CN117390281A - Project information recommendation method and system based on content interaction - Google Patents
Project information recommendation method and system based on content interaction Download PDFInfo
- Publication number
- CN117390281A CN117390281A CN202311439025.7A CN202311439025A CN117390281A CN 117390281 A CN117390281 A CN 117390281A CN 202311439025 A CN202311439025 A CN 202311439025A CN 117390281 A CN117390281 A CN 117390281A
- Authority
- CN
- China
- Prior art keywords
- model
- project
- recommendation
- information
- parameter set
- 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.)
- Withdrawn
Links
- 230000003993 interaction Effects 0.000 title claims abstract description 167
- 238000000034 method Methods 0.000 title claims abstract description 62
- 238000012549 training Methods 0.000 claims abstract description 142
- 238000012545 processing Methods 0.000 claims abstract description 27
- 230000010354 integration Effects 0.000 claims abstract description 24
- 238000004458 analytical method Methods 0.000 claims abstract description 20
- 230000006399 behavior Effects 0.000 claims description 78
- 230000005540 biological transmission Effects 0.000 claims description 18
- 239000013598 vector Substances 0.000 claims description 14
- 230000004927 fusion Effects 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 12
- 239000000463 material Substances 0.000 claims description 10
- 230000035945 sensitivity Effects 0.000 claims description 9
- 230000002776 aggregation Effects 0.000 claims description 8
- 238000004220 aggregation Methods 0.000 claims description 8
- 238000012795 verification Methods 0.000 claims description 8
- 230000000007 visual effect Effects 0.000 claims description 4
- 238000011160 research Methods 0.000 description 30
- 230000008569 process Effects 0.000 description 13
- 230000007613 environmental effect Effects 0.000 description 9
- 230000008859 change Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000002452 interceptive effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000008520 organization Effects 0.000 description 3
- 238000012935 Averaging Methods 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001010 compromised effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000013468 resource allocation Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Bioethics (AREA)
- Computer Security & Cryptography (AREA)
- Computer Hardware Design (AREA)
- Tourism & Hospitality (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Educational Administration (AREA)
- Medical Informatics (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to the field of data processing, and discloses a project information recommendation method and system based on content interaction, which are used for realizing intelligent recommendation analysis of content interaction, improving the accuracy of project information recommendation and improving the confidentiality of project information recommendation. The method comprises the following steps: carrying out model initialization through the project service cloud platform, generating an initialization model parameter set, and encrypting and distributing the initialization model parameter set to a plurality of first project information inquiry terminals; performing local model training and gradient parameter updating to generate a gradient updating model parameter set; encrypting the sensitive data to obtain the encrypted project declaration data; model feature training is carried out to obtain a feature model parameter set, and model collaborative updating is carried out to obtain a plurality of inquiry end recommendation models; performing model integration to generate a target global recommendation model; and inputting the content interaction behavior data into a target global recommendation model to conduct personalized item recommendation, and obtaining a target recommendation result.
Description
Technical Field
The invention relates to the field of data processing, in particular to a project information recommendation method and system based on content interaction.
Background
Government project declaration refers to a person, business, organization, or the like submitting an application to a government agency to obtain government funds, resources, or support for use in facilitating performance of a particular project or activity. These items relate to various fields such as scientific research, education, culture, environmental protection, entrepreneur, etc. The government can reasonably distribute social resources through a project declaration mode, and social development and innovation are promoted. Government project declarations typically require the applicant to submit a detailed application that includes information about the project's context, goals, plans, budgets, and the like. The applicant needs to clearly describe the meaning of the project, the expected outcome and the resources required. Government agencies will evaluate the feasibility, innovativeness, and contribution to society of an item based on information provided in the application, and then decide whether to give funds, support, or other forms of assistance. In government project declaration application scenarios, providing accurate, personalized project recommendations for users has become a critical task to improve user satisfaction and platform value.
However, the conventional recommendation method has some problems such as unsatisfactory recommendation effect for different types of users and items, and with increasing importance of user privacy protection and enhancement of data privacy regulations, the conventional centralized data collection and processing manner faces increasing challenges and exposes user privacy.
Disclosure of Invention
The invention provides a project information recommending method and system based on content interaction, which are used for realizing intelligent recommending analysis of the content interaction, improving the accuracy of project information recommendation and improving the confidentiality of project information recommendation.
The first aspect of the present invention provides a content interaction-based item information recommendation method, which includes:
constructing an item information interaction network based on a preset item service cloud platform and a plurality of first item information inquiry terminals, initializing a model through the item service cloud platform to generate an initialized model parameter set, and encrypting and distributing the initialized model parameter set to the plurality of first item information inquiry terminals according to the item information interaction network;
local model training is carried out on the plurality of first project information inquiry terminals through the initialization model parameter set, gradient parameter updating is carried out on gradient information obtained through training of each project information inquiry terminal, and a gradient updating model parameter set is generated;
acquiring government project declaration data through the plurality of first project information inquiry terminals, and performing sensitive data encryption on the government project declaration data to obtain encrypted project declaration data;
Model feature training is carried out through the encrypted project declaration data and the gradient updating model parameter set to obtain a feature model parameter set, and the feature model parameter set is sent to the plurality of first project information inquiry terminals to carry out model collaborative updating to obtain a plurality of inquiry terminal recommendation models;
carrying out model integration on the plurality of inquiry end recommendation models through the project service cloud platform to generate a target global recommendation model;
and acquiring content interaction behavior data of the target user based on the second item information inquiry terminal, and inputting the content interaction behavior data into the target global recommendation model to conduct personalized item recommendation to obtain a target recommendation result.
With reference to the first aspect, in a first implementation manner of the first aspect of the present invention, the building a project information interaction network based on a preset project service cloud platform and a plurality of first project information querying ends, performing model initialization through the project service cloud platform, generating an initialized model parameter set, and encrypting and distributing the initialized model parameter set to the plurality of first project information querying ends according to the project information interaction network, includes:
Carrying out item information inquiry end identification based on a preset item service cloud platform to obtain a plurality of first item information inquiry ends;
constructing a project information interaction network based on a preset project service cloud platform and a plurality of first project information inquiry ends;
respectively configuring a first training model to be trained in the project service cloud platform, and respectively creating a second training model to be trained in the plurality of first project information inquiry terminals;
carrying out model initialization on the first training model through the project service cloud platform to generate an initialized model parameter set;
performing parameter encryption processing on the initialized model parameter set to generate an initialized encrypted parameter set, and determining parameter transmission paths corresponding to the project service cloud platform and the plurality of first project information inquiry ends through the project information interaction network;
and distributing the initialized encryption parameter set to corresponding second training models in the plurality of first project information inquiry terminals through the parameter transmission path.
With reference to the first aspect, in a second implementation manner of the first aspect of the present invention, performing local model training on the plurality of first item information query ends through the initialization model parameter set, and performing gradient parameter update on gradient information obtained by training on each item information query end, to generate a gradient update model parameter set, including:
Performing parameter decryption processing on the initialized encryption parameter set through the plurality of first item information inquiry terminals to obtain the initialized model parameter set;
respectively matching the local training data set of each second training model through the plurality of first project information inquiry terminals;
performing local model training on the second training models of the plurality of first project information inquiry ends based on the local training data set and the initialization model parameter set to obtain first gradient information of each second training model;
creating corresponding first cloud platform nodes and second cloud platform nodes in the project service cloud platform, and creating corresponding first query end nodes and second query end nodes at each first project information query end;
encrypting the first gradient information through a first query end node of each first project information query end to obtain second gradient information, and transmitting the second gradient information to the project service cloud platform through the parameter transmission path;
decrypting the second gradient information through a second cloud platform node in the project service cloud platform to obtain third gradient information, and performing aggregation operation on the third gradient information to obtain target gradient information;
And updating the initialized model parameter set of the project service cloud platform through the target gradient information to obtain a gradient update model parameter set corresponding to the project service cloud platform.
With reference to the first aspect, in a third implementation manner of the first aspect of the present invention, the obtaining government project declaration data through the plurality of first project information querying ends, and encrypting sensitive data of the government project declaration data to obtain encrypted project declaration data includes:
acquiring government project declaration data through the plurality of first project information inquiry terminals, wherein the government project declaration data comprises: personal information filled in by a user and uploaded project declaration materials;
differential privacy analysis is carried out on the government project declaration data to obtain privacy budget, wherein the privacy budget comprises the following steps: privacy loss and sensitivity;
setting privacy parameters of the government project declaration data, wherein the privacy parameters comprise privacy protection level and noise intensity;
extracting characteristic information of the government project declaration data to obtain sensitive data information;
and carrying out noise addition on the sensitive data information according to the privacy budget and the privacy parameters to generate encrypted project declaration data.
With reference to the first aspect, in a fourth implementation manner of the first aspect of the present invention, performing model feature training through the encrypted project declaration data and the gradient update model parameter set to obtain a feature model parameter set, and sending the feature model parameter set to the plurality of first project information query ends to perform model collaborative update to obtain a plurality of query end recommendation models, where the method includes:
acquiring a plurality of target features corresponding to the first training model, wherein the target features comprise reporting histories and hobbies;
setting a target data format of the encrypted item declaration data according to the target features, and extracting a target feature data set corresponding to the encrypted item declaration data according to the target data format and the target features;
model training is carried out on the first training model through the target characteristic data set and the gradient updating model parameter set to obtain a characteristic model parameter set;
carrying out parameter encryption processing on the characteristic model parameter set to obtain an encrypted characteristic model parameter set, and transmitting the encrypted characteristic model parameter set to the plurality of first item information inquiry terminals;
And carrying out parameter decryption processing on the encrypted characteristic model parameter set through a second query end node in the plurality of first item information query ends, and carrying out parameter collaborative updating on a second training model in the plurality of first item information query ends through the characteristic model parameter set to obtain a plurality of query end recommendation models.
With reference to the first aspect, in a fifth implementation manner of the first aspect of the present invention, the performing, by the item service cloud platform, model integration on the plurality of query-side recommendation models, to generate a target global recommendation model includes:
according to preset model weights, carrying out model integration on the plurality of inquiry end recommendation models through the project service cloud platform to obtain an initial global recommendation model;
performing model verification on the initial global recommendation model to obtain a plurality of recommendation prediction results;
and carrying out user feedback analysis on the plurality of recommendation prediction results to obtain a plurality of user feedback data, wherein the user feedback data comprises: user ID, recommended item ID, interaction type, and time stamp;
and according to the plurality of user feedback data and the plurality of recommendation prediction results, performing model parameter optimization on the initial global recommendation model to generate a target global recommendation model.
With reference to the first aspect, in a sixth implementation manner of the first aspect of the present invention, the obtaining, based on the second item information query end, content interaction behavior data of the target user, and inputting the content interaction behavior data into the target global recommendation model to perform personalized item recommendation, and obtaining a target recommendation result includes:
acquiring content interaction behavior data of a target user based on a second item information inquiry end, and extracting features of the content interaction behavior data to obtain a plurality of content interaction behavior features;
performing feature coding processing on the plurality of content interaction behavior features to generate content interaction behavior coding vectors;
inputting the content interaction behavior coding vector into the target global recommendation model, and recommending personalized items to the content interaction behavior data through a plurality of inquiry end recommendation models in the target global recommendation model to obtain an initial recommendation result of each inquiry end recommendation model;
and carrying out result fusion on the initial recommendation result of the recommendation model of each query end, generating a target recommendation result, and pushing the target recommendation result to the second item information query end for visual display.
The second aspect of the present invention provides a content interaction-based item information recommendation system, comprising:
the initialization module is used for constructing an item information interaction network based on a preset item service cloud platform and a plurality of first item information inquiry terminals, carrying out model initialization through the item service cloud platform, generating an initialization model parameter set, and encrypting and distributing the initialization model parameter set to the plurality of first item information inquiry terminals according to the item information interaction network;
the training module is used for carrying out local model training on the plurality of first project information inquiry terminals through the initialization model parameter set, carrying out gradient parameter updating on gradient information obtained by training each project information inquiry terminal, and generating a gradient updating model parameter set;
the acquisition module is used for acquiring government project declaration data through the plurality of first project information inquiry terminals, and carrying out sensitive data encryption on the government project declaration data to acquire encrypted project declaration data;
the updating module is used for carrying out model feature training through the encryption project declaration data and the gradient updating model parameter set to obtain a feature model parameter set, and sending the feature model parameter set to the plurality of first project information inquiry terminals for model collaborative updating to obtain a plurality of inquiry terminal recommendation models;
The integration module is used for carrying out model integration on the plurality of inquiry end recommendation models through the project service cloud platform to generate a target global recommendation model;
and the recommendation module is used for acquiring content interaction behavior data of the target user based on the second item information inquiry end, inputting the content interaction behavior data into the target global recommendation model for personalized item recommendation, and obtaining a target recommendation result.
A third aspect of the present invention provides a content interaction-based item information recommendation apparatus, comprising: a memory and at least one processor, the memory having instructions stored therein; the at least one processor invokes the instructions in the memory to cause the content interaction based item information recommendation device to perform the content interaction based item information recommendation method described above.
A fourth aspect of the present invention provides a computer-readable storage medium having instructions stored therein which, when run on a computer, cause the computer to perform the above-described content interaction-based item information recommendation method.
According to the technical scheme provided by the invention, the project service cloud platform is used for carrying out model initialization, generating an initialization model parameter set, and encrypting and distributing the initialization model parameter set to a plurality of first project information inquiry terminals; performing local model training and gradient parameter updating to generate a gradient updating model parameter set; encrypting the sensitive data to obtain the encrypted project declaration data; model feature training is carried out to obtain a feature model parameter set, and model collaborative updating is carried out to obtain a plurality of inquiry end recommendation models; performing model integration to generate a target global recommendation model; the method and the device have the advantages that the content interaction behavior data are input into the target global recommendation model to conduct personalized project recommendation, and target recommendation results are obtained. Personal information and declaration materials of the user are transmitted and processed in an encrypted state, so that trust feeling of the user on the platform is enhanced, and the user is willing to participate and share data. The federal learning enables training of the model to be distributed on a plurality of item information inquiry ends, and distributed computing resources are fully utilized. The project service cloud platform can continuously optimize the recommendation model according to user feedback, so that the project service cloud platform can be quickly adapted to the change of user demands, intelligent recommendation analysis of content interaction is further realized, the accuracy of project information recommendation is improved, and meanwhile, the confidentiality of project information recommendation is improved.
Drawings
FIG. 1 is a schematic diagram of an embodiment of a method for recommending item information based on content interaction according to an embodiment of the present invention;
FIG. 2 is a flow chart of model collaborative updating in an embodiment of the present invention;
FIG. 3 is a flow chart of collaborative updating of parameters according to an embodiment of the present invention;
FIG. 4 is a flow chart of personalized project recommendation in an embodiment of the invention;
FIG. 5 is a schematic diagram of an embodiment of a content interaction-based project information recommendation system according to an embodiment of the present invention;
fig. 6 is a schematic diagram of an embodiment of a content interaction-based item information recommendation device according to an embodiment of the present invention.
Detailed Description
The embodiment of the invention provides a project information recommending method and system based on content interaction, which are used for realizing intelligent recommending analysis of the content interaction, improving the accuracy of project information recommendation and improving the confidentiality of project information recommendation. The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.
For easy understanding, a specific flow of an embodiment of the present invention is described below, referring to fig. 1, and an embodiment of a method for recommending item information based on content interaction in an embodiment of the present invention includes:
s101, constructing a project information interaction network based on a preset project service cloud platform and a plurality of first project information inquiry terminals, initializing a model through the project service cloud platform to generate an initialized model parameter set, and encrypting and distributing the initialized model parameter set to the plurality of first project information inquiry terminals according to the project information interaction network;
it will be appreciated that the execution subject of the present invention may be a content interaction-based project information recommendation system, and may also be a terminal or a server, which is not limited herein. The embodiment of the invention is described by taking a server as an execution main body as an example.
Specifically, the server identifies a plurality of first project information query ends based on a preset project service cloud platform, and the query ends represent the applicants in different fields, such as scientific research, environmental protection, culture and the like. And constructing an item information interaction network by means of the preset item service cloud platform and the identified inquiry terminal. This network will serve as a bridge for information transfer, connecting governments and applicants. And configuring a first training model to be trained on the project service cloud platform in the project information interaction network. This model will be used later for initialization. And initializing a first training model through the project service cloud platform to generate an initial model parameter set. These parameters will be used for subsequent collaborative updating of the model. In order to ensure the security of parameter transmission, the initial model parameter set needs to be encrypted. This employs modern encryption algorithms to ensure that parameters are not compromised during transmission. And determining a parameter transmission path between the project service cloud platform and each first project information inquiry terminal through the project information interaction network so as to ensure that the encryption parameter set can be safely transmitted to each inquiry terminal. And each first item information inquiry end uses the initialized encryption parameter set to carry out local model training to obtain a second training model. Under the background of government project declaration, query ends in different fields can perform local training on the model according to local data and characteristics, so that a more targeted model is obtained. Through the second training model, each inquiry end can provide personalized project recommendation for the applicant, and matching is carried out according to the background, the requirements and the interests of the applicant. Such personalized recommendations help to improve applicant satisfaction while optimizing resource allocation. For example, the server is exemplified in the education field. The government project service platform is used as a core and constructs a project information interaction network with a plurality of first project information inquiry terminals (applicants). These queries represent different educational institutions, such as universities, colleges, training institutions, etc. In the project information interaction network, a government project service platform is configured with a first training model to be trained for initialization. And initializing a first training model through the project service cloud platform to generate an initial model parameter set. These parameters will be encrypted to ensure the security of the transmission. And the first project information inquiry end of the different education institutions performs local model training by using the initialized encryption parameter set to obtain a second training model. For example, a university's end of inquiry may train a more accurate model based on local course, teaching and research directions. Through the second training model, the inquiring end of each education institution can provide individualized project recommendation for teachers and students. For example, a university may provide recommendations for its research students related to scientific projects, and for the undergraduates related to academic contests and practices. Such personalized recommendations help to improve satisfaction with educational institutions and applicants.
S102, carrying out local model training on a plurality of first project information inquiry terminals through an initialized model parameter set, and carrying out gradient parameter updating on gradient information obtained by training each project information inquiry terminal to generate a gradient updating model parameter set;
specifically, the server decrypts the initialized encryption parameter set through a plurality of first item information inquiry terminals to obtain an initialized model parameter set. These parameters will serve as the starting point for local model training. Each inquiry end is matched with a corresponding second training model according to the local data set, and is ready for local model training. And carrying out local model training by each first project information inquiry end by means of the local training data set and the decrypted initialization model parameter set to obtain first gradient information. The gradient information represents training results of the model on the local data, namely the fitting degree of the model to the data of the inquiring end. In order to achieve gradient updating on the premise of protecting data privacy, a first cloud platform node and a second cloud platform node are built in the project service cloud platform. Meanwhile, a first query end node and a second query end node are respectively arranged in each first item information query end. Each first query end node is responsible for converting the encrypted first gradient information into second gradient information, and transmitting the information to the project service cloud platform through a preset parameter transmission path. And in the project service cloud platform, the second cloud platform node decrypts the transmitted second gradient information, performs aggregation operation and merges the information into target gradient information. The aggregation operation can adopt methods such as weighted average, gradient accumulation and the like so as to ensure that the contributions of all the inquiry terminals are reasonably integrated. And updating the initialized model parameter set by the project service cloud platform through the target gradient information to generate a gradient update model parameter set. The parameter sets integrate the information of the local training results of all the inquiry terminals to form finer model parameters. For example, consider a research institution using this method to obtain recommendations related to a research project. Different scientific institutions will perform local model training based on their local data sets, generating first gradient information. These gradient information are finally integrated into target gradient information for updating model parameters of the project service cloud platform through encryption transmission and aggregation operation at the project service cloud platform. The scientific research institution will get more personalized project recommendations. In the embodiment, the data privacy is respected, and the contribution of each inquiry end is reasonably integrated, so that more accurate and privacy-protected project recommendation is realized.
S103, acquiring government project declaration data through a plurality of first project information inquiry terminals, and performing sensitive data encryption on the government project declaration data to acquire encrypted project declaration data;
the method is characterized in that a plurality of first project information inquiry terminals acquire government project reporting data through personal information filled in by a user and uploaded project reporting materials. Such data includes the applicant's name, contact, affiliated institution, project goal, plan, budget, etc. In order to protect user privacy, government project declaration data requires differential privacy analysis. This helps determine privacy budgets, including privacy loss and sensitivity. The privacy loss refers to the degree of privacy disclosure caused by revealing individual data, and the sensitivity measures the sensitivity degree of the data. Privacy parameters including privacy protection level and noise intensity are set for government project declared data. The level of privacy protection will determine the severity of the privacy protection, and the noise strength will affect the degree of encryption and perturbation of the data. Sensitive data information is extracted from government project declaration data, which is project budget, planning details, scientific achievements, etc., and needs to be specially protected from leakage. And carrying out noise addition on the sensitive data information according to the privacy budget and the privacy parameters. This noise is one way to preserve the privacy of an individual by adding noise such that the data becomes indistinguishable. After noise is added, the generated data is the encrypted project declaration data. For example, consider a research institution applying government support for a research project involving disease control. The applicant needs to provide personal information and project details, including disease study plans and associated budgets. By using the method, the government can establish an information interaction network with the organization, acquire project requirements of the organization, and then conduct differential privacy analysis on sensitive data information and set privacy parameters. And according to the privacy budget and the parameters, carrying out noise addition on the project details to generate encrypted project declaration data, so that the privacy of an applicant is ensured to be protected.
S104, model feature training is carried out through encryption project declaration data and gradient update model parameter sets to obtain feature model parameter sets, and the feature model parameter sets are sent to a plurality of first project information inquiry terminals to carry out model collaborative update to obtain a plurality of inquiry terminal recommendation models;
specifically, a plurality of target features are obtained from the first training model, and the features may include declaration history, hobbies, and the like. These target features will be used for personalized project recommendations to provide a recommendation scheme that better meets the needs of the user. And setting a target data format of the encrypted item declaration data according to the plurality of target features. Then, a target feature data set is extracted from the encrypted item declaration data according to the target data format and the target feature. This will provide key information about the user for subsequent model feature training. And performing feature training on the first training model by using the extracted target feature data set and the gradient update model parameter set. The method aims at generating a characteristic model parameter set by combining target characteristics and model parameters, so that the model can better capture interests and preferences of a user. And carrying out parameter encryption processing on the characteristic model parameter set to ensure safe transmission of model parameters. The encrypted characteristic model parameter set is transmitted to a plurality of first item information inquiry terminals and is ready for model collaborative updating. And carrying out parameter decryption processing on the encrypted characteristic model parameter set through a second query end node in the plurality of first item information query ends. And after decryption, carrying out collaborative updating on the second training model by using the parameters so as to obtain a personalized recommendation model. These recommendation models will combine the user's target features and model parameters to provide more accurate recommendation results for each query. For example, consider a research project application scenario where a government agency wishes to recommend appropriate project funding schemes to the applicant based on his history of research and interests. By the content interaction-based method, the government obtains encrypted declaration data and extracts the applicant's scientific history and interests. The government then uses these target feature data sets and gradients to update model parameters, which are used to perform feature training on the model to generate a feature model parameter set. The parameter sets are encrypted and then sent to the first item information inquiry terminals. And carrying out parameter decryption on each query end, and carrying out collaborative updating on the second training model by using the characteristic model parameters to obtain recommended models of a plurality of query ends. These models will combine the user's history of scientific research and interests to generate more accurate project recommendations for each query.
S105, carrying out model integration on a plurality of inquiry end recommendation models through the project service cloud platform to generate a target global recommendation model;
specifically, the server performs model integration on a plurality of query end recommendation models through the project service cloud platform according to preset model weights. The method aims at integrating recommendation models of different query ends to obtain an initial global recommendation model. Model integration may use a variety of methods, such as weighted averaging, model fusion, etc., to ensure model contribution equalization at different querying ends. Model verification is performed on the initial global recommendation model to evaluate its performance on unknown data. The reliability of the initial model is ensured. Through model verification, a plurality of recommendation prediction results can be obtained, and the results reflect recommendation effects of the initial global recommendation model on different samples. And providing a plurality of recommendation prediction results for the user, and enabling the user to feed back the recommendation results. User feedback analysis is a key element in further optimizing the recommendation model by collecting and analyzing feedback data of the user. The user feedback data includes a user ID, a recommendation item ID, an interaction type, and a time stamp. By analyzing the feedback of the user, the user can know which recommendations get the satisfaction of the user and which need improvement, so that the recommendation model is better optimized. And carrying out model parameter optimization on the initial global recommendation model based on the feedback data of the plurality of users and the recommendation prediction result. The goal of the optimization is to adjust the model parameters to better adapt to the interests and preferences of the user. Through this process, a target global recommendation model is generated that will more accurately provide personalized item recommendations to the user. For example, assume that a government sponsored platform is dedicated to providing personalized project recommendations for a research project applicant. By a content interaction-based method, governments integrate recommendation models of multiple query ends together to generate an initial global recommendation model. The platform then performs model verification on the initial model, evaluating its performance on the actual data. After model verification, the government recommended the applicant some items, and collected their feedback data, including which items were popular and which needed improvement. By analyzing the user's feedback data, the government has gained important insight into the user's preferences. For example, one applicant is interested in environmental protection type items, while another is more concerned with scientific innovation type items. Based on these feedback data, the government performs model parameter optimization on the initial global recommendation model, adjusting the model to better accommodate the applicant's needs. The government has generated a target global recommendation model that can more accurately provide the applicant with recommendations of items that meet his interests and preferences.
S106, acquiring content interaction behavior data of the target user based on the second item information inquiry end, and inputting the content interaction behavior data into a target global recommendation model to conduct personalized item recommendation, so as to obtain a target recommendation result.
Specifically, the content interaction behavior data of the target user is obtained through the second item information inquiry terminal. The behavior data may include interactive behavior of a user browsing records, search records, praise, comments, etc. on the platform. And carrying out feature extraction aiming at each interaction behavior to acquire a plurality of content interaction behavior features. These features may include time stamps, interaction types, keywords, tags, etc. And encoding the obtained plurality of content interaction behavior features to generate content interaction behavior encoding vectors, wherein the purpose is to convert the interaction behavior features into numerical representations for subsequent model calculation. Common encoding methods include word embedding, single-hot encoding, and the like. And inputting the generated content interaction behavior coding vector into a target global recommendation model. The model is generated in a previous step, aiming at providing personalized project recommendations for the user according to their content interaction behavior. And carrying out personalized item recommendation on the input content interaction behavior data through a plurality of inquiry end recommendation models in the target global recommendation model to obtain an initial recommendation result of each inquiry end recommendation model. And carrying out result fusion on the initial recommendation result of each query end recommendation model. The method of result fusion may include weighted averaging, rank fusion, etc., to ensure that recommendation opinions of multiple querying ends are comprehensively considered. After fusion, a target recommendation result is obtained, wherein the result represents personalized project recommendation generated according to the content interaction behavior of the user. For example, assume that a government project declaration platform is dedicated to providing personalized project recommendation services for researchers. And the platform acquires the content interaction behavior data of the user from the second item information inquiry end through a content interaction-based method. For example, user a browses a plurality of scientific articles about environmental protection in the past several months, and praise and comment. The platform performs feature extraction on the behavior data to obtain a plurality of content interaction behavior features such as time stamps, interaction types, keywords and the like. The platform converts the features into content interaction behavior encoding vectors and inputs the content interaction behavior encoding vectors into the target global recommendation model. The model considers the interactive behavior of the user, and generates personalized project recommendation for the user by utilizing the previous query end recommendation model. For user a, the model may recommend scientific projects related to environmental protection to meet its interests. After initial recommended results of all the inquiry terminals are obtained, the platform performs result fusion on the results. For example, the platform performs sorting fusion on the recommendation results of different query ends to obtain a final target recommendation result. The result will include a series of scientific items that match the interests of the user, which are generated by taking into account the user's content interaction behavior.
In the embodiment of the invention, the project service cloud platform is used for carrying out model initialization, generating an initialization model parameter set, and encrypting and distributing the initialization model parameter set to a plurality of first project information inquiry terminals; performing local model training and gradient parameter updating to generate a gradient updating model parameter set; encrypting the sensitive data to obtain the encrypted project declaration data; model feature training is carried out to obtain a feature model parameter set, and model collaborative updating is carried out to obtain a plurality of inquiry end recommendation models; performing model integration to generate a target global recommendation model; the method and the device have the advantages that the content interaction behavior data are input into the target global recommendation model to conduct personalized project recommendation, and target recommendation results are obtained. Personal information and declaration materials of the user are transmitted and processed in an encrypted state, so that trust feeling of the user on the platform is enhanced, and the user is willing to participate and share data. The federal learning enables training of the model to be distributed on a plurality of item information inquiry ends, and distributed computing resources are fully utilized. The project service cloud platform can continuously optimize the recommendation model according to user feedback, so that the project service cloud platform can be quickly adapted to the change of user demands, intelligent recommendation analysis of content interaction is further realized, the accuracy of project information recommendation is improved, and meanwhile, the confidentiality of project information recommendation is improved.
In a specific embodiment, the process of executing step S101 may specifically include the following steps:
(1) Carrying out item information inquiry end identification based on a preset item service cloud platform to obtain a plurality of first item information inquiry ends;
(2) Constructing a project information interaction network based on a preset project service cloud platform and a plurality of first project information inquiry ends;
(3) Respectively configuring a first training model to be trained in a project service cloud platform, and respectively creating a second training model to be trained in a plurality of first project information inquiry terminals;
(4) Carrying out model initialization on the first training model through the project service cloud platform to generate an initialized model parameter set;
(5) Carrying out parameter encryption processing on the initialized model parameter set to generate an initialized encrypted parameter set, and determining parameter transmission paths corresponding to the project service cloud platform and the plurality of first project information inquiry ends through a project information interaction network;
(6) And distributing the initialized encryption parameter set to corresponding second training models in the plurality of first item information inquiry terminals through the parameter transmission paths.
Specifically, the server identifies and verifies the user through a preset project service cloud platform. This may be achieved by means of login information, account binding, etc. After the user successfully logs in, the server can identify the user as the item information inquiry end, namely the first item information inquiry end. On the basis of the project service cloud platform, a project information interaction network is established with a plurality of first project information inquiry terminals. The network may be a distributed network for exchanging information between the various querying ends. Through the network, different inquiry terminals can share information and exchange data and participate in training and updating of the model together. And configuring a first training model to be trained in the project service cloud platform as an initial model. Meanwhile, second training models to be trained are respectively created in the plurality of first project information inquiry terminals. These models will be used for personalized local training to meet the needs of different querying ends. And initializing a first training model through the project service cloud platform to generate an initial model parameter set. These parameters are used as starting points for the training process. And carrying out parameter encryption processing on the initialized model parameter set to protect confidentiality of the model. And determining parameter transmission paths between the project service cloud platform and a plurality of first project information inquiry terminals through the project information interaction network. On this path, the encrypted initialization model parameter set will be transmitted to the corresponding second training model for subsequent local training. For example, consider a scientific project declaration platform where a government wishes to provide personalized project declaration advice to researchers in different fields. In this case, the government establishes a preset project based service cloud platform which is connected to a plurality of first project information inquiring ends. For example, researcher A logs into the platform and is identified as the first item information query. The server establishes an interactive network with a, allowing her to browse, search for item information on the platform. At the same time, the government has configured a generic first training model in the platform. In the platform, a performs some initial operations and the server then performs a personalized creation for her second training model. The project service cloud platform also initializes the first training model, and generates initial model parameters. These parameters are encrypted to protect their security and then transmitted over the interaction network to the second training model of a. The second training model of a now has some initial parameters, as well as her own local data. She can begin local model training and model optimization based on her own needs. Other researchers may perform similar operations on their respective first item information querying side to form an ecosystem that shares information and co-learns.
In a specific embodiment, the process of executing step S102 may specifically include the following steps:
(1) Performing parameter decryption processing on the initialized encryption parameter set through a plurality of first item information inquiry terminals to obtain an initialized model parameter set;
(2) Respectively matching a local training data set of each second training model through a plurality of first project information inquiry terminals;
(3) Performing local model training on the second training models of the plurality of first project information inquiry ends based on the local training data set and the initialized model parameter set to obtain first gradient information of each second training model;
(4) Creating corresponding first cloud platform nodes and second cloud platform nodes in the project service cloud platform, and creating corresponding first query end nodes and second query end nodes at each first project information query end;
(5) Encrypting the first gradient information through a first query end node of each first project information query end to obtain second gradient information, and transmitting the second gradient information to a project service cloud platform through a parameter transmission path;
(6) Decrypting the second gradient information through a second cloud platform node in the project service cloud platform to obtain third gradient information, and performing aggregation operation on the third gradient information to obtain target gradient information;
(7) And updating the initialized model parameter set of the project service cloud platform through the target gradient information to obtain a gradient update model parameter set corresponding to the project service cloud platform.
Specifically, the server performs parameter decryption processing on the initialized encryption parameter set through a plurality of first item information inquiry terminals to obtain an initialized model parameter set. Each querying end can decrypt these parameters for model training locally. Meanwhile, each query end is matched with a corresponding second training model according to the local training data set. Based on the local training data set and the initialized model parameter set, each first project information inquiry end carries out local model training on the second training model. In this process, each query calculates the first gradient information of the model training. And creating a first cloud platform node and a second cloud platform node which correspond to each other in the project service cloud platform. Meanwhile, a first query end node and a second query end node corresponding to each first item information query end are created. And encrypting the first gradient information obtained by calculation through a first query end node of each first item information query end, so as to obtain second gradient information. And transmitting the encrypted second gradient information to the project service cloud platform through the parameter transmission path. And decrypting the information of the second gradient information by a second cloud platform node in the project service cloud platform to obtain third gradient information. And performing aggregation operation on the plurality of third gradient information to obtain target gradient information. Through the target gradient information, the project service cloud platform can update the initialized model parameter set to obtain a corresponding gradient updated model parameter set. These updated parameters may be used in subsequent model recommendations and personalized project recommendations. For example, consider a government project declaration platform in the educational domain where a government desires to provide personalized project recommendations to an educator to facilitate educational innovation. In this platform, the educator logs in as a first project information query, and the government configures a first training model to be trained based on a preset model. For example, the education machine B logs on to the platform, and the server recognizes it as a first item information query end and creates a second training model for it. The platform initializes the first training model, encrypts the parameters of the first training model and transmits the parameters to the second training model of B. And B, locally performing model training on the second training model by using own local data, and calculating to obtain first gradient information. Then, the first query end node encrypts the first gradient information and transmits the first gradient information to the project service cloud platform through the network. And in the project service cloud platform, the second cloud platform node decrypts the second gradient information to obtain third gradient information. The platform collects a plurality of third gradient information and performs aggregation operation to obtain target gradient information. And updating the initialized model parameter set of the project service cloud platform through the target gradient information to form a gradient updated model parameter set, and providing support for subsequent project recommendation.
In a specific embodiment, the process of executing step S103 may specifically include the following steps:
(1) Acquiring government project declaration data through a plurality of first project information inquiry terminals, wherein the government project declaration data comprises: personal information filled in by a user and uploaded project declaration materials;
(2) Differential privacy analysis is carried out on government project declaration data to obtain privacy budget, wherein the privacy budget comprises the following steps: privacy loss and sensitivity;
(3) Setting privacy parameters of government project declaration data, wherein the privacy parameters comprise privacy protection level and noise intensity;
(4) Extracting characteristic information of government project declaration data to obtain sensitive data information;
(5) And carrying out noise addition on the sensitive data information according to the privacy budget and the privacy parameters to generate encrypted project declaration data.
Specifically, the server obtains government project declaration data through a plurality of first project information inquiry terminals. Such data includes personal information (e.g., name, contact) filled out by the user, and uploaded project declaration material (e.g., project plan, budget, etc.). Differential privacy analysis is performed on government project declaration data to evaluate privacy protection requirements. By differential privacy analysis, a privacy budget can be determined, including privacy loss and sensitivity. The privacy loss measures the degree of privacy disclosure under a certain query, and the sensitivity measures the degree of influence of a change of a certain item in the dataset on the query result. Based on the result of the differential privacy analysis, privacy parameters of government project declaration data are set. These parameters include the level of privacy protection (i.e., the maximum level of privacy disclosure allowed) and the noise strength (i.e., the added noise level). And extracting characteristic information of the government project declaration data, and obtaining sensitive data information from the characteristic information. For example, key information such as budget and project plan is extracted from the uploaded project declaration material. And carrying out noise addition on the sensitive data information according to the set privacy parameters. The introduction of noise can obscure the original data, thereby reducing the risk of privacy disclosure. The added noise may be random to ensure privacy security of the data. For example, consider a research project reporting platform where researchers need to submit applications to governments to obtain research funding support. The government wishes to protect the privacy of researchers while ensuring the accuracy of research applications. In this platform, a researcher submits a research project application to a government as a first project information query. For example, researcher A logs onto the platform and submits a research project application. The server obtains personal information (e.g., name, mailbox) she fills in and uploaded study materials (e.g., study plan, budget). Next, the server performs differential privacy analysis on the data, calculating privacy loss and sensitivity, and thus obtaining privacy budget. The government sets the privacy protection level and the noise intensity according to the result of the differential privacy analysis. For example, governments require that noise be added to sensitive data such as a research budget in cases where the level of privacy protection is high. Then, the server performs noise addition on sensitive data such as research budget and the like to generate encrypted research project declaration data. The introduction of noise blurs the original data, thereby reducing the risk of leakage of sensitive information. The government may obtain enough information to assess the feasibility and contribution of the research application, while protecting the privacy of the researchers.
In a specific embodiment, as shown in fig. 2, the process of executing step S104 may specifically include the following steps:
s201, acquiring a plurality of target features corresponding to a first training model, wherein the plurality of target features comprise declaration histories and hobbies;
s202, setting a target data format of the encrypted item declaration data according to a plurality of target features, and extracting a target feature data set corresponding to the encrypted item declaration data according to the target data format and the plurality of target features;
s203, performing model training on the first training model through the target feature data set and the gradient update model parameter set to obtain a feature model parameter set;
s204, carrying out parameter encryption processing on the feature model parameter set to obtain an encrypted feature model parameter set, and transmitting the encrypted feature model parameter set to a plurality of first item information inquiry terminals;
s205, performing parameter decryption processing on the encrypted characteristic model parameter set through a second query end node in the plurality of first item information query ends, and performing parameter collaborative updating on a second training model in the plurality of first item information query ends through the characteristic model parameter set to obtain a plurality of query end recommendation models.
Specifically, the server obtains a plurality of target features for the first training model, the features including a reporting history and interests. The declaration history may include a record of project applications submitted by the user in the past, and the hobbies may be browsing behavior, praise, comments, etc. of the user on the platform. Based on the obtained plurality of target features, the server sets a target data format of the encrypted item declaration data. This includes deciding how to organize the data and encode it so that it can be used in subsequent model training. Based on the target data format and the plurality of target features, the server extracts a target feature data set from the encrypted project declaration data for use in subsequent model training. And combining the target characteristic data set with the gradient updating model parameter set, and carrying out model training on the first training model. This step aims to optimize the model to better capture the interests and preferences of the user based on the user's target features. In this process, the server gets a set of feature model parameters that will be used to generate personalized recommendations. In order to protect the privacy of the feature model parameters, the feature model parameter set needs to be subjected to parameter encryption processing. The encrypted parameter set is transmitted to a plurality of first item information inquiring ends for use in a subsequent collaborative updating process. And performing parameter decryption processing on a second query end node in the plurality of first item information query ends, and decrypting the encrypted characteristic model parameter set into an original parameter. And carrying out parameter collaborative updating on the second training models in the plurality of first item information inquiry terminals by using the decrypted parameters. This will ensure that the recommendation model for each query is personalized optimized to better meet the needs of the user. For example, considering an online educational platform, a student may apply for various learning items. In this platform, the student is the first item information query, and the educational institution and course provider are the second item information query. The target characteristics of the students may include their learning history, hobbies and types of courses selected. For example, student A is a senior citizen who has learned multiple mathematical and physical courses. Her interests include scientific experimentation and programming. When she submits a learning item application, the platform sets a target data format of the encrypted item declaration data according to her target characteristics, and extracts information such as her learning history, interests and the like. The feature model parameter set will then be used for model training of the application of a to optimize the recommendation. These parameters are encrypted and transmitted to a plurality of educational institutions and course providers for decryption and collaborative updating of the parameters. This will result in the recommendation model for each institution and provider being personalized based on the characteristics of a, providing her with a more accurate recommendation of learning items.
In a specific embodiment, as shown in fig. 3, the process of executing step S105 may specifically include the following steps:
s301, carrying out model integration on a plurality of inquiry end recommendation models through a project service cloud platform according to preset model weights to obtain an initial global recommendation model;
s302, performing model verification on an initial global recommendation model to obtain a plurality of recommendation prediction results;
s303, performing user feedback analysis on the plurality of recommendation prediction results to obtain a plurality of user feedback data, wherein the user feedback data comprises: user ID, recommended item ID, interaction type, and time stamp;
s304, model parameter optimization is carried out on the initial global recommendation model according to the feedback data of the plurality of users and the recommendation prediction results, and a target global recommendation model is generated.
Specifically, the server performs model integration on a plurality of query end recommendation models according to preset model weights. The method adopts various integration technologies, such as weighted average, voting, stacking and the like, so as to synthesize the prediction results of the various inquiry end models and obtain an initial global recommendation model. And carrying out model verification on the initial global recommendation model, predicting by using the existing data set, and obtaining a plurality of recommendation prediction results. These predictions represent project recommendations for different users in actual use. A plurality of recommendation prediction results are provided to a user and user feedback data is collected. The user feedback data includes information such as user ID, recommended item ID, interaction type, and time stamp. These data reflect the user's acceptance of the recommendation, preferences, and interaction behavior. And performing model parameter optimization on the initial global recommendation model based on the plurality of user feedback data and the plurality of recommendation prediction results. This allows for machine-learned optimization algorithms, such as gradient descent, genetic algorithms, etc., to adjust the model parameters to more accurately adapt to the user's needs and feedback. Therefore, the server integrates the multiple query end recommendation models on the project service cloud platform according to the preset model weight, optimizes model parameters through user feedback data, and finally generates the global recommendation model of the target. The global recommendation model can more accurately conduct personalized project recommendation according to the requirements and feedback of users. For example, assume that on a government project declaration platform, a plurality of users respectively apply for projects in different fields, such as scientific research, environmental protection, education, and the like. Through model integration and parameter optimization, the server generates an initial global recommendation model. User A applies for environmental protection items, while user B applies for scientific research items. The server uses the initial model for prediction, recommending a series of environmental protection items to user a, and simultaneously recommending items suitable for the scientific research field to user B. User a selects an environmental item in the recommendation list and marks it as "interesting". This interaction is recorded as user feedback data, including user a's ID, item ID, type of interaction "interesting" and timestamp. And the server optimizes the parameters of the global recommendation model according to the feedback data of the user A so as to enable the parameters to more accurately understand the preference of the user. Next, when user B logs into the server again, the global recommendation model will recommend items to it that are more suitable for the scientific field based on its previous preferences and user feedback. By continuously collecting user feedback data and performing model optimization, the server gradually generates a more accurate and personalized global recommendation model, so that the user satisfaction is improved, and the efficiency and quality of reporting government projects are promoted.
In a specific embodiment, as shown in fig. 4, the process of executing step S106 may specifically include the following steps:
s401, acquiring content interaction behavior data of a target user based on a second item information inquiry end, and extracting features of the content interaction behavior data to obtain a plurality of content interaction behavior features;
s402, performing feature coding processing on the plurality of content interaction behavior features to generate content interaction behavior coding vectors;
s403, inputting the content interaction behavior coding vector into a target global recommendation model, and performing personalized item recommendation on the content interaction behavior data through a plurality of inquiry end recommendation models in the target global recommendation model to obtain an initial recommendation result of each inquiry end recommendation model;
s404, carrying out result fusion on the initial recommendation result of the recommendation model of each query end, generating a target recommendation result, and pushing the target recommendation result to the second item information query end for visual display.
Specifically, the server obtains content interaction behavior data of the target user through the second item information query terminal. Such data may include browsing history of the user on the platform, search records, interaction behavior, and the like. The content interaction behavior data are subjected to feature extraction, and a plurality of content interaction behavior features such as browsing times, search keywords, interaction types and the like are extracted from the content interaction behavior data. And carrying out feature coding processing on the acquired interactive behavior features of the plurality of contents. This may include single thermal encoding of the classification features, normalization of the succession of features, etc. And combining the processed characteristics to form a content interaction behavior coding vector which is used for representing the content interaction behavior mode of the user. And inputting the content interaction behavior coding vector into the target global recommendation model. The global recommendation model is obtained through the previous model integration and parameter optimization steps, and personalized project recommendation can be carried out according to the content interaction behavior characteristics of the user. And predicting the content interaction behavior data through a plurality of inquiry end recommendation models in the model to obtain an initial recommendation result of each inquiry end recommendation model. And carrying out result fusion on the initial recommendation result of each inquiry end recommendation model, and obtaining a comprehensive target recommendation result by using methods such as weighted average, voting and the like. The target recommendation result comprehensively considers the opinions of the recommendation models of the plurality of inquiry ends, and improves the accuracy and the comprehensiveness of recommendation. And finally, pushing the target recommendation result to a second item information inquiry end for visual display for viewing and selection by a user. For example, suppose that on a government project declaration platform, user A is a scientific research person who is often browsing for projects related to the scientific research field. Through the second item information inquiry end, the server acquires content interaction behavior data of the user A, including browsing a plurality of scientific research items, searching a specific research topic and the like. Through extracting the characteristics of the content interaction behavior data, the server obtains the content interaction behavior characteristics of the user A, such as browsing times, search word frequency and the like. These features are feature coded to generate content interaction behavior coding vectors. The content interaction behavior coding vector is input into a target global recommendation model, and the model is integrated through the model and optimized through parameters, so that personalized recommendation can be performed according to the content interaction behavior of the user. The model outputs an initial recommendation result of the user A, wherein the initial recommendation result comprises a plurality of projects suitable for the scientific research field. The server performs weighted fusion on initial recommendation results from a plurality of inquiry end recommendation models to obtain a comprehensive target recommendation result, and the comprehensive target recommendation result represents comprehensive recommendation opinion of the user A. This target recommendation is pushed to the second item information query and visually presented to user a, helping him to better understand the recommended item and make a selection.
The method for recommending item information based on content interaction in the embodiment of the present invention is described above, and the system for recommending item information based on content interaction in the embodiment of the present invention is described below, referring to fig. 5, an embodiment of the system for recommending item information based on content interaction in the embodiment of the present invention includes:
the initialization module 501 is configured to construct an item information interaction network based on a preset item service cloud platform and a plurality of first item information query terminals, perform model initialization through the item service cloud platform, generate an initialization model parameter set, and encrypt and distribute the initialization model parameter set to the plurality of first item information query terminals according to the item information interaction network;
the training module 502 is configured to perform local model training on the plurality of first item information query ends through the initialized model parameter set, perform gradient parameter update on gradient information obtained by training on each item information query end, and generate a gradient update model parameter set;
an obtaining module 503, configured to obtain government project declaration data through the plurality of first project information querying ends, and encrypt sensitive data of the government project declaration data to obtain encrypted project declaration data;
The updating module 504 is configured to perform model feature training through the encrypted project declaration data and the gradient updating model parameter set, obtain a feature model parameter set, and send the feature model parameter set to the plurality of first project information query ends to perform model collaborative updating, so as to obtain a plurality of query end recommendation models;
the integration module 505 is configured to perform model integration on the plurality of query end recommendation models through the project service cloud platform, and generate a target global recommendation model;
and the recommendation module 506 is configured to obtain content interaction behavior data of the target user based on the second item information query end, and input the content interaction behavior data into the target global recommendation model to perform personalized item recommendation, so as to obtain a target recommendation result.
Through the cooperative cooperation of the components, the project service cloud platform is used for carrying out model initialization to generate an initialized model parameter set, and the initialized model parameter set is distributed to a plurality of first project information inquiry terminals in an encrypted mode; performing local model training and gradient parameter updating to generate a gradient updating model parameter set; encrypting the sensitive data to obtain the encrypted project declaration data; model feature training is carried out to obtain a feature model parameter set, and model collaborative updating is carried out to obtain a plurality of inquiry end recommendation models; performing model integration to generate a target global recommendation model; the method and the device have the advantages that the content interaction behavior data are input into the target global recommendation model to conduct personalized project recommendation, and target recommendation results are obtained. Personal information and declaration materials of the user are transmitted and processed in an encrypted state, so that trust feeling of the user on the platform is enhanced, and the user is willing to participate and share data. The federal learning enables training of the model to be distributed on a plurality of item information inquiry ends, and distributed computing resources are fully utilized. The project service cloud platform can continuously optimize the recommendation model according to user feedback, so that the project service cloud platform can be quickly adapted to the change of user demands, intelligent recommendation analysis of content interaction is further realized, the accuracy of project information recommendation is improved, and meanwhile, the confidentiality of project information recommendation is improved.
Fig. 5 above describes the content interaction-based item information recommendation system in the embodiment of the present invention in detail from the point of view of the modularized functional entity, and the content interaction-based item information recommendation device in the embodiment of the present invention is described in detail from the point of view of hardware processing.
Fig. 6 is a schematic structural diagram of a content interaction-based item information recommendation device 600 according to an embodiment of the present invention, where the content interaction-based item information recommendation device 600 may have relatively large differences according to configuration or performance, and may include one or more processors (central processing units, CPU) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) storing application programs 633 or data 632. Wherein the memory 620 and the storage medium 630 may be transitory or persistent storage. The program stored in the storage medium 630 may include one or more modules (not shown), each of which may include a series of instruction operations on the content interaction-based item information recommendation device 600. Still further, the processor 610 may be configured to communicate with the storage medium 630 to execute a series of instruction operations in the storage medium 630 on the content interaction-based item information recommendation device 600.
The content interaction based project information recommendation device 600 may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input/output interfaces 660, and/or one or more operating systems 631, such as Windows Server, mac OS X, unix, linux, freeBSD, and the like. It will be appreciated by those skilled in the art that the content interaction based item information recommendation device structure shown in fig. 6 does not constitute a limitation of the content interaction based item information recommendation device, and may include more or less components than illustrated, or may combine certain components, or may be a different arrangement of components.
The present invention also provides a content interaction-based item information recommendation device, which includes a memory and a processor, where the memory stores computer readable instructions that, when executed by the processor, cause the processor to execute the steps of the content interaction-based item information recommendation method in the above embodiments.
The present invention also provides a computer readable storage medium, which may be a non-volatile computer readable storage medium, and may also be a volatile computer readable storage medium, where instructions are stored in the computer readable storage medium, when the instructions are executed on a computer, cause the computer to perform the steps of the content interaction-based project information recommendation method.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.
The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. The project information recommending method based on the content interaction is characterized by comprising the following steps of:
constructing an item information interaction network based on a preset item service cloud platform and a plurality of first item information inquiry terminals, initializing a model through the item service cloud platform to generate an initialized model parameter set, and encrypting and distributing the initialized model parameter set to the plurality of first item information inquiry terminals according to the item information interaction network;
local model training is carried out on the plurality of first project information inquiry terminals through the initialization model parameter set, gradient parameter updating is carried out on gradient information obtained through training of each project information inquiry terminal, and a gradient updating model parameter set is generated;
Acquiring government project declaration data through the plurality of first project information inquiry terminals, and performing sensitive data encryption on the government project declaration data to obtain encrypted project declaration data;
model feature training is carried out through the encrypted project declaration data and the gradient updating model parameter set to obtain a feature model parameter set, and the feature model parameter set is sent to the plurality of first project information inquiry terminals to carry out model collaborative updating to obtain a plurality of inquiry terminal recommendation models;
carrying out model integration on the plurality of inquiry end recommendation models through the project service cloud platform to generate a target global recommendation model;
and acquiring content interaction behavior data of the target user based on the second item information inquiry terminal, and inputting the content interaction behavior data into the target global recommendation model to conduct personalized item recommendation to obtain a target recommendation result.
2. The content interaction-based project information recommendation method according to claim 1, wherein the building a project information interaction network based on a preset project service cloud platform and a plurality of first project information querying terminals, and performing model initialization through the project service cloud platform, generating an initialization model parameter set, and encrypting and distributing the initialization model parameter set to the plurality of first project information querying terminals according to the project information interaction network comprises:
Carrying out item information inquiry end identification based on a preset item service cloud platform to obtain a plurality of first item information inquiry ends;
constructing a project information interaction network based on a preset project service cloud platform and a plurality of first project information inquiry ends;
respectively configuring a first training model to be trained in the project service cloud platform, and respectively creating a second training model to be trained in the plurality of first project information inquiry terminals;
carrying out model initialization on the first training model through the project service cloud platform to generate an initialized model parameter set;
performing parameter encryption processing on the initialized model parameter set to generate an initialized encrypted parameter set, and determining parameter transmission paths corresponding to the project service cloud platform and the plurality of first project information inquiry ends through the project information interaction network;
and distributing the initialized encryption parameter set to corresponding second training models in the plurality of first project information inquiry terminals through the parameter transmission path.
3. The content interaction-based project information recommendation method according to claim 2, wherein the performing local model training on the plurality of first project information querying terminals through the initialization model parameter set, and performing gradient parameter update on gradient information obtained by training on each project information querying terminal, generating a gradient update model parameter set includes:
Performing parameter decryption processing on the initialized encryption parameter set through the plurality of first item information inquiry terminals to obtain the initialized model parameter set;
respectively matching the local training data set of each second training model through the plurality of first project information inquiry terminals;
performing local model training on the second training models of the plurality of first project information inquiry ends based on the local training data set and the initialization model parameter set to obtain first gradient information of each second training model;
creating corresponding first cloud platform nodes and second cloud platform nodes in the project service cloud platform, and creating corresponding first query end nodes and second query end nodes at each first project information query end;
encrypting the first gradient information through a first query end node of each first project information query end to obtain second gradient information, and transmitting the second gradient information to the project service cloud platform through the parameter transmission path;
decrypting the second gradient information through a second cloud platform node in the project service cloud platform to obtain third gradient information, and performing aggregation operation on the third gradient information to obtain target gradient information;
And updating the initialized model parameter set of the project service cloud platform through the target gradient information to obtain a gradient update model parameter set corresponding to the project service cloud platform.
4. The content interaction-based project information recommendation method according to claim 1, wherein the obtaining government project declaration data through the plurality of first project information inquiring ends and performing sensitive data encryption on the government project declaration data to obtain encrypted project declaration data includes:
acquiring government project declaration data through the plurality of first project information inquiry terminals, wherein the government project declaration data comprises: personal information filled in by a user and uploaded project declaration materials;
differential privacy analysis is carried out on the government project declaration data to obtain privacy budget, wherein the privacy budget comprises the following steps: privacy loss and sensitivity;
setting privacy parameters of the government project declaration data, wherein the privacy parameters comprise privacy protection level and noise intensity;
extracting characteristic information of the government project declaration data to obtain sensitive data information;
and carrying out noise addition on the sensitive data information according to the privacy budget and the privacy parameters to generate encrypted project declaration data.
5. The content interaction-based project information recommendation method according to claim 3, wherein the performing model feature training by the encrypted project declaration data and the gradient update model parameter set to obtain a feature model parameter set, and sending the feature model parameter set to the plurality of first project information query terminals to perform model collaborative update to obtain a plurality of query terminal recommendation models includes:
acquiring a plurality of target features corresponding to the first training model, wherein the target features comprise reporting histories and hobbies;
setting a target data format of the encrypted item declaration data according to the target features, and extracting a target feature data set corresponding to the encrypted item declaration data according to the target data format and the target features;
model training is carried out on the first training model through the target characteristic data set and the gradient updating model parameter set to obtain a characteristic model parameter set;
carrying out parameter encryption processing on the characteristic model parameter set to obtain an encrypted characteristic model parameter set, and transmitting the encrypted characteristic model parameter set to the plurality of first item information inquiry terminals;
And carrying out parameter decryption processing on the encrypted characteristic model parameter set through a second query end node in the plurality of first item information query ends, and carrying out parameter collaborative updating on a second training model in the plurality of first item information query ends through the characteristic model parameter set to obtain a plurality of query end recommendation models.
6. The content interaction-based project information recommendation method according to claim 1, wherein the performing model integration on the plurality of query-side recommendation models through the project service cloud platform to generate a target global recommendation model comprises:
according to preset model weights, carrying out model integration on the plurality of inquiry end recommendation models through the project service cloud platform to obtain an initial global recommendation model;
performing model verification on the initial global recommendation model to obtain a plurality of recommendation prediction results;
and carrying out user feedback analysis on the plurality of recommendation prediction results to obtain a plurality of user feedback data, wherein the user feedback data comprises: user ID, recommended item ID, interaction type, and time stamp;
and according to the plurality of user feedback data and the plurality of recommendation prediction results, performing model parameter optimization on the initial global recommendation model to generate a target global recommendation model.
7. The content interaction-based item information recommendation method according to claim 1, wherein the obtaining content interaction behavior data of a target user based on the second item information query terminal, and inputting the content interaction behavior data into the target global recommendation model for personalized item recommendation, and obtaining a target recommendation result, includes:
acquiring content interaction behavior data of a target user based on a second item information inquiry end, and extracting features of the content interaction behavior data to obtain a plurality of content interaction behavior features;
performing feature coding processing on the plurality of content interaction behavior features to generate content interaction behavior coding vectors;
inputting the content interaction behavior coding vector into the target global recommendation model, and recommending personalized items to the content interaction behavior data through a plurality of inquiry end recommendation models in the target global recommendation model to obtain an initial recommendation result of each inquiry end recommendation model;
and carrying out result fusion on the initial recommendation result of the recommendation model of each query end, generating a target recommendation result, and pushing the target recommendation result to the second item information query end for visual display.
8. A content interaction-based item information recommendation system, comprising:
the initialization module is used for constructing an item information interaction network based on a preset item service cloud platform and a plurality of first item information inquiry terminals, carrying out model initialization through the item service cloud platform, generating an initialization model parameter set, and encrypting and distributing the initialization model parameter set to the plurality of first item information inquiry terminals according to the item information interaction network;
the training module is used for carrying out local model training on the plurality of first project information inquiry terminals through the initialization model parameter set, carrying out gradient parameter updating on gradient information obtained by training each project information inquiry terminal, and generating a gradient updating model parameter set;
the acquisition module is used for acquiring government project declaration data through the plurality of first project information inquiry terminals, and carrying out sensitive data encryption on the government project declaration data to acquire encrypted project declaration data;
the updating module is used for carrying out model feature training through the encryption project declaration data and the gradient updating model parameter set to obtain a feature model parameter set, and sending the feature model parameter set to the plurality of first project information inquiry terminals for model collaborative updating to obtain a plurality of inquiry terminal recommendation models;
The integration module is used for carrying out model integration on the plurality of inquiry end recommendation models through the project service cloud platform to generate a target global recommendation model;
and the recommendation module is used for acquiring content interaction behavior data of the target user based on the second item information inquiry end, inputting the content interaction behavior data into the target global recommendation model for personalized item recommendation, and obtaining a target recommendation result.
9. A content interaction-based item information recommendation apparatus, comprising: a memory and at least one processor, the memory having instructions stored therein;
the at least one processor invoking the instructions in the memory to cause the content interaction based item information recommendation device to perform the content interaction based item information recommendation method of any of claims 1-7.
10. A computer readable storage medium having instructions stored thereon, which when executed by a processor, implement the content interaction based item information recommendation method of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311439025.7A CN117390281A (en) | 2023-11-01 | 2023-11-01 | Project information recommendation method and system based on content interaction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311439025.7A CN117390281A (en) | 2023-11-01 | 2023-11-01 | Project information recommendation method and system based on content interaction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117390281A true CN117390281A (en) | 2024-01-12 |
Family
ID=89437203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311439025.7A Withdrawn CN117390281A (en) | 2023-11-01 | 2023-11-01 | Project information recommendation method and system based on content interaction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117390281A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117891531A (en) * | 2024-03-14 | 2024-04-16 | 蒲惠智造科技股份有限公司 | System parameter configuration method, system, medium and electronic equipment for SAAS software |
-
2023
- 2023-11-01 CN CN202311439025.7A patent/CN117390281A/en not_active Withdrawn
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117891531A (en) * | 2024-03-14 | 2024-04-16 | 蒲惠智造科技股份有限公司 | System parameter configuration method, system, medium and electronic equipment for SAAS software |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110189192B (en) | Information recommendation model generation method and device | |
| Imran et al. | PLORS: a personalized learning object recommender system | |
| Law et al. | A global framework of reference on digital literacy skills for indicator 4.4. 2 | |
| Zaidan et al. | Multi-criteria analysis for OS-EMR software selection problem: A comparative study | |
| CN1818941B (en) | Information processing system, information providing apparatus, information providing method, information processing apparatus, information processing method, and program | |
| CN109800885A (en) | It is determined for the rule of black box machine learning model | |
| Lu et al. | A novel TOPSIS evaluation scheme for cloud service trustworthiness combining objective and subjective aspects | |
| US10102478B2 (en) | Distributed and privacy-preserving prediction method | |
| CN101669108A (en) | A system and device for social shopping on-line | |
| CN103210415A (en) | Apparatus and method for diagnosing learning ability | |
| CN117390281A (en) | Project information recommendation method and system based on content interaction | |
| WO2019111508A1 (en) | Information processing device, information processing method, and program | |
| CN114398553A (en) | Object recommendation method and device, electronic equipment and storage medium | |
| Zhong et al. | Design of a personalized recommendation system for learning resources based on collaborative filtering | |
| Son et al. | A social learning management system supporting feedback for incorrect answers based on social network services | |
| Jibril et al. | The role of the social media brand community on consumers’ purchasing attitude | |
| CN115631008B (en) | Commodity recommendation method, device, equipment and medium | |
| CN115249082A (en) | User interest prediction method, device, storage medium and electronic equipment | |
| Huang et al. | Expert recommendation via tensor factorization with regularizing hierarchical topical relationships | |
| Aunimo et al. | Big data governance in agile and data-driven software development: A market entry case in the educational game industry | |
| KR20020062030A (en) | Business method for providing information favorites server page and computer readable medium having stored thereon computer executable instruction for performing the method | |
| RADOIU | Organization and constraints of a recommender system for MOOCs | |
| Zheng | Preference corrections: capturing student and instructor perceptions in educational recommendations | |
| AL‐Kaabi | The impact of e‐government services on customer satisfaction in the private sector: A case study of the Kingdom of Bahrain (SIJILAT), an online commercial registration | |
| JP7109749B2 (en) | CRYPTOCURRENCY TRADING SYSTEM, METHOD AND PROGRAM |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20240112 |