CN112199604A - Power user service recommendation method based on GAT - Google Patents

Power user service recommendation method based on GAT Download PDF

Info

Publication number
CN112199604A
CN112199604A CN202010932425.1A CN202010932425A CN112199604A CN 112199604 A CN112199604 A CN 112199604A CN 202010932425 A CN202010932425 A CN 202010932425A CN 112199604 A CN112199604 A CN 112199604A
Authority
CN
China
Prior art keywords
data
service
user
model
power
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010932425.1A
Other languages
Chinese (zh)
Inventor
颜宏文
王韬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Changsha University of Science and Technology
Original Assignee
Changsha University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Changsha University of Science and Technology filed Critical Changsha University of Science and Technology
Priority to CN202010932425.1A priority Critical patent/CN112199604A/en
Publication of CN112199604A publication Critical patent/CN112199604A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9536—Search customisation based on social or collaborative filtering
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply

Abstract

The application relates to the technical field of electric power, and discloses a power user service recommendation method and system architecture based on GAT. The method comprises the following steps: (S1) acquiring data related to the user in the power system. (S2) the acquired data is preprocessed to conform to the input of the model. (S3) training data and test data are constructed. (S4) establishing a power user characteristic extraction model, extracting user characteristics and preparing for recommending service for users. (S5) establishing a service feature extraction model to extract service features, thereby performing matching according to the user features to perform service recommendation. (S6) establishing a service recommendation model, and recommending the service according to the extracted user characteristics to enable the recommended service to meet the requirements of the user, thereby achieving accurate recommendation. (S7) the built model is checked using the test data, thereby evaluating the validity of the model. By the method and the device, the user requirements can be better mined, so that the user can be better served.

Description

Power user service recommendation method based on GAT
Technical Field
The application relates to the technical field of electric power, in particular to a power user service recommendation method based on GAT.
Background
In recent years, with the continuous application and penetration of information technology in a power grid, a power grid system becomes a complex information-physical system, the daily operation of the power grid generates massive data, the massive data provides abundant information for power services, and meanwhile, the method also provides a challenge for extracting knowledge from the data more effectively so as to better serve users.
The main characteristics of the users are extracted from a large amount of historical record information, which is the key point for service recommendation of power users. In the recommendation system, the initial collaborative filtering recommendation algorithm is the mainstream because it is simple and easy to understand, and in the later application, the data has sparsity, which causes that the collaborative filtering recommendation algorithm is difficult to recommend the service items needed by the user for the user; in order to solve the sparsity problem, a deep learning model is introduced into a recommendation system, and in the deep learning model, the model can more deeply mine potential features of a user and can accurately provide recommendation for the user to a certain extent. With the continuous provision of data complexity, the deep learning-based recommendation system gradually develops a soft situation, and is slightly insufficient in the capture of user features.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a power service recommendation method based on GAT (graph attachment networks), which models the characteristics of users through GAT and probability matrix decomposition, thereby realizing service recommendation to the users according to the user requirements.
The method comprises the following concrete implementation steps:
(S1) acquiring historical data, service work order data, power service data, customer complaint data and complaint work order data of a customer in the power system.
(S2) the acquired data is preprocessed to conform to the input of the model.
(S3) training data and test data are constructed.
(S4) establishing a power user characteristic extraction model, extracting user characteristics based on the historical data of the user, and preparing for recommending service for the user.
(S5) establishing a service feature extraction model, extracting service features based on the electric power service data and the user complaint data, and then matching according to the user features to recommend services.
(S6) establishing a service recommendation model, recommending the service according to the extracted user characteristics, and enabling the recommended service to meet the requirements of the user, thereby achieving accurate recommendation
(S7) the built model is checked using the test data, thereby evaluating the validity of the model.
In the step (S1), the historical data of the user includes a user number, a user name, a user gender, a station area where the user is located, a user telephone number, an account age, and a user occupation.
In the step (S1), the service work order data includes a work order number, a user number, a service type, and a service number.
In the step (S1), the power service data includes a service number, a service type, and a service name.
In the step (S1), the customer complaint data includes a complaint number, a complaint type, and a complaint name.
In the step (S1), the complaint work order data includes a work order number, a customer number, a complaint type, and a complaint number
In the step (S2), the data is preprocessed, invalid values and abnormal values in the data are removed, and the default values are supplemented by taking the average value of the attributes. And (5) adopting normalization processing for data with larger difference.
In the step (S3), for the preprocessed data, the data is divided by 8:2, 80% of the data is used for training the model, and 20% of the data is used for testing and verifying the model.
In the step (S4), a user feature extraction model is established, and in this stage, a GAT model is constructed, specifically including the following steps:
(S4-1) constructing graph data from the user data in the step (S1), correlating the interactions between the users, and thereby constructing the entire relational network graph data.
(S4-2) embedding the constructed graph data, namely converting the data to ensure that the data meets the requirement of the model and the user characteristics are convenient to extract.
(S4-3) constructing a user feature extraction model, constructing a GAT model, and extracting the selected node features according to the interaction of each node.
(S4-4) performing a user feature extraction operation.
In the step (S5), a service feature extraction model is established, and the step is constructed by using a GAT model, and specifically includes the following steps:
(S5-1) constructing graph data based on the service data in the step (S1), mainly based on the service used by the user or the complaint made.
(S5-2) embedding the constructed service graph data, namely converting the data to ensure that the data meets the requirement of the model and the service characteristics are convenient to extract.
(S5-3) constructing a service characteristic extraction model, constructing a GAT model, and extracting the selected node characteristics according to the interaction of each node.
(S5-4) performing a service feature extraction operation.
In the step (S6), a service recommendation model is established, and similarity calculation is performed according to the user characteristics and the service characteristics extracted in the steps (S4) and (S5), so that the service most needed by the user is predicted and recommended to the user.
In the step (S7), model verification is performed on the trained model to ensure reliability and accuracy of the model.
In the step (S4-1), graph data is constructed for the user data, which is mainly based on the social information of the user in the APP, that is, the user in the social list, and the user graph is constructed by using the relationship, and the users are related and influenced with each other.
In the step (S4-2), graph data is subjected to embedding processing, and node information in the graph data is mainly subjected to embedding processing to form data available for a model, so that user features can be extracted conveniently.
In the step (S5-1), the post-view data of the service data is constructed into a binary graph structure of the user-service according to the usage record, mainly based on the service record used by the user.
After the step (S5-2), graph data constructed by the service data is subjected to embedding processing, and node information in the graph data is mainly subjected to embedding processing to form data available for the model, so that service features are extracted.
Drawings
FIG. 1 is a model flowchart of a GAT-based power user service recommendation method of the present invention
FIG. 2 is a user graph data model of the GAT-based power user service recommendation method of the present invention
FIG. 3 is a service graph data model of a GAT-based power user service recommendation method of the present invention
FIG. 4 is a schematic diagram of a user service recommendation system of a GAT-based power user service recommendation method according to the present invention
Detailed Description
The invention will be further explained with reference to the drawings
As shown in fig. 1, the step (S1) acquires historical data, service work order data, power service data, customer complaint data and complaint work order data of the customer in the power system.
In the step (S1), the historical data of the user includes a user number, a user name, a user gender, a station area where the user is located, a user telephone number, an account age, and a user occupation.
In the step (S1), the service work order data includes a work order number, a user number, a service type, and a service number.
In the step (S1), the power service data includes a service number, a service type, and a service name.
In the step (S1), the customer complaint data includes a complaint number, a complaint type, and a complaint name.
In the step (S1), the complaint work order data includes a work order number, a customer number, a complaint type, and a complaint number
The acquired data is preprocessed to conform to the input of the model as shown in step (S2) of fig. 1. And removing invalid values and abnormal values in the data, and supplementing the default values by taking the average value in the attributes. And (5) adopting normalization processing for data with larger difference.
Figure BDA0002670686950000041
The above formula is DijData after normalization, dijMin (d) as sample data before normalizationj) Is the minimum value of the original sample data, max (d)j) Is the maximum value of the original sample data.
Training data and test data are constructed as shown in step (S3) of fig. 1. The data set was partitioned at an 8:2 ratio, 80% as training data set and 20% as testing data set.
As shown in fig. 1, in step (S4), a power user feature extraction model is constructed, and based on user history data, user features are extracted to prepare for user recommendation service. Wherein the step (S4) is divided into the following four parts
The step (S4-1) constructs graph data from the user data in the step (S1), and associates the interactions between the users, thereby constructing the entire relational network graph data. The graph model data is shown in fig. 2.
Step (S4-2) of embedding the constructed graph data, namely converting the data to enable the data to meet the requirement of a model and facilitate the extraction of user features, wherein p'nIs the data after the embedding.
p′n＝Embed(un) (2)
And (S4-3) constructing a user feature extraction model, constructing a GAT model, and extracting the selected node features according to the interaction of each node.
Figure BDA0002670686950000042
Figure BDA0002670686950000043
Figure BDA0002670686950000044
Wherein
Figure BDA0002670686950000045
Connecting the data in the friend set of the current user with the user data, inputting the data into a two-layer neural network, then performing Softmax processing to obtain the interactive weight between the users, and finally obtaining the user characteristics generated by the interaction between the users by combining the Attention weight as shown in a formula 4.
Step (S4-4) is to perform a user feature extraction operation.
In the step (S5) shown in fig. 1, a service feature extraction model is constructed, and service features are extracted based on the power service data and the user complaint data, and matching is performed according to the user features, so that service recommendation is performed. Wherein the step (S5) is divided into the following four parts
The step (S5-1) constructs service graph data from the service data in the step (S1), and associates the services used by the user, thereby constructing the entire service usage graph data. The graph model data is shown in fig. 3.
Step (S5-2) is to carry out embedding processing on the constructed graph data, namely to convert the service data into the form of the service model input requirement, so as to extract the service characteristic, e'nThe data after the service data is embedded.
e′n＝Embed(in) (6)
And (S5-3) constructing a GAT model according to the constructed service feature extraction model, carrying out updating according to the service used by each node user, and finally extracting service features.
Figure BDA0002670686950000051
Figure BDA0002670686950000052
Figure BDA0002670686950000053
Wherein
Figure BDA0002670686950000054
For the user who has used the service, it will
Figure BDA0002670686950000055
And e'iConnected and then input into a two-layer neural network to obtain a characteristic beta 'of each user using the service'ioObtaining an Attenttion value through Softmax processing, and finally obtaining the characteristic e of the service according to different weightsi。
Step (S5-4) performs a service feature extraction operation.
As shown in step (S6) of fig. 1, a service recommendation model is established, and service recommendation is performed according to the extracted user characteristics, so that the recommended service meets the requirements of the user, thereby achieving accurate recommendation.
Formula (II)
The established model is checked using the test data as in step (S7) shown in fig. 1, thereby evaluating the validity of the model.
A user service recommendation system architecture diagram of a power user service recommendation method based on GAT, the structure of which is shown in fig. 4. The system mainly comprises four layers, namely a data layer, a middleware layer, an algorithm layer and a service layer from bottom to top.
The data layer is mainly used for acquiring relevant data of power grid users, and is mainly acquired from a power customer marketing system, a 95598 work order system, a power utilization acquisition system and a power distribution management system at the present stage.
The middleware layer is mainly used for caching data, and the middleware is adopted for caching data due to the fact that the data volume generated by the power grid system every day is huge, so that the operation burden of the server is reduced.
The algorithm layer is mainly used for processing data and training models, and finally, the trained models are applied to service recommendation to form recommendation sequences.
The service layer is mainly used for managing the whole system, and the functions of the service layer comprise modules of user management, recommendation model management, system test management, user behavior analysis, data management, authority management and the like.
The invention provides a GAT-based power user service recommendation method and system architecture, which can better mine user characteristics and service characteristics by utilizing a neural network of a graph in consideration of the fact that the mutual influence among users can control final service decision, so that proper service can be recommended for the users, and the satisfaction degree of the users to enterprises is improved.

Claims (10)

1. A power user service recommendation method based on GAT is characterized by comprising a technical scheme and a system architecture.
2. The technical solution of the GAT-based power user service recommendation method according to claim 1 includes the steps of:
(S1) acquiring historical data, service work order data, power service data, customer complaint data and complaint work order data of a user in the power system;
(S2) preprocessing the acquired data to conform to the input of the model;
(S3) constructing training data and test data;
(S4) establishing a power user characteristic extraction model, extracting user characteristics based on historical data of the user, and preparing for recommending service for the user;
(S5) establishing a service feature extraction model, extracting service features based on the electric power service data and the user complaint data, and performing matching according to the user features to recommend services;
(S6) establishing a service recommendation model, and recommending the service according to the extracted user characteristics to enable the recommended service to meet the requirements of the user, thereby achieving accurate recommendation;
(S7) the built model is checked using the test data, thereby evaluating the validity of the model.
3. The step (S1) of claim 2, wherein the historical data of the user includes a user number, a user name, a user gender, a region where the user is located, a user telephone number, an account age, a user occupation;
in the step (S1), the service work order data includes a work order number, a user number, a service type, and a service number;
in the step (S1), the power service data includes a service number, a service type, and a service name;
in the step (S1), the customer complaint data includes a complaint number, a complaint type, and a complaint name;
in the step (S1), the complaint work order data includes a work order number, a customer number, a complaint type, and a complaint number.
4. The step (S2) of claim 2, wherein the data is preprocessed to remove invalid and abnormal values from the data, and to supplement the default values by averaging the attributes; and (5) adopting normalization processing for data with larger difference.
5. The step (S3) of claim 2, for the pre-processed data, dividing the data by 8:2, 80% being used for training the model, and 20% being used for testing and verifying the model.
6. The step (S4) of claim 2, wherein a power consumer feature extraction model is constructed to extract consumer features based on the consumer history data in preparation for the consumer to recommend services; wherein the step (S4) is divided into the following four parts:
a step (S4-1) of constructing graph data from the user data in the step (S1), correlating the interactions between the users, and thereby constructing the entire relational network graph data; graph model data is shown in FIG. 2;
step (S4-2) of embedding the constructed graph data, namely converting the data to enable the data to meet the requirement of a model and facilitate the extraction of user features, wherein p'nThe data after embedding;
p′n＝Embed(un) (2)
step (S4-3) of constructing a user characteristic extraction model, constructing a GAT model, and extracting the selected node characteristics according to the interaction of each node;
Figure FDA0002670686940000021
Figure FDA0002670686940000022
Figure FDA0002670686940000023
wherein
Figure FDA0002670686940000024
Connecting data in a friend set of a current user with the user data, inputting the data into a two-layer neural network, then performing Softmax processing to obtain interactive weight between the users, and finally obtaining user characteristics generated by interaction between the users by combining with the Attention weight as shown in a formula 4;
step (S4-4) is to perform a user feature extraction operation.
7. The step (S5) of claim 2, wherein a service feature extraction model is constructed to extract service features based on the power service data and the user complaint data, so as to perform matching according to the user features for service recommendation; wherein the step (S5) is divided into the following four parts:
a step (S5-1) of constructing service graph data from the service data in the step (S1), associating services used by the user, thereby constructing the entire service usage graph data; graph model data is shown in FIG. 3;
step (S5-2) is to carry out embedding processing on the constructed graph data, namely to convert the service data into the form of the service model input requirement, so as to extract the service characteristic, e'nEmbedding service dataThe latter data;
e′n＝Embed(in) (6)
step (S5-3) of constructing a GAT model according to the constructed service feature extraction model, carrying out updating according to the service used by each node user, and finally extracting service features;
Figure FDA0002670686940000025
Figure FDA0002670686940000031
Figure FDA0002670686940000032
wherein
Figure FDA0002670686940000033
For the user who has used the service, it will
Figure FDA0002670686940000034
And e'iConnected and then input into a two-layer neural network to obtain a characteristic beta 'of each user using the service'ioObtaining an Attenttion value through Softmax processing, and finally obtaining the characteristic e of the service according to different weightsi；
Step (S5-4) performs a service feature extraction operation.
8. The step (S6) of claim 2, wherein a service recommendation model is established to recommend the service according to the extracted user features, so that the recommended service meets the user' S requirements, thereby achieving accurate recommendation.
9. The step (S7) of claim 2, wherein the model created is examined using the test data to evaluate the validity of the model.
10. The system architecture of the GAT-based power user service recommendation method of claim 1, comprising:
the system mainly comprises four layers, namely a data layer, a middleware layer, an algorithm layer and a service layer from bottom to top;
the data layer is mainly used for acquiring relevant data of power grid users, and the data is mainly acquired from a power customer marketing system, a 95598 work order system, a power utilization acquisition system and a power distribution management system at the present stage;
the middleware layer is mainly used for caching data, and because the data volume generated by the power grid system every day is huge, the middleware is adopted for caching the data, so that the operation burden of the server is reduced;
the algorithm layer is mainly used for processing data and training models, and finally, service recommendation is carried out on the trained models to form a recommendation sequence;
the service layer is mainly used for managing the whole system, and the functions of the service layer comprise modules of user management, recommendation model management, system test management, user behavior analysis, data management, authority management and the like.
CN202010932425.1A 2020-09-08 2020-09-08 Power user service recommendation method based on GAT Pending CN112199604A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010932425.1A CN112199604A (en) 2020-09-08 2020-09-08 Power user service recommendation method based on GAT

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010932425.1A CN112199604A (en) 2020-09-08 2020-09-08 Power user service recommendation method based on GAT

Publications (1)

Publication Number Publication Date
CN112199604A true CN112199604A (en) 2021-01-08

Family

ID=74005366

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010932425.1A Pending CN112199604A (en) 2020-09-08 2020-09-08 Power user service recommendation method based on GAT

Country Status (1)

Country Link
CN (1) CN112199604A (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108764663A (en) * 2018-05-15 2018-11-06 广东电网有限责任公司信息中心 A kind of power customer portrait generates the method and system of management
US20200104729A1 (en) * 2018-09-27 2020-04-02 Babylon Partners Limited Method and system for extracting information from graphs
CN111046297A (en) * 2020-03-12 2020-04-21 深圳市成功快车科技有限公司 Service intelligent matching recommendation method, device, equipment and storage medium based on machine learning algorithm

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108764663A (en) * 2018-05-15 2018-11-06 广东电网有限责任公司信息中心 A kind of power customer portrait generates the method and system of management
US20200104729A1 (en) * 2018-09-27 2020-04-02 Babylon Partners Limited Method and system for extracting information from graphs
CN111046297A (en) * 2020-03-12 2020-04-21 深圳市成功快车科技有限公司 Service intelligent matching recommendation method, device, equipment and storage medium based on machine learning algorithm

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
吴国栋等: "图神经网络推荐研究进展", 《智能系统学报》 *

Similar Documents

Publication Publication Date Title
CN111444236B (en) Mobile terminal user portrait construction method and system based on big data
CN109447364B (en) Label-based electric power customer complaint prediction method
CN109753602B (en) Cross-social network user identity recognition method and system based on machine learning
CN112905891B (en) Scientific research knowledge map talent recommendation method and device based on graph neural network
CN103761254B (en) Method for matching and recommending service themes in various fields
CN112633962B (en) Service recommendation method and device, computer equipment and storage medium
CN109255369A (en) Using the method and device of neural network recognization picture, medium and calculate equipment
CN110598070B (en) Application type identification method and device, server and storage medium
CN107193974A (en) Localized information based on artificial intelligence determines method and apparatus
CN107368521B (en) Knowledge recommendation method and system based on big data and deep learning
CN110955828B (en) Multi-factor embedded personalized package recommendation method based on deep neural network
CN111159387A (en) Recommendation method based on multi-dimensional alarm information text similarity analysis
CN107402859A (en) Software function verification system and verification method thereof
CN104820843A (en) Method for marking picture semantics based on Gauss mixture model
CN110766438A (en) Method for analyzing user behaviors of power grid users through artificial intelligence
CN113095948A (en) Multi-source heterogeneous network user alignment method based on graph neural network
CN107291775A (en) The reparation language material generation method and device of error sample
Ma et al. UniTranSeR: A unified transformer semantic representation framework for multimodal task-oriented dialog system
CN110503507A (en) Insurance products data push method, system and computer equipment based on big data
CN111193254B (en) Residential daily electricity load prediction method and device
CN112199604A (en) Power user service recommendation method based on GAT
CN116452241B (en) User loss probability calculation method based on multi-mode fusion neural network
CN107016561B (en) Information processing method and device
CN112241417B (en) Page data verification method and device, medium and electronic equipment
CN109949175B (en) User attribute inference method based on collaborative filtering and similarity measurement

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210108