A Smart Recommendation Method and System Based on Graph Attention Networks

Abstract

This invention relates to the field of intelligent recommendation technology, and discloses an intelligent recommendation method and system based on graph attention networks. The intelligent recommendation method based on graph attention networks includes the following steps: S1, collecting raw data, including: contextualized data of advertising machines, advertising content data, and human-computer interaction data; constructing a heterogeneous graph based on the raw data, using each advertising machine and each piece of advertising content as nodes and advertising clicks as edges; S2, modeling the heterogeneous graph structure based on graph attention networks, calculating the attention coefficients between nodes, and aggregating the features of neighboring nodes based on the attention coefficients to update the current node; this invention, by modeling the relationship between advertising machines and advertising content, can achieve accurate recommendations, significantly improve the targeting of advertising, and the system has dynamic adaptability, capable of updating data and adjusting recommendation results in real time.

Description

Technical Field

[0001] This invention relates to the field of intelligent recommendation technology, specifically to an intelligent recommendation method and system based on graph attention networks. Background Technology

[0002] An advertising display machine is a digital information display device, typically consisting of a display screen, a media player, and a content management system. It is used to play advertisements, promotional content, or public information. Widely used in public places such as shopping malls, airports, and subway stations, it supports various media formats including images, videos, and text, and can be remotely managed and updated via a network. Equipped with a touchscreen, viewers can interact with the content by clicking on the advertising window. Advertising machines not only enhance the interactivity of advertising but also provide viewers with real-time information, making them an important tool for modern commercial promotion and information dissemination.

[0003] In public places such as shopping malls, advertising machines are an important tool for businesses to advertise and promote their products. Traditional advertising methods are usually based on fixed rules or simple statistical analysis, which cannot fully consider multi-dimensional information such as the location of the advertising machine, the content of the advertisement, and the characteristics of the people on each floor, and cannot dynamically adjust the recommendation strategy for the advertisement content, resulting in poor advertising performance. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent recommendation method and system based on graph attention networks, thereby solving the above-mentioned technical problems:

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] An intelligent recommendation method based on graph attention networks, the method comprising the following steps:

[0007] S1. Collect raw data, including: contextual data of advertising machines, advertising content data, and human-computer interaction data; based on the raw data, construct a heterogeneous graph with each advertising machine and each advertising content as nodes and advertising clicks as edges.

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

[0009] S3. Based on the node characteristics of the updated ad machine nodes and ad content nodes, calculate the predicted click-through rate of the ad content on the ad machine, then sort the ad content in the corresponding ad machine according to the predicted value, and generate the corresponding recommendation list.

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

[0011] S4. The advertising machine displays the corresponding advertising content according to the order in the recommendation list, and collects the actual click-through rate of each advertising content in the recommendation list during the display process.

[0012] S5. Evaluate the recommendation effect by combining the predicted and actual click-through rates of the ad content.

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

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

[0015] Features are extracted from the contextualized data of the advertising machines, including: the floor number of the advertising machine location, the pedestrian traffic within the area of ​​the advertising machine location, and the types of merchants, to construct the initial feature vector of the advertising machine node.

[0016] Features are extracted from the ad content data, including ad type, ad theme, and ad audience, to construct the initial feature vector of the ad content node.

[0017] As a further technical solution, the process of calculating the attention coefficient between nodes and aggregating the features of neighboring nodes based on the attention coefficient includes:

[0018] Calculate the attention coefficient α between the ad machine node and the ad content node. ij The attention coefficient α of the ad content node to the ad machine node ji ;

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

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

[0021] in, , respectively, are the initial feature vectors of the advertising machine node and the advertising content node; N(i) is the set of neighboring nodes of the advertising machine node, and N(j) is the set of neighboring nodes of the advertising content node; W is the shared weight matrix; σ() is the nonlinear activation function.

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

[0023] The predicted click-through rate (CTR) of the ad content at the current location is calculated using the following formula. pred ;

[0024]

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

[0026] As a further technical solution, the process of evaluating recommendation effectiveness by combining the predicted and actual click-through rates (CTR) of the ad content includes:

[0027] Through formula Calculate the error parameter M;

[0028] Where X represents the number of advertisements; Let be the predicted click-through rate (CTR) value for the k-th ad content; This represents the actual click-through rate (CTR) of the k-th ad content.

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

[0030] An intelligent recommendation system based on graph attention networks, the intelligent recommendation system comprising:

[0031] The data acquisition module is used to collect contextual data, advertising content data, and human-computer interaction data from the advertising machine.

[0032] The graph construction module uses advertising machines and advertising content as nodes and advertising clicks as edges to construct a heterogeneous graph and assigns attribute features to each node.

[0033] The graph attention network module is used to model the graph structure using GAT and generate node embedding representations.

[0034] The recommendation module is used to input the embedded representations of ad machine nodes and ad content nodes into the MLP, output the predicted click-through rate of ad content, sort the ad content in the corresponding ad machine according to the predicted value, and generate the corresponding recommendation list.

[0035] An advertising display terminal is used to display advertising content according to the order in the recommendation list.

[0036] The beneficial effects of this invention are:

[0037] This invention achieves precise recommendations by modeling the relationship between advertising machines and advertising content, significantly improving the targeting of advertising. The system possesses dynamic adaptability, capable of updating data and adjusting recommendation results in real time, flexibly responding to changes in the shopping mall environment and demographics. Furthermore, the efficiency of graph attention networks enables it to quickly process large-scale graph-structured data, making it suitable for complex shopping mall scenarios. In addition, through an interpretable attention mechanism, the system can clearly demonstrate the relationship between advertising machines and advertising content, enhancing the credibility and transparency of the recommendation results. Attached Figure Description

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

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

[0040] Figure 2 This is a summary block diagram of the intelligent recommendation system based on graph attention network in this invention; Detailed Implementation

[0041] 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.

[0042] Please see Figure 1 As shown, an intelligent recommendation method based on graph attention networks includes the following steps:

[0043] S1. Collect raw data, including: contextual data of the advertising machine, advertising content data, and human-computer interaction data; based on the raw data, construct a heterogeneous graph with each advertising machine and each piece of advertising content as nodes and advertising clicks as edges. Contextual data of the advertising machine includes specific data such as location, time, and environment; advertising content data includes specific data such as advertising type, creative, and brand; and human-computer interaction data includes user clicks and dwell time. The node types of the heterogeneous graph include advertising machine nodes and advertising content nodes, and edges represent user click behavior on advertising content.

[0044] S2. A graph attention network (GAT) is used to model the heterogeneous graph structure, calculate the attention coefficients between nodes, and aggregate the features of neighboring nodes based on these coefficients to update the current node's feature representation. GAT determines the importance of each node to its neighbors by calculating the attention coefficients. Based on these attention coefficients, the features of neighboring nodes are aggregated, and the feature representation of the current node is updated. In this way, each node's features include not only its own information but also information about its neighboring nodes.

[0045] S3. Based on the updated node characteristics of the ad machine nodes and ad content nodes, calculate the predicted click-through rate (CTR) of the ad content on the ad machine. This can be achieved using a prediction model (such as a neural network). Then, based on the predicted CTR, sort the ad content in the corresponding ad machine and generate a corresponding recommendation list. The ad content in the recommendation list is arranged from highest to lowest based on the predicted CTR.

[0046] S4. The advertising machine displays the corresponding advertising content according to the order in the recommendation list, and collects the actual click-through rate of each advertising content in the recommendation list during the display process.

[0047] S5. Evaluate the recommendation effect by combining the predicted and actual click-through rates of the ad content.

[0048] Through the above technical solution, this embodiment achieves accurate recommendations by modeling the relationship between the advertising machine and the advertising content, significantly improving the targeting of advertising. The system has dynamic adaptability, capable of updating data and adjusting recommendation results in real time, flexibly responding to changes in the shopping mall environment and demographic characteristics. Simultaneously, the efficiency of graph attention networks enables it to quickly process large-scale graph structure data, making it suitable for complex shopping mall scenarios. Furthermore, through an interpretable attention mechanism, the system can clearly demonstrate the relationship between the advertising machine and the advertising content, enhancing the credibility and transparency of the recommendation results.

[0049] The contextualized data of the advertising machine includes: location information data of the advertising machine, population information data within the location range of the advertising machine, and merchant information data within the location range of the advertising machine; the advertising content data includes: advertising type, theme, and target audience. Location information data includes, for example, location type: the advertising machine is located in different locations such as mall entrance, next to elevators, cash register, parking lot, etc.; location characteristics: such as pedestrian traffic, dwell time, lighting conditions, etc.; location relevance: the relevance of the advertising machine to its surrounding environment, for example, an advertising machine near a food court may be more suitable for recommending food advertisements. Population information data within the location range of the advertising machine includes, for example, audience profile: including age, gender, occupation, spending power, etc.; population density: changes in population density at different times, such as differences between morning and evening peak hours. Merchant information data within the location range of the advertising machine includes, for example, merchant type: such as catering, clothing, electronics products, etc.; merchant relevance: whether the types and activities of merchants around the advertising machine are related to the advertising content. Advertising content data is information related to the advertisement itself, used to match the advertising machine's usage scenario with user needs. Advertising type, such as brand promotion, sales campaigns, new product launches, etc.; advertising theme, such as food, clothing, electronics, travel, etc.; target audience, such as age, gender, interests, spending power, etc.

[0050] Through the above technical solution, this embodiment achieves intelligent and precise ad recommendation by deeply mining the contextualized data and ad content data of the advertising machine and combining it with graph attention network technology. This recommendation method not only improves advertising effectiveness but also optimizes user experience.

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

[0052] Features are extracted from the contextualized data of the advertising machines, including: the floor number of the advertising machine's location, the pedestrian traffic within the area of ​​the advertising machine's location, and the types of businesses, to construct an initial feature vector for the advertising machine node. Specifically, numerical features of the floor where the advertising machine is located are extracted, such as the 1st, 2nd, and 3rd floors, which can be directly represented by numerical values. Pedestrian traffic data around the advertising machine is statistically analyzed, such as average hourly traffic or peak-hour traffic, and used as numerical features. Information on the types of businesses around the advertising machine is extracted, such as restaurants, clothing stores, and electronics stores, which can be represented using one-hot encoding or multi-label encoding.

[0053] Features are extracted from the ad content data, including ad type, ad theme, and ad audience, to construct the initial feature vector for each ad content node. Ad type, ad theme, and ad audience can all be represented using one-hot encoding.

[0054] Through the above technical solution, the process of constructing a heterogeneous graph in this embodiment includes extracting features such as floor number, foot traffic, and merchant type from the contextualized data of the advertising machines to construct the initial feature vector of the advertising machine node; extracting features such as advertising type, theme, and audience from the advertising content data to construct the initial feature vector of the advertising content node; and finally constructing a heterogeneous graph with advertising machines and advertising content as nodes and user click behavior as edges. This process, through feature extraction and vectorization, transforms complex scene and content data into structured graph data, providing a foundation for subsequent graph neural network modeling and advertising recommendation.

[0055] The process of calculating the attention coefficient between nodes and aggregating the features of neighboring nodes based on the attention coefficient includes:

[0056] Calculate the attention coefficient α between the ad machine node and the ad content node. ij The attention coefficient α of the ad content node to the ad machine node ji ;

[0057] For example, calculating the attention coefficient α between the ad machine node and the ad content node. ij The specific calculation process is as follows:

[0058]

[0059] α ij =softmax(e ij );

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

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

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

[0063] in, , respectively, are the initial feature vectors of the advertising machine node and the advertising content node; N(i) is the set of neighboring nodes of the advertising machine node, and N(j) is the set of neighboring nodes of the advertising content node; W is the shared weight matrix; σ() is the nonlinear activation function.

[0064] Through the above technical solution, this embodiment provides a specific process for calculating the attention coefficient between nodes and aggregating the features of neighboring nodes based on the attention coefficient.

[0065] The process of calculating the predicted click-through rate (CTR) of ad content at the current location for the ad device includes:

[0066] The predicted click-through rate (CTR) of the ad content at the current location is calculated using the following formula. pred ;

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

[0068]

[0069] Where ReLU() is the activation function; W1 and W2 are preset weight matrices; b1 and b2 are preset bias terms; and 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 a process for calculating the predicted click-through rate of advertising content at the current location of the advertising machine.

[0070] The process of evaluating recommendation effectiveness by combining predicted and actual click-through rates (CTR) for ad content includes:

[0071] Through formula Calculate the error parameter M;

[0072] Where X represents the number of advertisements; Let be the predicted click-through rate (CTR) value for the k-th ad content; This represents the actual click-through rate (CTR) of the k-th ad content.

[0073] If the error parameter M does not exceed the preset threshold M max If the recommendation effect meets the preset standard, it is determined that the recommendation effect meets the preset standard; otherwise, it means that the model's predictive performance has not met expectations and further analysis and optimization of the model are needed.

[0074] Through the above technical solution, this embodiment provides a process for evaluating the recommendation effect by combining the predicted click-through rate (CTR) value and the actual CTR value of the advertising content.

[0075] Please see Figure 2 As shown, an intelligent recommendation system based on graph attention networks is disclosed. The intelligent recommendation system includes:

[0076] The data acquisition module is used to collect contextual data, advertising content data, and human-computer interaction data from the advertising machine.

[0077] The graph construction module uses advertising machines and advertising content as nodes and advertising clicks as edges to construct a heterogeneous graph and assigns attribute features to each node.

[0078] The graph attention network module is used to model the graph structure using GAT and generate node embedding representations.

[0079] The recommendation module is used to input the embedded representations of ad machine nodes and ad content nodes into MLP (Multilayer Perceptron), output the predicted value of ad content click-through rate, sort the ad content in the corresponding ad machine according to the predicted value, and generate the corresponding recommendation list.

[0080] An advertising display terminal is used to display advertising content according to the order in the recommendation list.

[0081] Through the above technical solution, this embodiment collects data from advertising machines, advertising content, and human-computer interaction via a data acquisition module. A heterogeneous graph is constructed using a graph construction module, and node attributes are assigned. A graph attention network module generates node embedding representations. Finally, the MLP (Multi-Level Processing) of the recommendation module predicts click-through rates and generates a recommendation list. The advertising content is then dynamically displayed on the advertising machine terminal. This system can accurately model the complex relationship between advertising machines and advertising content, achieving dynamic adaptation, efficient processing, and interpretable advertising recommendations, significantly improving advertising effectiveness and user experience.

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

[0083] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

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

[0085] While the present invention has been disclosed above, its scope of protection 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 all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. An intelligent recommendation method based on graph attention networks, characterized in that, The method includes the following steps: S1. Collect raw data, including: contextual data of advertising machines, advertising content data, and human-computer interaction data; based on the raw 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 graph attention network, calculate the attention coefficient between nodes, aggregate the features of neighboring nodes according to the attention coefficient, and update the current node; S3. Based on the node characteristics of the updated ad machine nodes and ad content nodes, calculate the predicted click-through rate of the ad content on the ad machine, then sort the ad content in the corresponding ad machine according to the predicted value, and generate the corresponding recommendation list. The contextualized data of the advertising machine includes: location information data of the advertising machine, population information data within the location range of the advertising machine, and merchant information data within the location range of the advertising machine; the advertising content data includes: advertising type, theme, and target audience; The process of constructing a heterogeneous graph includes: Features are extracted from the contextualized data of the advertising machines, including: the floor number of the advertising machine location, the pedestrian traffic within the area of ​​the advertising machine location, and the types of merchants, to construct the initial feature vector of the advertising machine node. Features are extracted from the ad content data, including ad type, ad theme, and ad audience, to construct the initial feature vector of the ad content node.

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 order in the recommendation list, and collects the actual click-through rate of each advertising content in the recommendation list during the display process. S5. Evaluate the recommendation effect by combining the predicted and actual click-through rates of the ad content.

3. The intelligent recommendation method based on graph attention network according to claim 2, characterized in that, The process of calculating the attention coefficient between nodes and aggregating the features of neighboring nodes based on the attention coefficient includes: Calculate the attention coefficient between the ad machine node and the ad content node. Attention coefficient of ad content node to ad machine node ; The updated feature vector of the advertising machine node is: ; The updated feature vector of the ad content node is: ; in, , These are the initial feature vectors for the advertising machine node and the advertising content node, respectively; This is the set of neighboring nodes of the advertising machine node. The set of neighboring nodes of the advertising content node; For shared weight matrices; It is a non-linear activation function.

4. The intelligent recommendation method based on graph attention network according to claim 3, characterized in that, The process of calculating the predicted click-through rate (CTR) of ad content at the current location for the ad device includes: The predicted click-through rate (CTR) of the ad content at the current location is calculated using the following formula. ; ； ； in, For activation functions; , This is a preset weight matrix; , This is a preset bias term; This is the Sigmoid function.

5. The intelligent recommendation method based on graph attention network according to claim 4, characterized in that, The process of evaluating recommendation effectiveness by combining predicted and actual click-through rates (CTR) for ad content includes: Through formula Calculate error parameters ; in, The quantity of advertising content; Let be the predicted click-through rate (CTR) value for the k-th ad content; This represents the actual click-through rate (CTR) of the k-th ad content. If error parameter Not exceeding the preset threshold If the recommendation effect meets the preset criteria, then it is determined that the recommendation effect meets the preset criteria.

6. An intelligent recommendation system based on graph attention networks, characterized in that, The intelligent recommendation system is used to execute the intelligent recommendation method based on graph attention network as described in any one of claims 1-5, and the intelligent recommendation system includes: The data acquisition module is used to collect contextual data, advertising content data, and human-computer interaction data from the advertising machine. The graph construction module uses advertising machines and advertising content as nodes and advertising clicks as edges to construct a heterogeneous graph and assigns attribute features to each node. The graph attention network module is used to model the graph structure using GAT and generate node embedding representations. The recommendation module is used to input the embedded representations of ad machine nodes and ad content nodes into the MLP, output the predicted click-through rate of ad content, sort the ad content in the corresponding ad machine according to the predicted value, and generate the corresponding recommendation list. An advertising display terminal is used to display advertising content according to the order in the recommendation list.

Images