Recommendation method and device for mobile application store of electric power system

Abstract

The invention relates to an electric power system mobile application store recommendation method and device, and belongs to the technical field of computers, and the method comprises the following steps: obtaining user behavior data, calculating an application score, and constructing an application interaction matrix; acquiring application metadata, and taking the application metadata as input of a double-layer automatic encoder to obtain an application feature vector; and calculating a contribution degree based on the application interaction matrix and the application feature vector so as to obtain a user preference vector. Context data are obtained, and context feature vectors are also output through the double-layer automatic encoder. And obtaining a user application similarity matrix according to the user preference vector and the application feature vector. And then the user uses the similarity matrix and the context feature vector as the input of a multilayer perceptron (MLP), and outputs a recommendation score matrix. And finally recommending applications to each user according to the recommendation score matrix. Through a context sensing mechanism, the system can adjust the recommendation strategy in real time, personalized recommendation of thousands of people and thousands of faces is achieved, and the efficiency of obtaining the target application by the user is improved.

Description

Technical Field

[0001] This invention relates to a recommendation method and device for a mobile application store for power systems, belonging to the field of computer technology. Background Technology

[0002] With the continuous improvement of the intelligence and informatization level of power systems, mobile application stores for power systems have gradually become an important tool for the daily work of power industry personnel. These stores provide diverse application services for different roles, including power operation and maintenance personnel, dispatchers, and managers, covering areas such as grid monitoring, fault early warning, power consumption statistics, and equipment management. The high dependence of this work requires application stores not only to provide a vast amount of application resources but also to intelligently recommend relevant applications based on the specific needs of different users, thereby improving work efficiency.

[0003] However, unlike internet app stores, mobile app stores for the power system have their own unique characteristics and needs. On the one hand, applications in the power industry are typically highly specialized, and user preferences for specific functions are particularly evident in functional modules and business processes. On the other hand, the operation of the power system heavily relies on real-time data and dynamic context, such as time, location, electricity price, and load. This requires recommendation systems to not only consider users' static preferences but also dynamically capture the interaction characteristics between users and the context.

[0004] Existing technologies often rely on only a single type of data (such as a user's download history or rating information), while ignoring the rich and diverse data in the power system, such as contextual data (time, electricity price, location, etc.) and detailed metadata of the application (text description, category, tags, etc.). This simplistic data processing approach results in recommendations that lack specificity and comprehensiveness. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a recommendation method and device for a mobile application store for power systems.

[0006] The technical solution of the present invention is as follows:

[0007] On one hand, the present invention provides a recommendation method for a mobile application store for power systems, comprising the following steps:

[0008] Calculate an application score by acquiring user behavior data, and construct an application interaction matrix based on the application score;

[0009] Obtain the application's metadata and use it as input to a two-layer autoencoder to output an application feature vector;

[0010] The contribution is calculated based on the application interaction matrix and the application feature vector, and the user preference vector is obtained based on the contribution and the application feature vector.

[0011] Acquire context data and use the context data as input to a two-layer autoencoder to output a context feature vector;

[0012] A user application similarity matrix is ​​obtained based on the user preference vector and application feature vector;

[0013] The user application similarity matrix and context feature vector are used as inputs to the multilayer perceptron (MLP), and the output is a recommendation score matrix.

[0014] Recommended applications to each user based on the recommendation score matrix.

[0015] Preferably, the behavioral data includes the number of clicks, the number of downloads, ratings, and usage duration;

[0016] The application score is calculated based on the behavioral data, expressed by the formula:

[0017] R raw (u,a)=w1·R click (u,a)+w2·R download (u,a)+w3·R rating (u,a)+w4·R duration (u,a);

[0018] In the formula, R raw (u,a) represents the application score of the u-th user for application a, R click (u,a) represents the number of clicks made by the u-th user on application a, R download (u,a) represents the number of times the u-th user downloads application a, R rating (u,a) represents the rating of application a by the u-th user, R duration (u,a) represents the duration of user u's use of application a, w1 represents the click count weight, w2 represents the download count weight, w3 represents the rating weight, and w4 represents the usage duration weight.

[0019] Preferably, the method further includes normalizing each element of the application interaction matrix, expressed by the formula:

[0020]

[0021] In the formula, R(u,a) represents the application score of the u-th user for application a after normalization, and μ u σ represents the average application score for the u-th user across all applications. uLet ε represent the standard deviation of the application scores for each application by the u-th user, and let ε represent a preset constant, which is set to 10. -6 To prevent division by zero, act(u) represents the activity level of the u-th user, pop(a) represents the popularity of application a, γ represents the activity level adjustment coefficient, δ represents the popularity level adjustment coefficient, u′ represents a random user, and a′ represents a random application. The function represents the maximum value, act(u′) represents the activity level of a random user, and pop(a′) represents the popularity of a random application.

[0022] Preferably, the metadata includes text descriptions, categories, and tags;

[0023] The metadata is used as input to the two-layer autoencoder, expressed by the formula:

[0024] h1=tanh(W e1 M(a)+b e1 );

[0025] F a (a) = tanh(W) e2 h1+b e2 );

[0026] In the formula, h1 represents the output of the hidden layer, tanh represents the hyperbolic tangent activation function, and W... e1 Let M(a) represent the hidden layer weight matrix, and M(a) represent the metadata applied to a. e1 F represents the bias vector of the hidden layer. a (a) represents the feature vector of application a, W e2 b represents the output layer weight matrix. e2 This represents the bias vector of the output layer.

[0027] Preferably, the contribution is calculated based on the application interaction matrix and the application feature vector, expressed by the following formula:

[0028]

[0029] In the formula, α(u,a) represents the contribution of application a to the u-th user, and v T W represents the transpose of the weight vector. p W r Let F represent the weight matrix. a (a′) represents the application feature vector of random application a′, and R(u,a′) represents the application score of the u-th user after normalization of random application a′.

[0030] The user preference vector is obtained based on the contribution and the applied feature vector, expressed by the formula:

[0031]

[0032] In the formula, P u (u) represents the user preference vector of the u-th user.

[0033] Preferably, the context data includes time, location, and electricity price;

[0034] The context data is used as input to the two-layer autoencoder, as expressed by the formula:

[0035] h c =tanh(W c1 C+b c1 );

[0036] F c (C)=tanh(W c2 h c +b c2 );

[0037] In the formula, h c W represents the output of the hidden layer. c1 Let C represent the hidden layer weight matrix, and C denote the upper layer weight matrix.

[0038] The following data, b c1 F represents the bias vector of the hidden layer. c (C) represents the application feature vector of the context data C, W c2 b represents the output layer weight matrix. c2 This represents the bias vector of the output layer.

[0039] Preferably, the user-application similarity score is calculated based on the user preference vector and application feature vector, expressed by the formula:

[0040] S(u,a)=P u (u) T MF a (a)+v T tanh(W u P u (u)+W a F a (a));

[0041] In the formula, S(u,a) represents the similarity score between the u-th user and application a, and P u (u) T The transpose of the user preference vector of the u-th user, W u M, W a Represents the weight matrix;

[0042] Construct a user application similarity matrix based on the user application similarity scores.

[0043] Preferably, the user application similarity matrix and context feature vector are used as inputs to the multilayer perceptron (MLP), as expressed by the formula:

[0044] h = tanh(W1[S(u,a);F c (C)]+b1);

[0045]

[0046] In the formula, h represents the output of the hidden layer, W1 represents the weight matrix of the hidden layer, and b1 represents the bias vector of the hidden layer. b1 represents the transpose of the output layer weight matrix, b2 represents the bias vector of the output layer, and G(u,a) is the recommendation score of the u-th user for application a.

[0047] Construct a recommendation score matrix based on the recommendation scores.

[0048] Preferably, user behavior data, application metadata, and contextual data are obtained through the power system mobile application store.

[0049] In another aspect, the present invention also provides an electronic device having a computer program stored thereon, which, when executed by a processor, implements the recommendation method for a power system mobile application store as described in any embodiment of the present invention.

[0050] The present invention has the following beneficial effects:

[0051] 1. This invention significantly improves the accuracy of recommendation systems by integrating user behavior data, application metadata, and contextual data (such as time, location, and electricity price) to construct a multi-dimensional feature model. Compared to traditional recommendation methods that rely solely on ratings or download volume, this solution can deeply mine users' implicit preferences and dynamically adjust them in conjunction with the unique operating environment of the power industry. Through a context-aware mechanism, the system can adjust recommendation strategies in real time, achieving personalized recommendations tailored to each user, effectively reducing information overload and improving the efficiency with which users access their target applications.

[0052] 2. This invention, tailored to the characteristics of the power industry, introduces "application metadata" as one of its core inputs. This metadata encompasses information such as the application's text description, functional category, and professional tags. Through a two-layer autoencoder, deep semantic extraction is performed to understand the application's functional content and business scenarios. For example, for applications like "substation inspection," keywords can be identified and recommended to users with recent substation operation records. This semantic understanding-based recommendation method overcomes the limitations of the traditional collaborative filtering "black box" model, making the recommendation results more aligned with the actual business processes of the power industry and significantly improving application usage and business matching. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the implementation of the method in an embodiment of the present invention. Detailed Implementation

[0054] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0056] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0057] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0058] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0059] Example 1:

[0060] See Figure 1 This invention provides a recommendation method for a mobile application store in a power system, comprising the following steps:

[0061] Calculate an application score by acquiring user behavior data, and construct an application interaction matrix based on the application score;

[0062] Obtain the application's metadata and use it as input to a two-layer autoencoder to output an application feature vector;

[0063] The contribution is calculated based on the application interaction matrix and the application feature vector, and the user preference vector is obtained based on the contribution and the application feature vector.

[0064] Acquire context data and use the context data as input to a two-layer autoencoder to output a context feature vector;

[0065] A user application similarity matrix is ​​obtained based on the user preference vector and application feature vector;

[0066] The user application similarity matrix and context feature vector are used as inputs to the multilayer perceptron (MLP), and the output is a recommendation score matrix.

[0067] Based on the recommendation score matrix, applications are recommended to each user. For example, if 10 applications need to be recommended to the u-th user, the recommendation scores of the u-th user are sorted from highest to lowest, and the applications corresponding to the top 10 recommendation scores are recommended.

[0068] Preferably, the behavioral data includes the number of clicks, the number of downloads, ratings, and usage duration;

[0069] The application score is calculated based on the behavioral data, expressed by the formula:

[0070] R raw (u,a)=w1·R click (u,a)+w2·R download (u,a)+w3·R rating (u,a)+w4·R duration (u,a);

[0071] In the formula, R raw (u,a) represents the application score of the u-th user for application a, R click (u,a) represents the number of clicks made by the u-th user on application a, R download (u,a) represents the number of times the u-th user downloads application a, R rating (u,a) represents the rating of application a by the u-th user, R duration (u,a) represents the duration of user u's use of application a, w1 represents the click count weight, w2 represents the download count weight, w3 represents the rating weight, and w4 represents the usage duration weight.

[0072] Preferably, the method further includes normalizing each element of the application interaction matrix, expressed by the formula:

[0073]

[0074] In the formula, R(u,a) represents the application score of the u-th user for application a after normalization, and μ u σ represents the average application score for the u-th user across all applications. u Let represent the standard deviation of the application scores for each application by the u-th user, ε represent a preset constant, act(u) represent the activity level of the u-th user, pop(a) represent the popularity of application a, γ represent the activity level adjustment coefficient, δ represent the popularity level adjustment coefficient, u′ represent a random user, and a′ represent a random application. The function represents the maximum value, act(u′) represents the activity level of a random user, and pop(a′) represents the popularity of a random application.

[0075] The average application score is expressed by the formula:

[0076]

[0077] In the formula, A represents the number of applications.

[0078] The standard deviation of the application score is expressed by the formula:

[0079]

[0080] The activity level is expressed by the formula:

[0081]

[0082] The popularity is expressed by the formula:

[0083]

[0084] Preferably, the metadata includes text descriptions, categories, and tags;

[0085] The metadata is used as input to the two-layer autoencoder, expressed by the formula:

[0086] h1=tanh(W e1 M(a)+b e1 );

[0087] F a (a) = tanh(W) e2 h1+b e2 )

[0088] In the formula, h1 represents the output of the hidden layer, tanh represents the hyperbolic tangent activation function, and W... e1 Let M(a) represent the hidden layer weight matrix, and M(a) represent the metadata applied to a. e1 F represents the bias vector of the hidden layer. a (a) represents the feature vector of application a, W e2 b represents the output layer weight matrix. e2 This represents the bias vector of the output layer.

[0089] Preferably, the contribution is calculated based on the application interaction matrix and the application feature vector, expressed by the following formula:

[0090]

[0091] In the formula, α(u,a) represents the contribution of application a to the u-th user, and v T W represents the transpose of the weight vector. p W r Let F represent the weight matrix. a (a′) represents the application feature vector of random application a′, and R(u,a′) represents the application score of the u-th user after normalization of random application a′.

[0092] The user preference vector is obtained based on the contribution and the applied feature vector, expressed by the formula:

[0093]

[0094] In the formula, P u (u) represents the user preference vector of the u-th user.

[0095] Preferably, the context data includes time, location, and electricity price;

[0096] The context data is used as input to the two-layer autoencoder, as expressed by the formula:

[0097] h c =tanh(W c1 C+b c1 );

[0098] F c (C)=tanh(W c2 h c +b c2 );

[0099] In the formula, h c W represents the output of the hidden layer. c1 Let C represent the hidden layer weight matrix, and b represent the context data. c1 F represents the bias vector of the hidden layer. c (C) represents the application feature vector of the context data C, W c2 b represents the output layer weight matrix. c2 This represents the bias vector of the output layer.

[0100] Preferably, the user-application similarity score is calculated based on the user preference vector and application feature vector, expressed by the formula:

[0101] S(u,a)=P u (u) T MF a (a)+v T tanh(W u P u (u)+W a Fa (a));

[0102] In the formula, S(u,a) represents the similarity score between the u-th user and application a, and P u (u) T The transpose of the user preference vector of the u-th user, W u M, W a Represents the weight matrix;

[0103] Construct a user application similarity matrix based on the user application similarity scores.

[0104] Preferably, the user application similarity matrix and context feature vector are used as inputs to the multilayer perceptron (MLP), as expressed by the formula:

[0105] h = tanh(W1[S(u,a);F c (C)]+b1);

[0106]

[0107] In the formula, h represents the output of the hidden layer, W1 represents the weight matrix of the hidden layer, and b1 represents the bias vector of the hidden layer. b1 represents the transpose of the output layer weight matrix, b2 represents the bias vector of the output layer, and G(u,a) is the recommendation score of the u-th user for application a.

[0108] Construct a recommendation score matrix based on the recommendation scores.

[0109] Preferably, user behavior data, application metadata, and contextual data are obtained through the power system mobile application store.

[0110] Example 2:

[0111] This embodiment provides an electronic device having a computer program stored thereon. When the computer program is executed by a processor, it implements the recommendation method for a power system mobile application store as described in any embodiment of the present invention.

[0112] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0113] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0114] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0115] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A recommendation method for a mobile application store for power systems, characterized in that, Includes the following steps: Calculate an application score by acquiring user behavior data, and construct an application interaction matrix based on the application score; Obtain the application's metadata and use it as input to a two-layer autoencoder to output an application feature vector; The contribution is calculated based on the application interaction matrix and the application feature vector, and the user preference vector is obtained based on the contribution and the application feature vector. Acquire context data and use the context data as input to a two-layer autoencoder to output a context feature vector; A user application similarity matrix is ​​obtained based on the user preference vector and application feature vector; The user application similarity matrix and context feature vector are used as inputs to the multilayer perceptron, and the output is a recommendation score matrix; Recommended applications to each user based on the recommendation score matrix.

2. The recommendation method for a mobile application store in a power system according to claim 1, characterized in that, The behavioral data includes the number of clicks, the number of downloads, ratings, and usage duration; The application score is calculated based on the behavioral data, expressed by the formula: R raw (u,a)=w1·R click (u,a)+w2·R download (u,a)+w3·R rating (u,a)+w4·R duration (u,a); In the formula, R raw (u,a) represents the application score of the u-th user for application a, R click (u,a) represents the number of clicks made by the u-th user on application a, R download (u,a) represents the number of times the u-th user downloads application a, R rating (u,a) represents the rating of application a by the u-th user, R duration (u,a) represents the duration of user u's use of application a, w1 represents the click count weight, w2 represents the download count weight, w3 represents the rating weight, and w4 represents the usage duration weight.

3. The recommendation method for a power system mobile application store according to claim 2, characterized in that, The method further includes normalizing each element of the application interaction matrix, expressed by the formula: In the formula, R(u,a) represents the application score of the u-th user for application a after normalization, and μ u σ represents the average application score for the u-th user across all applications. u Let represent the standard deviation of the application scores for each application by the u-th user, ε represent a preset constant, act(u) represent the activity level of the u-th user, pop(a) represent the popularity of application a, γ represent the activity level adjustment coefficient, δ represent the popularity level adjustment coefficient, u′ represent a random user, and a′ represent a random application. The function represents the maximum value, act(u′) represents the activity level of a random user, and pop(a′) represents the popularity of a random application.

4. The recommendation method for a power system mobile application store according to claim 3, characterized in that, The metadata includes text descriptions, categories, and tags; The metadata is used as input to the two-layer autoencoder, expressed by the formula: h1 = tanh(W) e1 M(a)+b e1 ); F a (a)=tanh(W e2 h1+b e2 ); In the formula, h1 represents the output of the hidden layer, tanh represents the hyperbolic tangent activation function, and W... e1 Let M(a) represent the hidden layer weight matrix, and M(a) represent the metadata applied to a. e1 F represents the bias vector of the hidden layer. a (a) represents the feature vector of application a, W e2 b represents the output layer weight matrix. e2 This represents the bias vector of the output layer.

5. The recommendation method for a power system mobile application store according to claim 4, characterized in that, The contribution is calculated based on the application interaction matrix and application feature vector, expressed by the following formula: In the formula, α(u,a) represents the contribution of application a to the u-th user, and v T W represents the transpose of the weight vector. p W r Let F represent the weight matrix. a (a′) represents the application feature vector of random application a′, and R(u,a′) represents the application score of the u-th user after normalization of random application a′. The user preference vector is obtained based on the contribution and the applied feature vector, expressed by the formula: In the formula, P u (u) represents the user preference vector of the u-th user.

6. The recommendation method for a power system mobile application store according to claim 5, characterized in that, The context data includes time, location, and electricity price; The context data is used as input to the two-layer autoencoder, as expressed by the formula: h c = tanh(W c1 C+b c1 ); F c (C)=tanh(W c2 h c +b c2 ); In the formula, h c W represents the output of the hidden layer. c1 Let C represent the hidden layer weight matrix, and b represent the context data. c1 F represents the bias vector of the hidden layer. c (C) represents the application feature vector of the context data C, W c2 b represents the output layer weight matrix. c2 This represents the bias vector of the output layer.

7. The recommendation method for a power system mobile application store according to claim 6, characterized in that, The user-application similarity score is calculated based on the user preference vector and application feature vector, expressed by the formula: S(u,a)=P u (u) T MF a (a)+v T tanh(W u P u (u)+W a F a (a)); In the formula, S(u,a) represents the similarity score between the u-th user and application a, and P u (u) T The transpose of the user preference vector of the u-th user, W u M, W a Represents the weight matrix; Construct a user application similarity matrix based on the user application similarity scores.

8. The recommendation method for a power system mobile application store according to claim 7, characterized in that, The user application similarity matrix and context feature vector are used as inputs to the multilayer perceptron (MLP), as expressed by the formula: h=tanh(W1[S(u,a);F c (C)]+b1); In the formula, h represents the output of the hidden layer, W1 represents the weight matrix of the hidden layer, and b1 represents the bias vector of the hidden layer. b1 represents the transpose of the output layer weight matrix, b2 represents the bias vector of the output layer, and G(u,a) is the recommendation score of the u-th user for application a. Construct a recommendation score matrix based on the recommendation scores.

9. The recommendation method for a mobile application store in a power system according to claim 1, characterized in that, User behavior data, application metadata, and contextual data are obtained through the power system mobile application store.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the recommendation method for a power system mobile application store as described in any one of claims 1 to 9.

Images