Intelligent recommendation method and system based on graph attention network

Abstract

The invention relates to the technical field of intelligent recommendation, and discloses an intelligent recommendation method and system based on a graph attention network. The invention discloses an intelligent recommendation method based on a graph attention network, and the method comprises the following steps: S1, collecting original data which comprises scene data of an advertisement player, advertisement content data and man-machine interaction data; according to the original data, constructing a heterogeneous graph by taking each advertisement player and each advertisement content as nodes and advertisement clicks as edges; s2, modeling the heterogeneous graph structure based on a graph attention network, calculating attention coefficients between nodes, aggregating features of neighbor nodes according to the attention coefficients, and updating a current node; by modeling the relationship between the advertising machine and the advertisement content, accurate recommendation can be realized, the pertinence of advertisement putting is remarkably improved, and the system has dynamic adaptive capacity and can update data and adjust the recommendation result in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recommendation, and particularly relates to an intelligent recommendation method and system based on a graph attention network. Background Art

[0002] An advertising machine is a digital information display device, usually composed of a display screen, a media player, and a content management system, and is used to play advertisements, promotional content, or public information. It is widely used in public places such as shopping malls, airports, and subway stations, supports various media forms such as pictures, videos, and texts, and can be remotely managed and content-updated through the network. The advertising machine is equipped with a touch screen, and the audience can interact with the content by clicking on the advertisement window. The advertising machine not only improves the interactivity of advertisement placement but also provides real-time information for the audience, and is an important tool for modern commercial promotion and information dissemination.

[0003] In public places such as shopping malls, advertising machines are important tools for merchants to conduct publicity and promotion. Traditional advertisement placement methods usually rely on fixed rules or simple statistical analysis, and cannot fully consider multi-dimensional information such as the location of the advertising machine, advertisement content, and floor population characteristics to dynamically adjust the advertisement content recommendation strategy, resulting in poor advertisement placement effects. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent recommendation method and system based on a graph attention network to solve the above technical problems:

[0005] The purpose of the present invention can be achieved through the following technical solutions:

[0006] An intelligent recommendation method based on a graph attention network, the method includes the following steps:

[0007] S1. Collect raw data, where the raw data includes: the scene data of the advertising machine, the advertisement content data, and the human-machine interaction data; according to the raw data, taking each advertising machine and each advertisement content as nodes and advertisement clicks as edges, construct a heterogeneous graph;

[0008] S2. Model the heterogeneous graph structure based on the graph attention network, calculate the attention coefficients between nodes, and aggregate the features of neighbor nodes according to the attention coefficients to update the current node;

[0009] S3. According to the node features of the updated advertising machine nodes and advertisement content nodes, calculate the click-through rate prediction value of the advertisement content on the advertising machine, then sort the advertisement content in the corresponding advertising machine according to the prediction value, and generate a corresponding recommendation list.

[0010] As a further technical solution, the method further includes:

[0011] S4. The advertising machine displays the corresponding advertisement content according to the sorting in the recommended list, and during the process of displaying the advertisement content, collects the actual click-through rate values of each advertisement content in the recommended list;

[0012] S5. Combine the predicted click-through rate value and the actual click-through rate value of the advertisement content to evaluate the recommendation effect.

[0013] As a further technical solution, the scenario data of the advertising machine includes: the location information data of the advertising machine, the population information data within the range of the advertising machine's location, and the merchant information data within the range of the advertising machine's location; the advertisement content data includes: the type, theme, and target audience of the advertisement.

[0014] As a further technical solution, the process of constructing the heterogeneous graph includes:

[0015] Extract features from the scenario data of the advertising machine, including: the floor number of the advertising machine's location, the number of people flow within the range of the advertising machine's location, and the types of merchants, and construct the initial feature vector of the advertising machine node,

[0016] Extract features from the advertisement content data, including: advertisement type, advertisement theme, and advertisement audience, and construct the initial feature vector of the advertisement content node.

[0017] As a further technical solution, the process of calculating the attention coefficient between nodes and aggregating the features of neighbor nodes according to the attention coefficient includes:

[0018] Calculate and obtain the attention coefficient α of the advertising machine node to the advertisement content node ij , the attention coefficient α of the advertisement content node to the advertising machine node ji ;

[0019] The updated feature vector of the advertising machine node is:

[0020] The updated feature vector of the advertisement content node is:

[0021] Among them, are the initial feature vectors of the advertising machine node and the advertisement content node respectively; N(i) is the set of neighbor nodes of the advertising machine node, N(j) is the set of neighbor nodes of the advertisement content node; W is the shared weight matrix; σ() is the non-linear activation function.

[0022] As a further technical solution, the process of calculating the predicted click-through rate value of the advertisement content on the advertising machine at the current location includes:

[0023] Calculate the predicted click-through rate value CTR of the advertisement content on the advertising machine at the current location through the following formula pred ;

[0024]

[0025] Among them, ReLU() is the activation function; W1 and W2 are preset weight matrices; b1 and b2 are preset bias terms; S() is the Sigmoid function.

[0026] As a further technical solution, the process of evaluating the recommendation effect by combining the predicted click-through rate value and the actual click-through rate value of the advertisement content includes:

[0027] Through the formula Calculate the error parameter M;

[0028] Among them, X is the number of advertisement contents; is the predicted click-through rate value of the k-th advertisement content; is the actual click-through rate value of the k-th advertisement content;

[0029] If the error parameter M does not exceed the preset threshold M max , it is determined that the recommendation effect meets the preset standard.

[0030] An intelligent recommendation system based on a graph attention network, the intelligent recommendation system includes:

[0031] A data acquisition module for acquiring the contextual data, advertisement content data, and human-computer interaction data of the advertising machine;

[0032] A graph construction module that takes the advertising machine and the advertisement content as nodes and the advertisement click as an edge to construct a heterogeneous graph and assigns attribute features to each node;

[0033] A graph attention network module for modeling the graph structure through GAT to generate node embedding representations;

[0034] A recommendation module for inputting the embedding representations of the advertising machine node and the advertisement content node into the MLP, outputting the predicted click-through rate value of the advertisement content, sorting the advertisement content in the corresponding advertising machine according to the predicted value, and generating a corresponding recommendation list;

[0035] An advertising machine display terminal for displaying the advertisement content according to the sorting in the recommendation list.

[0036] The beneficial effects of the present invention:

[0037] By modeling the relationship between the advertising machine and the advertising content, the present invention can achieve precise recommendation, significantly improve the pertinence of advertising placement. The system has dynamic adaptability, can update data in real time and adjust the recommendation results, and flexibly respond to changes in the mall environment and population characteristics. At the same time, the efficiency of the graph attention network enables it to quickly process large-scale graph-structured data and is applicable to complex mall scenarios. In addition, through the interpretable attention mechanism, the system can clearly display the relationship between the advertising machine and the advertising content, enhancing the credibility and transparency of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The present invention will be further described below with reference to the accompanying drawings.

[0039] Figure 1 is a flowchart of an intelligent recommendation method based on a graph attention network in the present invention;

[0040] Figure 2 is a block diagram of the content summary of an intelligent recommendation system based on a graph attention network in the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Please refer to Figure 1 As shown, an intelligent recommendation method based on a graph attention network, the method includes the following steps:

[0043] S1. Collect raw data, where the raw data includes: scene data of the advertising machine, advertising content data, and human-computer interaction data; according to the raw data, taking each advertising machine and each advertising content as nodes and advertising clicks as edges, construct a heterogeneous graph. The scene data of the advertising machine is specifically such as location, time, environment, etc., the advertising content data is specifically such as advertising type, creativity, brand, etc., and the human-computer interaction data is such as user clicks, stay time, etc. The node types of the heterogeneous graph include advertising machine nodes and advertising content nodes, and the edges represent the click behavior of users on the advertising content.

[0044] S2. Model the heterogeneous graph structure based on the graph attention network, calculate the attention coefficients between nodes, and aggregate the features of neighbor nodes according to the attention coefficients to update the current node. GAT determines the importance of each node to its neighbor nodes by calculating the attention coefficients. According to the attention coefficients, aggregate the features of neighbor nodes and update the feature representation of the current node. In this way, the features of each node contain not only its own information but also the information of its neighbor nodes.

[0045] S3. According to the node features of the updated advertising machine nodes and advertising content nodes, calculate the predicted click-through rate value of the advertising content on the advertising machine, which can be specifically implemented through a prediction model (such as a neural network). Then, sort the advertising content in the corresponding advertising machine according to the predicted value and generate a corresponding recommendation list. The advertising content in the recommendation list is arranged in descending order of the predicted click-through rate.

[0046] S4. The advertising machine displays the corresponding advertising content according to the sorting in the recommendation list, and during the process of displaying the advertising content, collect the actual click-through rate values of each advertising content in the recommendation list;

[0047] S5. Combine the predicted click-through rate value and the actual click-through rate value of the advertising content to evaluate the recommendation effect.

[0048] Through the above technical solutions, in this embodiment, by modeling the relationship between the advertising machine and the advertising content, accurate recommendation can be achieved, significantly improving the pertinence of advertising placement. The system has dynamic adaptability, can update data in real time and adjust the recommendation results, and flexibly respond to changes in the mall environment and population characteristics. At the same time, the high efficiency of the graph attention network enables it to quickly process large-scale graph structure data and is applicable to complex mall scenarios. In addition, through the interpretable attention mechanism, the system can clearly display the relationship between the advertising machine and the advertising content, enhancing the credibility and transparency of the recommendation results.

[0049] The scenario data of the advertising machine includes: the location information data of the advertising machine, the crowd information data within the location range of the advertising machine, and the merchant information data within the location range of the advertising machine; the advertising content data includes: the type, theme and target audience of the advertisement. The location information data of the advertising machine, for example, the location type: the advertising machine is located at different locations such as the entrance of the shopping mall, next to the elevator, the cashier counter, and the parking lot; location characteristics: such as the size of the flow of people, the length of stay, the lighting conditions, etc.; location relevance: the relevance between the advertising machine and the surrounding environment, for example, the advertising machine near the dining area may be more suitable for recommending food advertisements. The crowd information data within the location range of the advertising machine, for example, crowd portrait: including age, gender, occupation, consumption ability, etc.; crowd density: changes in crowd density in different time periods, such as the difference between morning and evening peak hours. Merchant information data within the location range of the advertising machine, for example, merchant type: such as catering, clothing, electronic products, etc.; merchant relevance: whether the type and activities of merchants around the advertising machine are related to the advertising content. The advertising content data is the relevant information of the advertisement itself, which is used to match the usage scenario of the advertising machine with user needs. Ad type, such as brand promotion, promotional activities, new product launch, etc. Ad theme, such as food, clothing, electronic products, travel, etc. Ad audience, such as age, gender, interests, hobbies, spending power, etc.

[0050] Through the above technical solution, this embodiment can realize intelligent and accurate advertising recommendation by deeply mining the scenario data and advertising content data of the advertising machine and combining the graph attention network technology. This recommendation method can not only improve the advertising effect, but also optimize the user experience.

[0051] The process of building a heterogeneous graph includes:

[0052] Extract features from the scenario data of the advertising machine, including: the number of floors where the advertising machine is located, the flow of people and the types of businesses within the location of the advertising machine, and construct the initialization feature vector of the advertising machine node. Specifically, extract the numerical features of the floor where the advertising machine is located, for example, 1st floor, 2nd floor, 3rd floor, etc., which can be directly represented by numerical values. Count the flow of people around the advertising machine, such as the average flow of people per hour or the flow of people during peak hours, as numerical features. Extract the type information of businesses around the advertising machine, such as catering, clothing, electronic products, etc., and use unique hot encoding or multi-label encoding to represent the types of businesses.

[0053] Extract features from the advertising content data, including: advertising type, advertising theme and advertising audience, and construct the initialization feature vector of the advertising content node. Advertising type, advertising theme and advertising audience can all be represented using one-hot encoding.

[0054] Through the above technical solution, the process of constructing the heterogeneous graph in this embodiment includes extracting features such as the number of floors, the number of people flow, and the types of merchants from the scenario data of the advertising machine to construct the initial feature vector of the advertising machine node; extracting features such as the advertisement type, theme, and audience from the advertisement content data to construct the initial feature vector of the advertisement content node; finally, using the advertising machine and the advertisement content as nodes and the user click behavior as edges to construct the heterogeneous graph. Through this process of feature extraction and vector representation, complex scenario and content data are transformed into structured graph data, providing a basis for subsequent graph neural network modeling and advertisement recommendation.

[0055] The process of calculating the attention coefficients between nodes and aggregating the features of neighbor nodes according to the attention coefficients includes:

[0056] Calculating and obtaining the attention coefficient α of the advertising machine node to the advertisement content node ij , and the attention coefficient α of the advertisement content node to the advertising machine node ji ;

[0057] For example, the specific calculation process of calculating and obtaining the attention coefficient α of the advertising machine node to the advertisement content node ij is as follows:

[0058]

[0059] α ij = softmax(e ij );

[0060] LeakyReLU() is a non-linear activation function; is a learnable attention vector used to calculate the correlation between nodes, and W is a shared weight matrix. Similarly, α ji can be calculated and obtained.

[0061] The updated feature vector of the advertising machine node is:

[0062] The updated feature vector of the advertisement content node is:

[0063] Among them, are the initial feature vectors of the advertising machine node and the advertisement content node respectively; N(i) is the set of neighbor nodes of the advertising machine node, and N(j) is the set of neighbor nodes of the advertisement content node; W is a shared weight matrix; σ() is a non-linear activation function.

[0064] Through the above technical solution, this embodiment provides the specific process of calculating the attention coefficients between nodes and aggregating the features of neighbor nodes according to the attention coefficients

[0065] The process of calculating the predicted click-through rate of the advertisement content on the current advertising machine includes:

[0066] Calculate the predicted click-through rate CTR of the advertisement content on the current advertising machine through the following formula pred ;

[0067] is the joint feature representation of the advertising machine node and the advertisement content node. As shown in the formula, it is obtained by fusing the advertising machine node features and the advertisement content node features.

[0068]

[0069] Among them, ReLU() is the activation function; W1, W2 are preset weight matrices; b1, b2 are preset bias terms; S() is the Sigmoid function, which maps the output to the range of 0 to 1 to represent the click probability. Through the above technical solution, this embodiment provides the process of calculating the predicted click-through rate of the advertisement content on the current advertising machine.

[0070] Combining the predicted click-through rate and the actual click-through rate of the advertisement content, the process of evaluating the recommendation effect includes:

[0071] Calculate the error parameter M through the formula ;

[0072] Among them, X is the number of advertisement contents; is the predicted click-through rate of the kth advertisement content; is the actual click-through rate of the kth advertisement content;

[0073] If the error parameter M does not exceed the preset threshold M max , it is determined that the recommendation effect meets the preset standard, otherwise it indicates that the prediction performance of the model does not meet the expectation and the model needs to be further analyzed and optimized.

[0074] Through the above technical solution, this embodiment provides the process of combining the predicted click-through rate and the actual click-through rate of the advertisement content to evaluate the recommendation effect.

[0075] Please refer to Figure 2 shown, an intelligent recommendation system based on a graph attention network, the intelligent recommendation system includes:

[0076] A data acquisition module for collecting the scenario data, advertisement content data and human-computer interaction data of the advertising machine;

[0077] A graph construction module that constructs a heterogeneous graph with the advertising machine and the advertisement content as nodes and the advertisement click as an edge, and assigns attribute features to each node;

[0078] The graph attention network module is used to model the graph structure through GAT to generate node embedding representations;

[0079] The recommendation module is used to input the embedding representations of the advertising machine nodes and advertising content nodes into an MLP (multi-layer perceptron), output the predicted values of the advertising content click-through rate, sort the advertising content in the corresponding advertising machine according to the predicted values, and generate a corresponding recommendation list;

[0080] The advertising machine display terminal is used to display the advertising content according to the sorting in the recommendation list.

[0081] Through the above technical solutions, in this embodiment, the data collection module collects advertising machine, advertising content, and human-computer interaction data, the graph construction module constructs a heterogeneous graph and assigns node attribute features, the graph attention network module generates node embedding representations, then combines with the MLP of the recommendation module to predict the click-through rate and generate a recommendation list, and finally the advertising machine display terminal dynamically displays the advertising content. This system can accurately model the complex relationship between the advertising machine and the advertising content, realize dynamic adaptation, efficient processing, and interpretable advertising recommendations, and significantly improve the advertising delivery effect and user experience.

[0082] The present invention can be a system, a method, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0083] The computer-readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. The computer-readable storage medium can be, for example, - but not limited to - an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the above. More specific examples (non-exhaustive list) of the computer-readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disc (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punched card or raised structures in grooves having instructions stored thereon, and any suitable combination of the above. The computer-readable storage medium used herein is not construed as an instantaneous signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through an optical fiber cable) or electrical signals transmitted through wires.

[0084] The computer-readable program instructions described herein can be downloaded to various computing / processing devices from a computer-readable storage medium or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in each computing / processing device.

[0085] Although the present invention has been disclosed as above, the scope of protection of the present invention is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will all fall within the scope of protection of the present invention.

Claims

1. An intelligent recommendation method based on a graph attention network, characterized in that, The method includes the following steps: S1. Collect original data, where the original data includes: the scenario data of the advertising machine, the advertising content data, and the human-computer interaction data; according to the original data, construct a heterogeneous graph with each advertising machine and each advertising content as nodes and advertising clicks as edges. S2. Model the heterogeneous graph structure based on the graph attention network, calculate the attention coefficients between nodes, and aggregate the features of neighbor nodes according to the attention coefficients to update the current node. S3. According to the node features of the updated advertising machine nodes and advertising content nodes, calculate the predicted click-through rate value of the advertising content on the advertising machine, then sort the advertising content in the corresponding advertising machine according to the predicted value, and generate a corresponding recommendation list.

2. The intelligent recommendation method based on graph attention network according to claim 1, characterized in that, The method further includes: S4. The advertising machine displays the corresponding advertising content according to the sorting in the recommendation list, and during the process of displaying the advertising content, collect the actual click-through rate values of each advertising content in the recommendation list. S5. Combine the predicted click-through rate value and the actual click-through rate value of the advertising content to evaluate the recommendation effect.

3. An intelligent recommendation method based on a graph attention network according to claim 2, characterized in that, The scenario data of the advertising machine includes: the location information data of the advertising machine, the population information data within the range of the advertising machine's location, and the merchant information data within the range of the advertising machine's location; the advertising content data includes: the type, theme, and target audience of the advertisement.

4. An intelligent recommendation method based on a graph attention network according to claim 3, characterized in that, The process of constructing the heterogeneous graph includes: Extract features from the scenario data of the advertising machine, including: the floor number of the advertising machine's location, the number of people flow within the range of the advertising machine's location, and the types of merchants, and construct the initial feature vector of the advertising machine node. Extract features from the advertising content data, including: the advertisement type, advertisement theme, and advertisement audience, and construct the initial feature vector of the advertising content node.

5. An intelligent recommendation method based on a graph attention network according to claim 1, characterized in that, The process of calculating the attention coefficients between nodes and aggregating the features of neighbor nodes according to the attention coefficients includes: Calculate and obtain the attention coefficient α of the advertising machine node to the advertising content node ij , the attention coefficient α of the advertising content node to the advertising machine node ji ; The feature vector after the update of the advertising machine node is as follows: The feature vector after the update of the advertisement content node is as follows: wherein, are respectively the initial feature vectors of the advertising machine node and the advertising content node; N(i) is the set of neighbor nodes of the advertising machine node, N(j) is the set of neighbor nodes of the advertising content node; W is the shared weight matrix; σ() is the non-linear activation function.

6. An intelligent recommendation method based on a graph attention network according to claim 1, characterized in that The process of calculating the predicted click-through rate value of the advertising content on the advertising machine at the current location includes: The click-through rate prediction value CTR of the advertisement content on the current advertising machine is calculated by the following formula pred ; Where ReLU() is the activation function; W1, W2 are preset weight matrices; b1, b2 are preset bias terms; S() is the Sigmoid function.

7. An intelligent recommendation method based on a graph attention network according to claim 1, characterized in that, The process of combining the predicted click-through rate value and the actual click-through rate value of the advertising content to evaluate the recommendation effect includes: Calculate the error parameter M through the formula ​ Where X is the number of advertisement contents; is the predicted click-through rate value of the k-th advertisement content; is the actual click-through rate value of the k-th advertisement content; If the error parameter M does not exceed the preset threshold M max , it is determined that the recommendation effect meets the preset standard.

8. An intelligent recommendation system based on a graph attention network, characterized in that, The intelligent recommendation system is used to execute the intelligent recommendation method based on the graph attention network according to any one of claims 1-7. The intelligent recommendation system includes: A data collection module, which is used to collect the scenario data of the advertising machine, the advertising content data, and the human-computer interaction data. A graph construction module, which constructs a heterogeneous graph with the advertising machine and the advertising content as nodes and advertising clicks as edges, and assigns attribute features to each node. A graph attention network module, which is used to model the graph structure through GAT to generate node embedding representations. A recommendation module, which is used to input the embedding representations of the advertising machine nodes and the advertising content nodes into the MLP, output the predicted click-through rate value of the advertising content, then sort the advertising content in the corresponding advertising machine according to the predicted value, and generate a corresponding recommendation list. An advertising machine display terminal, which is used to display the advertising content according to the sorting in the recommendation list.

Images