by the source domain can be effectively migrated to the target domain, so as to implement cross-domain recommendation for the target domain. Of course, in practical application, any one of the data fields may be used as a source field or a target field, which is not specifically limited herein.
In practical application, when a target domain trains a recommended model, source domain prediction data can be acquired first, wherein the source domain prediction data is acquired according to a generated model, and the training process of the generated model is realized in a source domain, and the embodiment is not limited to the method; the source domain prediction data may include a user information vector of a registered user in the source domain, and further, the user information vector in the source domain is extracted from the source domain prediction data; further, acquiring target domain training data, wherein the target domain training data can comprise user information vectors of registered users in a target domain and project information vectors of recommended projects for the users in the target domain, and further extracting the user information vectors and the project information vectors in the target domain from the target domain training data; then, realizing knowledge migration according to the user information vector in the source domain and the user information vector in the target domain, and obtaining a cross-domain information vector; and finally, training a recommendation model according to the cross-domain information vector and the target domain project information vector.
It should be noted that, the user information vector in the source domain extracted in the target domain is a vector after privacy protection processing, and in this embodiment, the source domain prediction data may be obtained by inputting noise data corresponding to the original user data in the source domain into the generation model.
Fig. 2 shows a flowchart of a privacy protection cross-domain recommendation method based on a generative model according to an embodiment of the present application, which specifically includes the following steps:
it should be noted that, the execution subject of the privacy protection cross-domain recommendation method based on the generation model provided in this embodiment may be understood as an execution subject in a target domain, and by acquiring user knowledge in a source domain and migrating the user knowledge to the target domain, the recommendation model in the target domain is trained; the recommendation model in this embodiment may be adapted to recommend corresponding items for the user, where the items may refer to service items provided by the platform of the target domain, such as merchandise, books, music, movies, and the like.
Step 202: source domain prediction data and target domain training data are obtained, wherein the source domain prediction data are generated by inputting noise data into a generation model.
The source domain prediction data may be understood as training data output from a source domain, and the source domain prediction data is generated by inputting noise data into a generation model, where the training data includes user information data registered in the source domain, project information recommended for a registered user, and the like; the target domain training data may be understood as data for training a recommendation model in the target domain, and may include user information data registered in the target domain, project information recommended for a registered user, and the like.
It should be noted that, the user information mentioned in the source domain and the target domain characterizes the attribute of the user, such as the user name, the user identity, the user gender, etc., and the item information characterizes the attribute of the item, such as the item name, the item tag, the time of generation, etc., in the embodiment of the present application, the item refers to the service item provided by the platform of the data domain (source domain or target domain), and may specifically be a commodity, a book, music, a movie, etc.
In practical application, the execution body may acquire source domain prediction data and target domain training data, where the source domain prediction data is generated by inputting noise data into a generation model, and it should be noted that the noise data may be understood as random noise data of the training generation model, and the generation model is used for predicting user information data in the source domain.
Further, in order to improve performance of the recommendation model in the target domain, a corresponding user information vector can be acquired from the source domain so as to train the recommendation model in the target domain better, and meanwhile, in order to avoid privacy leakage of the user information in the source domain when the user information in the source domain is migrated to the target domain, source domain prediction data can be obtained by adding noise to the original sample; specifically, before the source domain prediction data is obtained, the method further includes:
Acquiring original data, wherein the original data comprises user information, project information and associated information between the user information and the project information;
randomly sampling noise data associated with the structure of the raw data;
inputting the noise data into a to-be-trained generation model to obtain prediction data corresponding to the noise data, wherein the prediction data comprises user information vectors;
and calculating similarity based on the original data and the predicted data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain a generation model.
The original data may be understood as initial data in the source domain platform, including user information registered in the source domain platform, item information provided by the source domain platform, association information between the user information and the item information, and the like, where the association information may include association information for recommending corresponding items for the user, for example, user 1-item 2, and may be characterized as items commonly used by the user 1 in the source domain platform are item 1 and item 2.
In practical application, original data can be acquired in a source domain, wherein the original data comprises user information of registered users in the source domain, project information provided by a source domain platform, association information between the user information and the project information and the like; further, noise data with the same shape as the original data can be sampled from random distribution, then the noise data is input into a to-be-trained generation model, and prediction data corresponding to the noise data is obtained, wherein the prediction data comprises user information vectors of users in a source domain platform; and finally, calculating the similarity according to the original data and the predicted data, and adjusting parameters in the to-be-trained generation model according to the similarity until the training stopping condition is reached, so as to obtain the generation model.
It should be noted that, in the process of training the generated model in this embodiment, the method may also be implemented by using a discriminant model, and calculate the similarity between the original data and the predicted data, so as to continuously adjust the model parameters of the generated model to be trained; specifically, the calculating the similarity based on the original data and the predicted data, and adjusting parameters in the to-be-trained generating model according to the similarity, to obtain the generating model, includes:
carrying out noise adding processing on the original data to obtain a noise-added original sample;
and calculating the similarity between the noise-added original sample and the prediction data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain the generation model.
In an alternative embodiment, in order to protect user privacy data, differential privacy encryption processing can be performed on the original data to obtain a noisy original sample, a discrimination model is used to calculate a vector distance between the noisy original sample and the predicted data, similarity between vectors is determined, and the performance of a generation model is trained according to the similarity, so that the generation model in a source domain outputs source domain predicted data similar to the original data, the user information vector of the original data can be represented, and the privacy data of the original data can be prevented from being leaked. For example, the raw data represents With user id information->For the condition, the noise data can be expressed as +.>And (2) original data->Distributed similarly spurious data representation +.>Generating a model representation +.>The discrimination model is obtained by comparison->And->The differences in distribution between minimizing the estimate of real raw data and maximizing the estimate of spurious data can be referred to as equation 1 below:
equation 1
The discrimination model completes the generation of the model by calculating the loss gradient informationThe training process of the system adopts a mode of adding noise to the loss gradient information of the whole framework to carry out privacy protection modeling; namely, the original data of the source domain is directly connected with a discrimination model, then a generation model obtains the real data of the source domain through the discrimination model, and the loss function gradient of the discrimination model is decomposed according to a chain rule to obtain the loss function gradient of the discrimination model>Product of the sum of jacobian matrices +.>Reference is made to the following equation 2:
equation 2
By gradient the loss functionEncryption is carried out by adopting a Gaussian mechanism, the whole discrimination model and a later generation model are in a post-processing mode, privacy is guaranteed not to be leaked, and the modeling process of the discrimination model for obtaining the following privacy protection is shown in the following formula 3:
Equation 3
Wherein,,noise size for privacy loss, +.>For restraining->Has +.>This upper bound.
Based on the method, the Gaussian mechanism is applied to the gradient part of the loss function in the privacy modeling process, and meanwhile, the limit of privacy loss of the Renyi differential privacy constraint is adopted, so that economical privacy protection modeling is realized, namely, the introduced noise is minimized, and meanwhile, the privacy of a user is effectively protected, and the method is particularly important for a target domain to utilize source domain knowledge.
Step 204: and extracting the characteristic vector of the source domain prediction data to obtain a source domain user information vector.
In practical application, the executing body extracts a feature vector in source domain predicted data to obtain a source domain user information vector, wherein the source domain user information vector can represent source domain false data corresponding to original data in a source domain platform, but can represent user information in the original data in the source domain to a certain extent.
Further, in order to mine user knowledge from source domain false data, the vector can be encoded by using an encoding reconstruction technology so as to gather user information and obtain a source domain user information vector; specifically, the extracting the feature vector of the source domain prediction data to obtain a source domain user information vector includes:
Extracting a feature vector of the source domain prediction data;
and encoding the feature vector to obtain a hidden layer representation of the feature vector, and determining the hidden layer representation as a source domain user information vector.
In practical application, feature vectors in the source domain prediction data can be extracted, the feature vectors are encoded, hidden layer representation of the feature vectors is obtained, and the hidden layer representation is determined to be a source domain user information vector; in an alternative embodiment, by using the reconstruction capability of the automatic encoder, the hidden layer representation is extracted by reconstructing N users' false scoring data r ̃ _s in the source domain, and the following training objectives are obtained when extracting the source domain user information vector as the user knowledge of the source domain (see the following formula 4):
equation 4
Wherein the reconstructed object is evaluated in the form of the sum of squares of the Frobenius norms.
Based on this, the feature vectors of the source domain prediction data are reconstructed such that the user information aggregation is adjusted to dense data to better characterize the user information vectors.
Step 206: and extracting the feature vector of the target domain training data to obtain a target domain user information vector and a target domain project information vector.
In practical application, after the target domain training data is obtained, the feature vector of the target domain training data can be extracted to obtain a target domain user information vector and a target domain project information vector, wherein the target domain user information vector is a vector for representing user information in a target domain platform, and the target domain project information vector is a vector for representing project information recommended by a user in the target domain platform.
Step 208: a recommendation model is trained based on the cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
Further, in order to learn the source domain user information vector in the target domain, the similarity between the source domain user information vector and the target domain user information vector can be calculated, so that the characteristics of the two information vectors tend to be similar, the user knowledge migration is realized, and the cross-domain information vector, namely the domain aligned information vector, is obtained; further, the recommendation model is trained using the cross-domain information vector and the target domain project information vector.
In an alternative embodiment, the user knowledge migration between the source domain and the target domain is implemented in the form of the sum of squares of Frobenius norms as a bridge, resulting in a domain alignment target (see equation 5 below):
equation 5
Specifically, the training recommendation model based on the cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain item information vector includes:
determining a target loss value based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector;
And training a recommendation model based on the target loss value.
In practical application, a target loss value corresponding to the model is determined based on a cross-domain information vector between a source domain user information vector and a target domain project information vector, and a recommendation model is trained according to the target loss value.
Further, in this embodiment, when the recommendation model is trained according to the cross-domain information vector and the target domain project information vector, the target loss value may be determined according to the training target corresponding to the cross-domain information vector and the training target between the cross-domain information vector and the target domain information vector; specifically, the determining a target loss value based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain item information vector includes:
determining a cross-domain information vector between the source domain user information vector and the target domain user information vector;
determining a cross-domain loss value based on the cross-domain information vector;
determining a project prediction loss value based on the cross-domain information vector and the target domain project information vector;
and determining a target loss value according to the cross-domain loss value and the project predicted loss value.
In practical application, in this embodiment, a cross-domain loss value may be determined according to a cross-domain information vector, and a project prediction loss value may be determined according to the cross-domain information vector and a target domain project information vector; further, a target loss value is determined according to the cross-domain loss value and the project predicted loss value, so that parameters of the recommendation model can be adjusted according to the target loss value.
Furthermore, in order to enhance the robustness of the cross-domain recommended item under the modeling of privacy protection, the embodiment can adopt the noise processing of the model parameters in the recommended model to resist the robustness scoring prediction of the recommended model; specifically, the determining the project prediction loss value based on the cross-domain information vector and the target domain project information vector includes:
inputting the cross-domain information vector and the target domain project information vector into a recommendation model to be trained to obtain initial recommendation information;
noise is added to the model parameters in the recommendation model to be trained, and an alternative recommendation model is obtained;
inputting the cross-domain information vector and the target domain project information vector into the alternative recommendation model to obtain alternative recommendation information;
and determining an item prediction loss value based on the initial recommendation information and the alternative recommendation information.
In practical application, a cross-domain information vector and a target domain project information vector can be input into a recommendation model to be trained, initial recommendation information can be obtained, model parameters in the recommendation model to be trained are further subjected to noise adding to obtain an alternative recommendation model, and then the cross-domain information vector and the target domain project information vector are input into the alternative recommendation model to obtain alternative recommendation information; determining project prediction loss values according to the initial recommendation information and the alternative recommendation information; the recommendation information output by the recommendation model may be understood as item information, etc. for recommending items for the user in the target domain.
In an alternative embodiment, to enhance the robustness of cross-domain recommendations under privacy modeling, robust scoring predictions are made with antagonistic personalized rankings. First, a target domain preference set T is constructed for a favorite commodity i and a dislike commodity j of a user u, and Bayesian personalized ranking is calculated, see the following formula 6:
equation 6
Wherein z represents the user knowledge representation, v represents the commodity knowledge representation,and->，/>As constraint target domain model parameters +.>Further, when the model is disturbed, the disturbance range of the model can be designated as +. >The robustness of the target domain model prediction is enhanced by adding a form of anti-disturbance to the model parameters, namely the following formula 7 can be referred to:
equation 7
Based on the method, in order to improve the robustness of the recommendation model, when determining the training target of the project prediction loss value, the project prediction loss value can be calculated according to the recommendation information output by the recommendation model before and after the noise addition by carrying out the noise addition processing on the model parameters, so that the project prediction loss value can be better trained in the target domain to obtain the recommendation model.
In addition, in the embodiment, when the recommendation model is trained, the loss value corresponding to the source domain user information vector can be used for training; specifically, before determining the target loss value according to the cross-domain loss value and the project predicted loss value, the method further includes:
determining a source domain loss value based on the source domain user information vector;
accordingly, the determining a target loss value according to the cross-domain loss value and the project predicted loss value includes:
and determining a target loss value according to the source domain loss value, the cross-domain loss value and the project predicted loss value.
In practical application, in the training process of acquiring the user knowledge of the source domain, the source domain loss value can be determined according to the source domain user information vector, and the source domain loss value is added into the training of the recommendation model, so that the target loss value is determined according to the source domain loss value, the cross-domain loss value and the project prediction loss value, and model parameters are adjusted by utilizing the target loss value, thereby realizing the training of the recommendation model. Based on this, in maintaining a robust recommendation scoring strategy for target domain recommendation performance, the robust recommendation predictive training target in the target domain may refer to equation 8 as follows:
Equation 8
In summary, in the embodiment of the application, the generation model in the source domain is utilized to isolate the target domain to directly query the original user data in the source domain, so that the user privacy is prevented from being revealed, and furthermore, on the premise that the user data in the source domain is protected by privacy, an economical privacy protection modeling strategy and a robust recommendation scoring strategy can be adopted to maintain good recommendation performance in the target domain, so that the method for protecting the cross-domain recommendation item by the privacy of the generation model is realized, the cross-domain recommendation performance of the recommendation model is improved, and the factor protection of the user data in the source domain is realized.
Referring to fig. 3, fig. 3 shows a process flow diagram of a privacy protection cross-domain recommendation method based on a generation model according to an embodiment of the present application, which specifically includes the following steps:
step 302: source domain data is acquired in a source domain.
Specifically, the source domain data is user information, project information, and association information between the user information and the project information in the source domain platform.
Step 304: random noise similar to the source domain data distribution structure is acquired.
Step 306: the random noise input generates a model.
Step 308: source domain spurious data is obtained.
Step 310: and calculating the similarity between the source domain data and the source domain false data by using the discriminant model so as to train and generate the model.
Step 312: feature vectors in the source domain spurious data are extracted.
Step 314: and acquiring target domain data.
Step 316: and extracting the feature vector in the target domain data.
Step 318: and carrying out coding processing on the source domain feature vector according to the feature reconstruction mode to obtain source domain user knowledge.
Step 320: and obtaining the user knowledge of the target domain based on the feature vector in the target domain data.
Step 322: and obtaining target domain commodity knowledge based on the feature vector in the target domain data.
Step 324: and carrying out domain knowledge migration based on the source domain user knowledge and the target domain user knowledge to obtain cross-domain user knowledge.
Step 326: and according to the cross-domain user knowledge and the target domain commodity knowledge, performing robust scoring prediction, and training the recommendation in the target domain to improve the cross-domain recommendation performance.
In summary, the embodiment uses the generation model of privacy modeling to perform privacy protection cross-domain recommendation, and maintains the robust recommendation scoring strategy of the recommendation performance of the target domain while using the knowledge of the privacy protection source domain.
Fig. 4 shows a flowchart of a method for recommending items according to an embodiment of the present application, which specifically includes the following steps:
it should be noted that, the item recommendation method provided in this embodiment may be applied to item recommendation for a user in a target domain, including service items provided for a platform of the target domain, such as merchandise, books, music, movies, and the like, which is not limited in this embodiment.
Step 402: and acquiring user information of the target user.
The user information may understand the user information in the data domain platform, such as a user name, a user identity, a user gender, etc., which is not limited in this embodiment.
Step 404: and inputting the user information into a recommendation model to obtain project recommendation information for the target user, wherein the recommendation model is obtained by training by using the recommendation model training method.
In practical application, the obtained user information of the target user is input into a recommendation model, so that item recommendation information recommended for the target user can be obtained, for example, other service items, including but not limited to other books, music and the like, are recommended for the target user in a target domain; it should be noted that, the recommendation model in this embodiment may be obtained by using the method for training the recommendation model in the foregoing embodiment, and the specific training process of the recommendation model is not described in detail herein.
In summary, the project recommendation method provided by the embodiment of the application can realize that not only the privacy of user data can be protected, but also suitable projects can be recommended for users under the cross-domain condition, and the use experience of users on the cross-domain platform is improved, so that the cross-domain platform can more accurately acquire the user demands, and the long-term development of a recommendation model is ensured.
Corresponding to the method embodiment, the application further provides a privacy protection cross-domain recommending device embodiment based on the generation model, and fig. 5 shows a schematic structural diagram of the privacy protection cross-domain recommending device based on the generation model provided by an embodiment of the application. As shown in fig. 5, the apparatus includes:
a training data acquisition module 502 configured to acquire source domain prediction data and target domain training data, wherein the source domain prediction data is obtained by generating a model based on noise data;
a first feature vector extraction module 504 configured to extract feature vectors of the source domain prediction data, obtaining source domain user information vectors;
a second feature vector extraction module 506 configured to extract feature vectors of the target domain training data to obtain a target domain user information vector and a target domain item information vector;
a model training module 508 is configured to train a recommendation model based on the cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
Optionally, the apparatus further comprises:
the system comprises a generation model training module, a generation model generation module and a generation model generation module, wherein the generation model training module is configured to acquire original data, wherein the original data comprises user information, project information and associated information between the user information and the project information;
Randomly sampling noise data associated with the structure of the raw data;
inputting the noise data into a to-be-trained generation model to obtain prediction data corresponding to the noise data, wherein the prediction data comprises user information vectors;
and calculating similarity based on the original data and the predicted data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain a generation model.
Optionally, the generative model training module is further configured to:
carrying out noise adding processing on the original data to obtain a noise-added original sample;
and calculating the similarity between the noise-added original sample and the prediction data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain the generation model.
Optionally, the first feature vector extraction module 504 is further configured to:
extracting a feature vector of the source domain prediction data;
and encoding the feature vector to obtain a hidden layer representation of the feature vector, and determining the hidden layer representation as a source domain user information vector.
Optionally, the model training module 508 is further configured to:
determining a target loss value based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector;
And training a recommendation model based on the target loss value.
Optionally, the model training module 508 is further configured to:
determining a cross-domain information vector between the source domain user information vector and the target domain user information vector;
determining a cross-domain loss value based on the cross-domain information vector;
determining a project prediction loss value based on the cross-domain information vector and the target domain project information vector;
and determining a target loss value according to the cross-domain loss value and the project predicted loss value.
Optionally, the model training module 508 is further configured to:
inputting the cross-domain information vector and the target domain project information vector into a recommendation model to be trained to obtain initial recommendation information;
noise is added to the model parameters in the recommendation model to be trained, and an alternative recommendation model is obtained;
inputting the cross-domain information vector and the target domain project information vector into the alternative recommendation model to obtain alternative recommendation information;
and determining an item prediction loss value based on the initial recommendation information and the alternative recommendation information.
Optionally, the apparatus further comprises:
a source domain loss value determination module configured to determine a source domain loss value based on the source domain user information vector;
Optionally, the model training module 508 is further configured to:
and determining a target loss value according to the source domain loss value, the cross-domain loss value and the project predicted loss value.
According to the privacy protection cross-domain recommendation device based on the generation model, the source domain prediction data generated by the generation model according to the noise data is obtained, the target domain training data is obtained, the source domain user information vector in the source domain prediction data is extracted, and the target domain user information vector and the target domain project information vector in the target domain training data are extracted, so that the recommendation model is trained in the target domain according to the cross-domain information vector between the source domain user information vector and the target domain project information vector; the source domain prediction data obtained by the generation model can not directly touch the original information of the user in the source domain, so that the privacy protection of the source domain user data is realized; in addition, the input data for training the recommendation model also comprises a cross-domain information vector, and the user knowledge of the source domain is transferred to the target domain, so that the recommendation performance of the recommendation model in the target domain is improved.
The foregoing is a schematic scheme of a privacy protection cross-domain recommendation device based on a generation model in this embodiment. It should be noted that, the technical solution of the privacy protection cross-domain recommendation device based on the generation model and the technical solution of the privacy protection cross-domain recommendation method based on the generation model belong to the same concept, and details of the technical solution of the privacy protection cross-domain recommendation device based on the generation model, which are not described in detail, can be referred to the description of the technical solution of the privacy protection cross-domain recommendation method based on the generation model.
Fig. 6 illustrates a block diagram of a computing device 600 provided in accordance with an embodiment of the present application. The components of computing device 600 include, but are not limited to, memory 610 and processor 620. The processor 620 is coupled to the memory 610 via a bus 630 and a database 650 is used to hold data.
Computing device 600 also includes access device 640, access device 640 enabling computing device 600 to communicate via one or more networks 660. Examples of such networks include public switched telephone networks (PSTN, public Switched Telephone Network), local area networks (LAN, local Area Network), wide area networks (WAN, wide Area Network), personal area networks (PAN, personal Area Network), or combinations of communication networks such as the internet. The access device 640 may include one or more of any type of network interface, wired or wireless, such as a network interface card (NIC, network interface controller), such as an IEEE802.11 wireless local area network (WLAN, wireless Local Area Network) wireless interface, a worldwide interoperability for microwave access (Wi-MAX, worldwide Interoperability for Microwave Access) interface, an ethernet interface, a universal serial bus (USB, universal Serial Bus) interface, a cellular network interface, a bluetooth interface, a near field communication (NFC, near Field Communication) interface, and so forth.
In one embodiment of the application, the above-described components of computing device 600, as well as other components not shown in FIG. 6, may also be connected to each other, such as by a bus. It should be understood that the block diagram of the computing device illustrated in FIG. 6 is for exemplary purposes only and is not intended to limit the scope of the present application. Those skilled in the art may add or replace other components as desired.
Computing device 600 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., tablet, personal digital assistant, laptop, notebook, netbook, etc.), mobile phone (e.g., smart phone), wearable computing device (e.g., smart watch, smart glasses, etc.), or other type of mobile device, or a stationary computing device such as a desktop computer or personal computer (PC, personal Computer). Computing device 600 may also be a mobile or stationary server.
Wherein the processor 620, when executing the computer instructions, implements the steps of the generating model-based privacy preserving cross-domain recommendation method.
The foregoing is a schematic illustration of a computing device of this embodiment. It should be noted that, the technical solution of the computing device and the technical solution of the privacy protection cross-domain recommendation method based on the generation model belong to the same concept, and details of the technical solution of the computing device, which are not described in detail, can be referred to the description of the technical solution of the privacy protection cross-domain recommendation method based on the generation model.
An embodiment of the present application also provides a computer-readable storage medium storing computer instructions that, when executed by a processor, implement the steps of a privacy preserving cross-domain recommendation method based on a generative model as described above.
The above is an exemplary version of a computer-readable storage medium of the present embodiment. It should be noted that, the technical solution of the storage medium and the technical solution of the privacy protection cross-domain recommendation method based on the generation model belong to the same concept, and details of the technical solution of the storage medium, which are not described in detail, can be referred to the description of the technical solution of the privacy protection cross-domain recommendation method based on the generation model.
The foregoing describes certain embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
The computer instructions include computer program code that may be in source code form, object code form, executable file or some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth. It should be noted that the computer readable medium contains content that can be appropriately scaled according to the requirements of jurisdictions in which such content is subject to legislation and patent practice, such as in certain jurisdictions in which such content is subject to legislation and patent practice, the computer readable medium does not include electrical carrier signals and telecommunication signals.
It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present application is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present application. Further, those skilled in the art will appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily all required for the present application.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.
The preferred embodiments of the application disclosed above are intended only to assist in the explanation of the application. Alternative embodiments are not intended to be exhaustive or to limit the application to the precise form disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the application and the practical application, to thereby enable others skilled in the art to best understand and utilize the application. The application is limited only by the claims and the full scope and equivalents thereof.
Claims (11)
1. The privacy protection cross-domain recommendation method based on the generation model is characterized by comprising the following steps of:
acquiring original data, wherein the original data comprises user information of registered users in a source domain, project information provided by a source domain platform and associated information between the user information and the project information;
randomly sampling noise data associated with the structure of the raw data;
inputting the noise data into a to-be-trained generation model to obtain prediction data corresponding to the noise data, wherein the prediction data comprises user information vectors of users in the source domain platform;
Calculating similarity based on the original data and the predicted data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain a generation model;
acquiring source domain prediction data and target domain training data, wherein the source domain prediction data is generated by inputting noise data into the generation model;
extracting the characteristic vector of the source domain prediction data to obtain a source domain user information vector;
extracting feature vectors of the target domain training data to obtain a target domain user information vector and a target domain project information vector;
a recommendation model is trained based on the cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
2. The method according to claim 1, wherein the calculating the similarity based on the raw data and the predicted data, adjusting parameters in the to-be-trained generation model according to the similarity, and obtaining the generation model, comprises:
carrying out noise adding processing on the original data to obtain a noise-added original sample;
and calculating the similarity between the noise-added original sample and the prediction data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain the generation model.
3. The method of claim 1, wherein extracting the feature vector of the source domain prediction data to obtain a source domain user information vector comprises:
extracting a feature vector of the source domain prediction data;
and encoding the feature vector to obtain a hidden layer representation of the feature vector, and determining the hidden layer representation as a source domain user information vector.
4. The method of claim 1, wherein the training a recommendation model based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain item information vector, comprises:
determining a target loss value based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector;
and training a recommendation model based on the target loss value.
5. The method of claim 4, wherein the determining a target loss value based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain item information vector, comprises:
Determining a cross-domain information vector between the source domain user information vector and the target domain user information vector;
determining a cross-domain loss value based on the cross-domain information vector;
determining a project prediction loss value based on the cross-domain information vector and the target domain project information vector;
and determining a target loss value according to the cross-domain loss value and the project predicted loss value.
6. The method of claim 5, wherein determining an item prediction loss value based on the cross-domain information vector and the target domain item information vector comprises:
inputting the cross-domain information vector and the target domain project information vector into a recommendation model to be trained to obtain initial recommendation information;
noise is added to the model parameters in the recommendation model to be trained, and an alternative recommendation model is obtained;
inputting the cross-domain information vector and the target domain project information vector into the alternative recommendation model to obtain alternative recommendation information;
and determining an item prediction loss value based on the initial recommendation information and the alternative recommendation information.
7. The method of claim 5, wherein prior to determining a target loss value from the cross-domain loss value and the project predicted loss value, further comprising:
Determining a source domain loss value based on the source domain user information vector;
accordingly, the determining a target loss value according to the cross-domain loss value and the project predicted loss value includes:
and determining a target loss value according to the source domain loss value, the cross-domain loss value and the project predicted loss value.
8. A method of recommending items, comprising:
acquiring user information of a target user;
inputting the user information into a recommendation model to obtain project recommendation information for the target user, wherein the recommendation model is trained by using the privacy protection cross-domain recommendation method based on the generation model according to any one of claims 1-7.
9. A privacy preserving cross-domain recommendation device based on a generative model, comprising:
the training data acquisition module is configured to acquire original data, wherein the original data comprises user information of registered users in a source domain, project information provided by a source domain platform and associated information between the user information and the project information; randomly sampling noise data associated with the structure of the raw data; inputting the noise data into a to-be-trained generation model to obtain prediction data corresponding to the noise data, wherein the prediction data comprises user information vectors of users in the source domain platform; calculating similarity based on the original data and the predicted data, and adjusting parameters in the to-be-trained generation model according to the similarity to obtain a generation model; acquiring source domain prediction data and target domain training data, wherein the source domain prediction data is obtained through the generation model based on noise data;
The first feature vector extraction module is configured to extract the feature vector of the source domain prediction data to obtain a source domain user information vector;
the second feature vector extraction module is configured to extract feature vectors of the target domain training data to obtain target domain user information vectors and target domain project information vectors;
a model training module configured to train a recommendation model based on a cross-domain information vector between the source domain user information vector and the target domain user information vector, and the target domain project information vector.
10. A computing device comprising a memory, a processor, and computer instructions stored on the memory and executable on the processor, wherein the processor, when executing the computer instructions, performs the steps of the method of any one of claims 1-8.
11. A computer readable storage medium storing computer instructions which, when executed by a processor, implement the steps of the method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310742959.1A CN116502271B (en) | 2023-06-21 | 2023-06-21 | Privacy protection cross-domain recommendation method based on generation model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310742959.1A CN116502271B (en) | 2023-06-21 | 2023-06-21 | Privacy protection cross-domain recommendation method based on generation model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116502271A CN116502271A (en) | 2023-07-28 |
| CN116502271B true CN116502271B (en) | 2023-09-19 |
Family
ID=87326912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310742959.1A Active CN116502271B (en) | 2023-06-21 | 2023-06-21 | Privacy protection cross-domain recommendation method based on generation model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116502271B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112116025A (en) * | 2020-09-28 | 2020-12-22 | 北京嘀嘀无限科技发展有限公司 | User classification model training method and device, electronic equipment and storage medium |
| CN112417293A (en) * | 2020-12-03 | 2021-02-26 | 京东数字科技控股股份有限公司 | Information pushing method and system, model training method and related equipment |
| CN113312644A (en) * | 2021-06-15 | 2021-08-27 | 杭州金智塔科技有限公司 | Cross-domain recommendation model training method and system based on privacy protection |
| CN115033916A (en) * | 2022-06-07 | 2022-09-09 | 支付宝(杭州)信息技术有限公司 | Multi-party combined data based push model training and information push method and device |
| CN115438732A (en) * | 2022-09-06 | 2022-12-06 | 重庆理工大学 | Cross-domain recommendation method for cold start user based on classification preference migration |
| CN115952360A (en) * | 2023-03-14 | 2023-04-11 | 中国海洋大学 | Domain-adaptive cross-domain recommendation method and system based on user and article commonality modeling |
-
2023
- 2023-06-21 CN CN202310742959.1A patent/CN116502271B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112116025A (en) * | 2020-09-28 | 2020-12-22 | 北京嘀嘀无限科技发展有限公司 | User classification model training method and device, electronic equipment and storage medium |
| CN112417293A (en) * | 2020-12-03 | 2021-02-26 | 京东数字科技控股股份有限公司 | Information pushing method and system, model training method and related equipment |
| WO2022116862A1 (en) * | 2020-12-03 | 2022-06-09 | 京东科技控股股份有限公司 | Information pushing method and system, model training method, and related devices |
| CN113312644A (en) * | 2021-06-15 | 2021-08-27 | 杭州金智塔科技有限公司 | Cross-domain recommendation model training method and system based on privacy protection |
| CN115033916A (en) * | 2022-06-07 | 2022-09-09 | 支付宝(杭州)信息技术有限公司 | Multi-party combined data based push model training and information push method and device |
| CN115438732A (en) * | 2022-09-06 | 2022-12-06 | 重庆理工大学 | Cross-domain recommendation method for cold start user based on classification preference migration |
| CN115952360A (en) * | 2023-03-14 | 2023-04-11 | 中国海洋大学 | Domain-adaptive cross-domain recommendation method and system based on user and article commonality modeling |
Non-Patent Citations (2)
| Title |
|---|
| 一种基于差分隐私保护的协同过滤推荐方法;何明;常盟盟;吴小飞;;计算机研究与发展(第07期);全文 * |
| 跨域推荐中安全可信的数据协同计算方法研究;王利娥;《中国博士学位论文全文数据库 电子期刊》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116502271A (en) | 2023-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111931062B (en) | Training method and related device of information recommendation model | |
| CN111967615B (en) | Multi-model training method and device based on feature extraction, electronic equipment and medium | |
| CN111949886B (en) | Sample data generation method and related device for information recommendation | |
| Plant et al. | CAPE: Context-aware private embeddings for private language learning | |
| CN110889759A (en) | Credit data determination method, device and storage medium | |
| CN114330966A (en) | Risk prediction method, device, equipment and readable storage medium | |
| CN112084307A (en) | Data processing method and device, server and computer readable storage medium | |
| CN107506479B (en) | A kind of object recommendation method and apparatus | |
| Chen et al. | Deformable convolutional matrix factorization for document context-aware recommendation in social networks | |
| CN116756278A (en) | Machine question-answering method and device | |
| CN116975349A (en) | Image retrieval method, device, electronic equipment and storage medium | |
| CN113704509B (en) | Multimedia recommendation method and device, electronic equipment and storage medium | |
| CN116502271B (en) | Privacy protection cross-domain recommendation method based on generation model | |
| CN108460811B (en) | Face image processing method and device and computer equipment | |
| CN117272937B (en) | Text coding model training method, device, equipment and storage medium | |
| CN115098771B (en) | Recommendation model updating method, recommendation model training method and computing equipment | |
| CN116501978A (en) | Recommendation model generation method and device based on privacy protection machine forgetting algorithm | |
| CN116704581A (en) | Face recognition method, device, equipment and storage medium | |
| CN116384405A (en) | Text processing method, text classification method and emotion recognition method | |
| CN115204299A (en) | Data processing method and device, storage medium and electronic equipment | |
| Jafari et al. | Detection of phishing addresses and pages with a data set balancing approach by generative adversarial network (GAN) and convolutional neural network (CNN) optimized with swarm intelligence | |
| CN114970670A (en) | Model fairness assessment method and device | |
| CN115758368B (en) | Prediction method and device for malicious cracking software, electronic equipment and storage medium | |
| Yang et al. | Privacy Through Diffusion: A White-Listing Approach to Sensor Data Anonymization | |
| CN116070695B (en) | Training method of image detection model, image detection method and electronic equipment |
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 |