Short message prediction distribution method based on graph structure, storage medium and device

Abstract

The application discloses a short message prediction distribution method based on a graph structure, a storage medium and an apparatus, and belongs to the technical field of communication. In the method, a user set associated with a content category of a target short message is determined; the user set is divided into a plurality of user subsets according to a short message contact frequency between any two users in the user set; graph structure data corresponding to each user subset is constructed; graph structure features corresponding to each graph structure data are determined; whether a starting user corresponding to an edge feature of the graph structure features forwards the target short message to an ending user is predicted according to the edge feature; according to a prediction result, a user who does not need to distribute the target short message is deleted from the user set, and the target short message is distributed to a target user. Thus, the application realizes the technical effect of accurately screening the target user, reduces unnecessary push volume to control group sending cost, ensures that information reaches the real user in need to optimize information transmission efficiency.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a short message prediction and distribution method, storage medium and device based on graph structure. Background Technology

[0002] Short messages (SMS), with their advantages of immediacy, high reach, and wide coverage, have been widely used in areas such as commercial marketing, government announcements, public service notifications, and internal enterprise collaboration. They have become a core means of efficient communication and ensuring timely information delivery between information providers (such as enterprises and government agencies) and target users. Currently, SMS distribution mainly adopts either full-scale mass messaging or group-based mass messaging based on simple user attributes such as region and age. Full-scale mass messaging directly pushes the SMS to all preset user groups without determining whether users need to receive it or can obtain the information through other channels. While simple attribute-based group-based mass messaging can initially screen users, it only focuses on individual static attributes and does not consider the dynamic interaction relationships between users and differences in personalized needs. Both models have significant accuracy deficiencies.

[0003] The existing distribution model directly leads to high costs for mass SMS messaging, specifically in three aspects: First, ineffective pushes waste resources, with a large number of SMS messages being pushed to users who have already obtained information through forwarding by friends and family or who have no interest in the content, consuming SMS credits, server processing, and bandwidth resources, thus increasing operating costs; second, duplicate pushes damage user experience, with the same user potentially receiving both forwarded messages and system pushes, causing negative perceptions and even unsubscribing, requiring additional costs to recall users; third, low accuracy leads to a cost-benefit imbalance, with poor matching between content and user needs, resulting in low open rates and conversion rates, requiring information publishers to increase the volume of mass messaging to compensate for the deficiencies, creating a "high investment, low return" cycle, further exacerbating cost pressures.

[0004] To address these issues, some existing technologies attempt to introduce user interest tags to optimize the scope of mass messaging. However, such solutions only focus on individual user interest characteristics and do not analyze the interaction relationships between users, such as the frequency of SMS communication and historical forwarding behavior. This makes it difficult to exclude users who do not require proactive system push notifications, and the core problem of high SMS mass messaging costs remains unresolved.

[0005] There are currently no effective solutions to the technical problems in the existing short message distribution technology, such as ineffective push resource waste, repeated push affecting user experience, high cost of mass sending and imbalance of information transmission efficiency. Summary of the Invention

[0006] The embodiments of this disclosure provide a graph-based short message prediction and distribution method, storage medium, and apparatus to at least solve the technical problems in the prior art, such as wasted resources from ineffective push notifications, repeated pushes affecting user experience, high costs of mass messaging, and imbalanced information transmission efficiency.

[0007] According to one aspect of the present disclosure, a graph-based method for predicting and distributing short messages is provided, comprising: determining a set of users associated with the content category of a target short message; dividing the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set; constructing graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of short message sending by users in the corresponding user subset and the frequency of short message sending between users; inputting the graph structure data into a pre-trained graph neural network model to determine graph structure features corresponding to each graph structure data; predicting, using the pre-trained neural network model, whether a starting user corresponding to a side feature will forward the target short message to an ending user based on the edge features of the graph structure features; deleting users from the user set who do not need to receive the target short message based on the prediction result, and determining the target user of the target short message; and distributing the target short message to the target user.

[0008] According to another aspect of the present disclosure, a storage medium is also provided, the storage medium including a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0009] According to another aspect of the present disclosure, a graph-based short message prediction and distribution apparatus is also provided, comprising: a set determination module for determining a set of users associated with the content category of a target short message; a partitioning module for partitioning the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set; a construction module for constructing graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of short message sending by users in the corresponding user subset and the frequency of short message sending between users; a graph structure feature determination module for inputting the graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data; a prediction module for predicting, based on the edge features of the graph structure features and using the pre-trained neural network model, whether the starting user corresponding to the edge feature will forward the target short message to the ending user; a target user determination module for deleting users from the user set who do not need to receive the target short message based on the prediction results and determining the target user of the target short message; and a distribution module for distributing the target short message to the target user.

[0010] According to another aspect of the present disclosure, a graph-based short message prediction and distribution apparatus is also provided, comprising: a processor; and a memory connected to the processor, configured to provide the processor with instructions for processing the following steps: determining a set of users associated with the content category of a target short message; dividing the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set; constructing graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of short message sending by users in the corresponding user subset and the frequency of short message sending between users; inputting the graph structure data into a pre-trained graph neural network model to determine graph structure features corresponding to each graph structure data; predicting, using the pre-trained neural network model, whether a starting user corresponding to a side feature will forward the target short message to an ending user based on the edge features of the graph structure features; deleting users from the user set who do not need to receive the target short message based on the prediction result, and determining the target user of the target short message; and distributing the target short message to the target user.

[0011] This application first determines the content category of the target SMS message, and then filters out users with relevant interactive behaviors from the database to form a user set. Next, it counts the frequency of SMS messages between users in the set and uses a clustering algorithm to divide the user subsets with close interactions. Subsequently, it constructs a graph structure data for each user subset, with users as nodes, user interaction relationships as directed edges, and corresponding sending frequency attributes. It inputs this data into a pre-trained graph neural network model to extract graph structure features that represent interaction patterns and forwarding potential. Then, it inputs the edge features in the graph structure features into a pre-trained fully connected neural network model to obtain the forwarding probability value from the starting user to the ending user. Based on this, users who can obtain messages through spontaneous forwarding are deleted to determine the target users. Finally, the target SMS message is distributed to the target users. Therefore, this application achieves the technical effects of accurately screening target users to reduce invalid pushes, saving distribution resources, avoiding repeated pushes to users who can obtain messages through spontaneous forwarding to improve user experience, reducing unnecessary push volume to control mass sending costs, and ensuring that information reaches the users who really need it to optimize information transmission efficiency. It also solves the technical problems of invalid push waste, repeated push affecting user experience, high mass sending costs and unbalanced information transmission efficiency in the existing short message distribution technology. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this disclosure and form part of this application, illustrate exemplary embodiments of this disclosure and are used to explain this disclosure, but do not constitute an undue limitation of this disclosure. In the drawings:

[0013] Figure 1 This is a hardware structure block diagram of a computing device for implementing the method described in Embodiment 1 of this disclosure;

[0014] Figure 2 This is a schematic diagram of a graph-based short message prediction and distribution system according to Embodiment 1 of this disclosure;

[0015] Figure 3 This is a flowchart illustrating the short message prediction and distribution method based on graph structure according to Embodiment 1 of this disclosure;

[0016] Figure 4 This is a schematic diagram of user set partitioning according to the graph-based short message prediction and distribution method described in Embodiment 1 of this disclosure;

[0017] Figure 5 This is a schematic diagram illustrating the process of determining the target short message content category according to the graph-based short message prediction and distribution method described in Embodiment 1 of this disclosure;

[0018] Figure 6 This is a schematic diagram of edge feature extraction in the graph-based short message prediction and distribution method according to Embodiment 1 of this disclosure;

[0019] Figure 7 This is a schematic diagram of the graph structure of the short message prediction and distribution method based on graph structure according to Embodiment 1 of this disclosure;

[0020] Figure 8 This is a schematic diagram of graph structure message passing according to the graph structure-based short message prediction and distribution method described in Embodiment 1 of this disclosure;

[0021] Figure 9 This is a schematic diagram of a neural network model for a graph-based short message prediction and distribution method according to Embodiment 1 of this disclosure;

[0022] Figure 10 This is a schematic diagram of a graph-based short message prediction and distribution device according to Embodiment 2 of this disclosure; and

[0023] Figure 11 This is a schematic diagram of a graph-based short message prediction and distribution device according to Embodiment 3 of this disclosure. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Example 1

[0027] According to this embodiment, a method for predicting and distributing short messages based on a graph structure is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0028] The method embodiments provided in this example can be executed on mobile terminals, computer terminals, servers, or similar computing devices. Figure 1 A hardware block diagram of a computing device for implementing a graph-based short message prediction and distribution method is shown. Figure 1 As shown, a computing device may include one or more processors (processors may include, but are not limited to, microprocessors such as MCUs or programmable logic devices such as FPGAs), a memory for storing data, a transmission device for communication functions, and an input / output interface. The memory, transmission device, and input / output interface are connected to the processor via a bus. In addition, it may also include a display, keyboard, and cursor control device connected to the input / output interface. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, a computing device may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0029] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element in a computing device. As involved in the embodiments of this disclosure, the data processing circuits serve as processor control (e.g., selection of a variable resistor termination path connected to an interface).

[0030] The memory can be used to store software programs and modules of application software, such as the program instruction / data storage device corresponding to the graph-based SMS prediction and distribution method in this embodiment of the present disclosure. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the graph-based SMS prediction and distribution method of the aforementioned application. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the computing device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0031] The transmission device is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the computing device's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0032] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows users to interact with the user interface of the computing device.

[0033] It should be noted here that, in some optional embodiments, the above... Figure 1 The computing device shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computing devices.

[0034] Figure 2This is a schematic diagram of a graph-based short message prediction and distribution system according to this embodiment. (Refer to...) Figure 2 As shown, the system includes: a database 100, a server 200, and a user terminal 300. The database 100 stores various types of data that support SMS prediction and distribution, including basic user information and preference vectors, SMS content data, and intermediate results generated by the server 200; it also receives user behavior data fed back by the user terminal 300.

[0035] Server 200 is used to retrieve data from database 100 to determine the target SMS content category, filter related user sets, divide user subsets using clustering algorithms, and construct graph structure data for each subset; call a pre-trained graph neural network model to extract graph structure features, and predict the forwarding probability between users based on edge features; filter target users according to the prediction results and generate SMS distribution instructions; and store key data from the processing back into database 100.

[0036] User terminal 300 is used to receive target short messages sent by server 200 and present them to the user; record the user's behavior and operations on the short messages, and transmit the behavior data back to database 100 in real time.

[0037] Under the aforementioned operating environment, according to the first aspect of this embodiment, a short message prediction and distribution method based on a graph structure is provided. This method comprises... Figure 2 The database 100 and server 200 shown are implemented together. Figure 3 A flowchart illustrating the method is shown below. (Refer to...) Figure 3 As shown, the method includes:

[0038] S302: Determine the set of users associated with the content category of the target SMS message;

[0039] S304: Divide the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set;

[0040] S306: Construct graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of SMS messages sent by users in the corresponding user subset and the frequency of SMS messages sent between users;

[0041] S308: Input the graph structure data into the pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data.

[0042] S310: Based on the edge features of the graph structure, use a pre-trained neural network model to predict whether the starting user corresponding to the edge feature will forward the target short message to the ending user.

[0043] S312: Based on the prediction results, remove users from the user set who do not need to receive the target SMS, and determine the target users of the target SMS; and

[0044] S314: Distribute the target SMS to the target user.

[0045] Specifically, refer to Figure 2 As shown, server 200 first determines the content category of the target SMS message to be distributed, then filters the corresponding users in database 100 based on the content category, ultimately forming a user set (corresponding to step S302). (See reference) Figure 4 As shown, in an embodiment of the present invention, all user groups are divided into multiple user sets (each set is defined by a corresponding dashed circular area) by clustering user information, and each user set corresponds to a preset content category. Each user's information is a preference vector comprising L elements, where each element corresponds to a preset interest element, and each element value represents the user's preference for that preset interest element.

[0046] Figure 5 A flowchart illustrating the process of determining the target SMS content category (see reference). Figure 5 As shown, the target short message text is first segmented into words, resulting in a segmentation sequence w1, w2, w3, ..., w n Then, the segmented sequence is input into a preset segmentation model, and the content category is determined through multi-module processing. Specifically, the segmented sequence w1, w2, w3, ..., w n The input word embedding layer transforms the text into semantically meaningful word vectors. These vectors are then fed into the BERT model to generate contextual vectors that incorporate contextual information. Next, these contextual vectors are input into a fully connected layer, which outputs an integral value corresponding to each preset content category. Finally, the integral value with each content category is input into a softmax classifier, which calculates the probability that the target message belongs to each content category. The category with the highest probability is the final content category of the target message.

[0047] It should be noted that the target SMS message's preset content category can be, for example, commercial marketing (such as clothing promotion notifications) or government services (such as social security payment reminders). The corresponding user set is filtered using the following criteria: user behavior data within a preset time period (such as the past 30 days) is extracted from database 100, and users who have received messages of this content category, clicked on similar content, or actively marked their interests are selected, forming a user set associated with the target SMS message's content category. For example, in a commercial marketing scenario, if the target SMS message is about weekend discounts at restaurants, server 200 will filter users who have clicked on restaurant-related messages within the past 30 days and have offline restaurant consumption records, aggregating them into a user set corresponding to that discount message, ensuring a high correlation between the users in the set and the target SMS message.

[0048] Furthermore, server 200 counts the total number of short message interactions between any two users in the user set within a preset time period (e.g., the last 30 days) as the contact frequency. Using a clustering algorithm (e.g., K-means), users whose short message contact frequency is higher than a preset threshold (e.g., the preset threshold is 10) are grouped into the same user subset, while those below the preset threshold are divided into different user subsets (corresponding to step S304), making the short message interactions between users within each user subset closer.

[0049] Subsequently, server 200 constructs graph structure data corresponding to each user subset (corresponding to step S306). In the graph structure data of each user subset, nodes correspond to users in the user subset, and node attributes include the total frequency of short messages sent by the user to other users in the user subset. Directed edges between nodes represent the short message interaction relationship between users, and the edge attributes record the specific sending frequency between two users (such as the number of times user A sends to user B). Thus, the short message sending frequency of users themselves and between users is intuitively presented through the attributes of nodes and edges.

[0050] Next, refer to Figure 6 As shown, server 200 inputs the graph structure data corresponding to each user subset into the pre-trained graph neural network model one by one, thereby determining the graph structure features corresponding to each graph structure data (corresponding to step S308). Specifically, the graph neural network model performs multi-dimensional feature fusion processing on the input graph structure data: it integrates the basic user interaction information corresponding to the nodes (such as the cumulative number of interactions of users in the user subset), and combines the short message interaction relationship between users represented by the graph edges and edge attributes (such as interaction frequency and interaction direction). Through the built-in mechanism of the model (such as neighborhood aggregation, attention weight allocation, etc.), it extracts and generates high-dimensional structured information that can represent the overall interaction pattern within the user subset and the short message forwarding potential between user pairs (node ​​pairs). This high-dimensional structured information is the graph structure feature.

[0051] Furthermore, the graph structure features include node features and edge features. Node features are used to characterize the interaction characteristics of an individual user (such as the frequency of active sending by a user in a user subset, the range of users covered by the interaction, etc.); edge features are used to quantify the short message forwarding correlation between two users (node ​​pairs) (i.e., the potential probability of both parties engaging in forwarding behavior).

[0052] Next, based on the edge features of the graph structure, a pre-trained neural network model is used to predict whether the starting user corresponding to the edge feature will forward the target SMS to the ending user (corresponding to step 310). The pre-trained neural network model is a fully connected neural network model. Its training process uses historical edge feature data (with the same format and dimension as the current edge features) as input samples and the actual forwarding labels of the user pairs involved in the corresponding historical edge features (binary labels indicating whether the user pair has forwarded SMS in the historical scenario, 1 indicates forwarding and 0 indicates not forwarding) as supervision data. The prediction bias is calculated through the cross-entropy loss function, and the weight parameters of each layer of the model are iteratively adjusted in combination with the Adam optimizer until the forwarding prediction accuracy and AUC value (area under the ROC curve) of the model on the independent test set meet the preset performance thresholds (such as accuracy ≥ 90% and AUC ≥ 0.85), ensuring that the model has the reliable ability to infer forwarding behavior based on edge features.

[0053] Then, each edge feature (corresponding to a pair of starting and ending users) is individually input into the neural network model. This model, through a process of linear mapping in the input layer, nonlinear fitting using the ReLU activation function in the hidden layer, and probability mapping using the Sigmoid function in the output layer, outputs a predicted probability value ranging from [0,1]. This probability value directly quantifies the likelihood that the starting user corresponding to the edge feature will spontaneously forward the target SMS message to the ending user. The closer the probability value is to 1, the higher the probability of this forwarding behavior occurring; conversely, the lower the probability, providing a quantitative basis for subsequent judgments on whether the system needs to actively push the target SMS message.

[0054] Further, based on the probability value of the starting user forwarding the target SMS to the ending user, the server 200 removes users from the user set who do not need to receive the target SMS and determines the target user (corresponding to step S312). In this embodiment of the invention, the server 200 first locates each ending user in the user set, wherein the user pointed to by the directed edge in the graph structure data of the ending user is the starting user of the directed edge (i.e., the user who may forward the target SMS to it). For each ending user, the server 200 checks the forwarding probability value (prediction result) of all its corresponding starting users one by one, and determines whether the ending user can obtain the target SMS content through the spontaneous forwarding of the starting user. When the prediction result indicates that the current ending user does not need the server 200 to actively distribute the target SMS to know the SMS content, the ending user is removed from the user set; when the prediction result indicates that the current ending user needs the server 200 to actively distribute the target SMS to know the SMS content, the ending user is retained in the user set; and the remaining users in the user set are taken as the target users of the target SMS.

[0055] Finally, the target SMS message is distributed to the target users in the user set (corresponding to step S314).

[0056] As described in the background section above, the existing distribution model directly leads to high costs for mass SMS messaging, specifically in three aspects: First, ineffective pushes waste resources, with a large number of SMS messages being pushed to users who have already obtained information through forwarding from friends and family or who have no interest in the content, consuming SMS credits, server processing, and bandwidth resources, thus increasing operating costs; second, duplicate pushes damage user experience, with the same user potentially receiving both forwarded messages and system pushes, causing negative perceptions and even unsubscribing, requiring additional costs to recall users; third, low accuracy leads to a cost-benefit imbalance, with poor matching between content and user needs, resulting in low open rates and conversion rates, requiring information publishers to increase the volume of mass messaging to compensate for the deficiencies, creating a "high investment, low return" cycle, further exacerbating cost pressures.

[0057] In view of this, this application first determines the content category of the target SMS message, and then selects users with relevant interactive behaviors from the database to form a user set. Next, it counts the frequency of SMS messages between users in the set and uses a clustering algorithm to divide the user subsets with close interactions. Subsequently, it constructs a graph structure data for each user subset, with users as nodes, user interaction relationships as directed edges, and corresponding sending frequency attributes. It inputs the graph structure data into a pre-trained graph neural network model to extract graph structure features that represent interaction patterns and forwarding potential. Then, it inputs the edge features in the graph structure features into a pre-trained fully connected neural network model to obtain the forwarding probability value from the starting user to the ending user. Based on this, users who can obtain messages through spontaneous forwarding are deleted to determine the target users. Finally, the target SMS message is distributed to the target users. Therefore, this application achieves the technical effects of accurately screening target users to reduce invalid pushes, saving distribution resources, avoiding repeated pushes to users who can obtain messages through spontaneous forwarding to improve user experience, reducing unnecessary push volume to control mass sending costs, and ensuring that information reaches the users who really need it to optimize information transmission efficiency. It also solves the technical problems of invalid push waste, repeated push affecting user experience, high mass sending costs and unbalanced information transmission efficiency in the existing short message distribution technology.

[0058] Optionally, the operation of constructing graph structure data corresponding to each user subset includes: constructing a directed graph structure corresponding to each user subset, wherein the nodes in the directed graph structure correspond to the users in the corresponding user subset, and the directed edges in the directed graph structure are used to indicate that the frequency of the starting user corresponding to the starting point sending short messages to the ending user corresponding to the ending point meets a preset condition; and constructing graph structure data corresponding to each directed graph structure, wherein the node attributes of the nodes in the graph structure data represent the frequency of the corresponding user sending short messages, and the edge attributes of the directed edges in the graph structure data represent the similarity between the starting user corresponding to the starting point and the ending user corresponding to the ending point and the frequency of the starting user sending short messages to the ending user.

[0059] Specifically, constructing the graph structure data corresponding to each user subset requires two steps and can be combined with... Figure 7 Intuitive understanding of directed graph examples:

[0060] The first step is to construct a directed graph structure. For each subset of users, using a single user within the subset (e.g., ...) Figure 7 In the graph, N1~N5 are used as independent nodes, each bound to a unique user ID to ensure a one-to-one correspondence between user identity and node; then, directed edges are constructed based on preset conditions (e.g., the frequency of the starting user sending SMS messages to the ending user ≥ 3 times in the past 30 days). Figure 7(E1~E8 in the original text). The direction of the directed edges is completely consistent with the direction of the short message sending (for example, E1 is N2 to N1, that is, N2 sends a short message to N1; E2 is N1 to N2, that is, N1 sends a short message to N2). This is used to filter out user pairs with effective interaction in the subset.

[0061] Then, graph structure data is constructed based on the directed graph structure, supplementing the nodes and directed edges with specific attribute information to form complete graph structure data. The node attribute values ​​of the nodes in the graph structure data directly represent the total frequency of short messages sent by the corresponding user in that user subset (e.g., if N2 sends a total of 15 short messages to N1, N3, and N4 in the subset, then the node attribute value of N2 is 15), intuitively reflecting the scale of user active interaction. In addition, the edge attributes of the nodes in the graph structure data represent the similarity between the starting user and the ending user, as well as the specific frequency of short messages sent from the starting user to the ending user (e.g., if N2 sends 8 short messages to N1 corresponding to E1, then the frequency attribute value of the directed edge E1 is 8), thereby quantifying the interaction strength and the closeness of the association between user pairs.

[0062] Through the above two steps, the graph structure data not only fully depicts the interaction relationship network within the user subset, but also quantifies the interaction scale, similarity, and frequency through attribute dimensions, providing structured and accurate data support for subsequent graph neural network models to extract features and predict short message forwarding behavior between users.

[0063] Optionally, the operation of constructing a directed graph structure corresponding to each user subset includes: mapping users of the user subset to nodes of the corresponding directed graph structure; and determining whether to construct a directed edge starting from the first user and ending at the second user based on the frequency of the first user sending short messages to the second user in the user subset.

[0064] Specifically, when constructing the directed graph structure corresponding to each user subset, firstly, each user in the user subset is mapped one-to-one with an independent node in the directed graph structure, such as... Figure 7 Nodes N1 to N5 in the graph are identified by their unique IDs, ensuring that each user has a unique node mapping in the directed graph, achieving precise association between users and graph nodes. Then, for any two users in the user subset (e.g., user N1 and user N2), the frequency of short messages sent from user N1 to user N2 is counted (e.g., the number of messages sent in the last 30 days). If this frequency reaches a preset threshold (e.g., ≥3 times), a directed graph is constructed with user N1's corresponding node as the starting point and user N2's corresponding node as the ending point, such as... Figure 7The directed edges are E1 (N2 points to N1), E2 (N1 points to N2), E3 (N3 points to N2), E4 (N2 points to N3), E5 (N4 points to N2), E6 (N2 points to N4), E7 (N5 points to N4), and E8 (N4 points to N5). If the frequency threshold is not reached, the directed edge is not constructed.

[0065] By using the above methods, we can ensure that the edges in the directed graph only retain short message transmission relationships between users that have a valid interaction frequency.

[0066] Optionally, the operation of constructing graph structure data corresponding to each directed graph structure includes: determining the node attributes of the corresponding nodes in the graph structure data based on the frequency of SMS messages sent by users corresponding to the nodes of the directed graph structure; and determining the edge attributes of the corresponding directed edges in the graph structure data based on the frequency of SMS messages sent from the starting user to the ending user corresponding to the directed edge of the directed graph structure and the deviation information between the user information of the starting user and the ending user.

[0067] Specifically, when constructing graph structure data corresponding to each directed graph structure, node attributes and edge attributes need to be determined separately. First, for each node in the directed graph structure, the node attributes are determined based on the total frequency of short messages sent by its corresponding user within its user subset. For example... Figure 7 For node N2, the total frequency of short messages sent to users N1, N3, N4, etc. (e.g., a total of 15 messages sent) is counted. This total is the attribute value of node N2, which intuitively reflects the scale of users' active interaction.

[0068] Next, for each directed edge in the directed graph structure, the edge attributes consist of two parts: one part is the specific frequency of short messages sent from the starting user to the ending user (e.g., N1 sends 8 short messages to N2 corresponding to E1, and this frequency is part of the edge attributes); the other part is the deviation information between the user information of the starting user and the ending user (e.g., cosine similarity calculated based on user interest preference vectors, differences in historical interaction behavior, etc.; the smaller the deviation, the stronger the connection between users). Integrating these two parts of information constitutes the edge attributes of the directed edge, used to comprehensively quantify the interaction strength and the degree of connection between user pairs.

[0069] The above methods provide structured and high-value data support for subsequent extraction of deep interaction patterns and accurate prediction of short message forwarding behavior between users through graph neural network models.

[0070] Optionally, the method further includes determining deviation information between the user information of the starting user and the ending user by performing the following operations, wherein the deviation information is used to quantify the preference differences between the two users in the dimension of preset interest elements: obtaining the preference degree vectors of the starting user and the ending user; calculating the preference differences between the starting user and the ending user in each preset interest element based on the preference degree vectors; constructing a deviation information vector based on the preference differences; and determining the deviation information between the user information of the starting user and the ending user based on the deviation information vector.

[0071] Specifically, to determine the deviation information between the user information of the starting user and the ending user, it is first necessary to obtain the preference vector I of the starting user. q =[u q,1 ,u q,2 ,...u q,L ] T Preference vector I of the endpoint user z =[u z,1 ,u z,2 ,...u z,L ] T Wherein, the vector dimension L is the same as the number of preset interest elements, u q,1 ~u q,L This represents the preference of the starting user q for each preset interest element (such as "dining", "clothing", "government services", etc.), with a value range of [0,1]. The closer the value is to 1, the higher the preference. z,1 ~u z,L This represents the preference of the endpoint user z for each preset interest element. Its value ranges from [0,1], and the closer the value is to 1, the higher the preference.

[0072] Then, for each preset interest element (from the 1st to the Lth), a uniform calculation method (such as the absolute difference method) is used. , ,..., ), calculate the preference differences between the starting user and the ending user on the preset interest element respectively:

[0073] in, This indicates the difference in preferences between the starting user and the ending user regarding the first preset interest element;

[0074] This indicates the difference in preferences between the starting user and the ending user regarding the second preset interest element;

[0075] And so on, This represents the preference difference between the starting user and the ending user for the Lth preset interest element;

[0076] Furthermore, based on the calculated differences in preferences among L preset interest elements, a deviation information vector is constructed. =[ , ,..., ] T The deviation information vector In a structured form, it represents the difference in preference between the starting user and the ending user on each preset interest element; the Δ at a certain position in this deviation information vector. i ( , ,..., The larger the absolute value difference (one of the values ​​in the equation), the greater the difference in preference between the two individuals for the corresponding interest element; the smaller the value, the more similar their preferences.

[0077] In other words, based on this deviation information vector, the deviation information between the user information of the starting user and the ending user is determined.

[0078] By using the above methods, the degree of correlation between user pairs can be quantified, providing accurate preference difference data support for graph neural network models to extract edge features and predict forwarding behavior.

[0079] Optionally, the operation of inputting graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data includes: using the message passing layer of the graph neural network model to pass messages to the node attributes of the graph structure data to determine the node features of the graph structure features; and using the message passing layer to pass messages to the edge attributes of the graph structure data to determine the edge features of the graph structure features.

[0080] Specifically, refer to Figure 8 As shown, the process of inputting graph structure data into a graph neural network model and processing the data through message passing layers (including message passing layer 1 and message passing layer 2) to extract graph structure features, and finally generating node features and edge features respectively, is as follows:

[0081] First, by aggregating node attributes using message passing layer 1, the node attributes (such as the total frequency of user messages within a subset) of each node in the graph structure data are updated. Message passing layer 1 allows each node to actively collect information from its neighboring nodes (i.e., other nodes directly connected by directed edges). For example, node N2 receives node attributes from its originating users (such as N1 and N3 who sent messages to N2) and its destination neighbors (such as N1 and N3 who sent messages to other users). These neighboring node attribute values ​​are combined with the weights of the corresponding edges (such as the interaction frequency in the edge attributes; higher frequency means higher weight). Aggregation is performed using methods such as weighted summation, mean calculation, or neural network mapping. This aggregation is then fused with the node's own node attributes to ultimately generate node features that reflect the node's interactive position in the entire subgraph and the strength of its association with neighboring nodes. In other words, node features are a comprehensive reflection of the node's own attributes and information from neighboring nodes.

[0082] Furthermore, message passing is performed on the edge attributes of the graph structure data based on message passing layer 2. When processing edges, message passing layer 2 utilizes both the edge's own attributes and the features of the nodes at both ends of the edge. For example, for edge E2 (N1→N2), message passing layer 2 fuses the edge attributes of E2 (such as the frequency of transmission from N1 to N2, and the preference deviation vector between N1 and N2) with the node features of N1 and N2 (e.g., by concatenating them before inputting into a fully connected layer), generating edge features that simultaneously reflect the original interaction features of the edge and the influence of the characteristics of the nodes at both ends on the interaction. In other words, the edge features are the result of the combined effect of the edge's own attributes and the nodes at both ends, and can more accurately characterize the forwarding association potential between two users.

[0083] Through the above methods, the node and edge features extracted from graph structure data by the graph neural network model not only retain the original interaction information but also incorporate global correlation patterns, providing deep and structured feature support for subsequent prediction of user forwarding behavior.

[0084] Optionally, based on the edge features of the graph structure features, a pre-trained neural network model is used to predict whether the starting user corresponding to the edge feature will forward the target SMS to the ending user. This includes: inputting the edge features of the graph structure features into the pre-trained neural network model to determine the probability value of the starting user corresponding to the edge feature forwarding the target SMS to the ending user; and determining that the starting user corresponding to the edge feature forwards the target SMS to the ending user if the probability value is greater than a preset probability threshold.

[0085] Specifically, refer to Figure 9As shown, the edge features of the graph structure are input into a pre-trained neural network model to determine the probability value of the starting user forwarding the target SMS to the ending user corresponding to the edge feature. In the embodiments of the present invention, the judgment criterion is based on a preset probability threshold (this threshold is determined based on historical SMS forwarding data statistics, for example, set to 0.7, corresponding to a historical scenario where the actual forwarding success rate of the starting user under this threshold is ≥85%): If, among all the corresponding starting users of a certain ending user, there is at least one starting user whose forwarding probability value is greater than the preset threshold, it is determined that the ending user does not need the server 200 to actively distribute the target SMS content, and the server 200 removes the ending user from the user set; if the forwarding probability values ​​of all the corresponding starting users of a certain ending user are less than or equal to the preset threshold, it is determined that the ending user cannot obtain the target SMS through spontaneous forwarding, and the server 200 needs to actively distribute it, and the server 200 retains it in the user set.

[0086] Therefore, this application achieves the technical effects of accurately screening target users to reduce invalid pushes, saving distribution resources, avoiding repeated pushes to users who can obtain messages through spontaneous forwarding to improve user experience, reducing unnecessary push volume to control mass sending costs, and ensuring that information reaches the users who really need it to optimize information transmission efficiency. It also solves the technical problems of invalid push waste, repeated push affecting user experience, high mass sending costs and unbalanced information transmission efficiency in the existing short message distribution technology.

[0087] In addition, refer to Figure 1 As shown, according to a second aspect of this embodiment, a storage medium is provided. The storage medium includes a stored program, wherein, when the program is executed, a processor performs any of the methods described above.

[0088] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0090] Example 2

[0091] Figure 10 A graph-based short message prediction and distribution apparatus according to this embodiment is shown, which corresponds to the method described in Embodiment 1. (Reference) Figure 10 As shown, the device includes: a set determination module 1010, used to determine a set of users associated with the content category of the target short message; a partitioning module 1020, used to partition the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set; a construction module 1030, used to construct graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of short message sending by users in the corresponding user subset and the frequency of short message sending between users; a graph structure feature determination module 1040, used to input the graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data; a prediction module 1050, used to predict whether the starting user corresponding to the edge feature will forward the target short message to the ending user based on the edge features of the graph structure features using a pre-trained neural network model; a target user determination module 1060, used to delete users who do not need to receive the target short message from the user set based on the prediction results and determine the target user of the target short message; and a distribution module 1070, used to distribute the target short message to the target user.

[0092] Optionally, the operation of constructing graph structure data corresponding to each user subset includes: constructing a directed graph structure corresponding to each user subset, wherein the nodes in the directed graph structure correspond to the users in the corresponding user subset, and the directed edges in the directed graph structure are used to indicate that the frequency of the starting user corresponding to the starting point sending short messages to the ending user corresponding to the ending point meets a preset condition; and constructing graph structure data corresponding to each directed graph structure, wherein the node attributes of the nodes in the graph structure data represent the frequency of the corresponding user sending short messages, and the edge attributes of the directed edges in the graph structure data represent the similarity between the starting user corresponding to the starting point and the ending user corresponding to the ending point and the frequency of the starting user sending short messages to the ending user.

[0093] Optionally, the operation of constructing a directed graph structure corresponding to each user subset includes: mapping users of the user subset to nodes of the corresponding directed graph structure; and determining whether to construct a directed edge starting from the first user and ending at the second user based on the frequency of the first user sending short messages to the second user in the user subset.

[0094] Optionally, the operation of constructing graph structure data corresponding to each directed graph structure includes: determining the node attributes of the corresponding nodes in the graph structure data based on the frequency of SMS messages sent by users corresponding to the nodes of the directed graph structure; and determining the edge attributes of the corresponding directed edges in the graph structure data based on the frequency of SMS messages sent from the starting user to the ending user corresponding to the directed edge of the directed graph structure and the deviation information between the user information of the starting user and the ending user.

[0095] Optionally, the device further includes a deviation information determination module, used to determine deviation information between the user information of the starting user and the ending user according to the following operations, wherein the deviation information is used to quantify the preference difference between the two in the dimension of preset interest elements: obtaining the preference degree vectors of the starting user and the ending user; calculating the preference difference between the starting user and the ending user on each preset interest element according to the preference degree vectors; constructing a deviation information vector according to the preference difference; and determining the deviation information between the user information of the starting user and the ending user according to the deviation information vector.

[0096] Optionally, the operation of inputting graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data includes: using the message passing layer of the graph neural network model to pass messages to the node attributes of the graph structure data to determine the node features of the graph structure features; and using the message passing layer to pass messages to the edge attributes of the graph structure data to determine the edge features of the graph structure features.

[0097] Optionally, based on the edge features of the graph structure features, a pre-trained neural network model is used to predict whether the starting user corresponding to the edge feature will forward the target SMS to the ending user. This includes: inputting the edge features of the graph structure features into the pre-trained neural network model to determine the probability value of the starting user corresponding to the edge feature forwarding the target SMS to the ending user; and determining that the starting user corresponding to the edge feature forwards the target SMS to the ending user if the probability value is greater than a preset probability threshold.

[0098] Therefore, this application achieves the technical effects of accurately screening target users to reduce invalid pushes, saving distribution resources, avoiding repeated pushes to users who can obtain messages through spontaneous forwarding to improve user experience, reducing unnecessary push volume to control mass sending costs, and ensuring that information reaches the users who really need it to optimize information transmission efficiency. It also solves the technical problems of invalid push waste, repeated push affecting user experience, high mass sending costs and unbalanced information transmission efficiency in the existing short message distribution technology.

[0099] Example 3

[0100] Figure 11 A graph-based short message prediction and distribution apparatus according to this embodiment is shown, which corresponds to the method described in Embodiment 1. Reference Figure 11 As shown, the device includes: a processor 1110; and a memory 1120 connected to the processor 1110, for providing the processor 1110 with instructions to process the following steps: determining a set of users associated with the content category of the target short message; dividing the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set; constructing graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of short message sending by users in the corresponding user subset and the frequency of short message sending between users; inputting the graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data; predicting, using the pre-trained neural network model, whether the starting user corresponding to the edge feature will forward the target short message to the ending user based on the edge features of the graph structure features; deleting users from the user set who do not need to receive the target short message based on the prediction results, and determining the target user of the target short message; and distributing the target short message to the target user.

[0101] Optionally, the operation of constructing graph structure data corresponding to each user subset includes: constructing a directed graph structure corresponding to each user subset, wherein the nodes in the directed graph structure correspond to the users in the corresponding user subset, and the directed edges in the directed graph structure are used to indicate that the frequency of the starting user corresponding to the starting point sending short messages to the ending user corresponding to the ending point meets a preset condition; and constructing graph structure data corresponding to each directed graph structure, wherein the node attributes of the nodes in the graph structure data represent the frequency of the corresponding user sending short messages, and the edge attributes of the directed edges in the graph structure data represent the similarity between the starting user corresponding to the starting point and the ending user corresponding to the ending point and the frequency of the starting user sending short messages to the ending user.

[0102] Optionally, the operation of constructing a directed graph structure corresponding to each user subset includes: mapping users of the user subset to nodes of the corresponding directed graph structure; and determining whether to construct a directed edge starting from the first user and ending at the second user based on the frequency of the first user sending short messages to the second user in the user subset.

[0103] Optionally, the operation of constructing graph structure data corresponding to each directed graph structure includes: determining the node attributes of the corresponding nodes in the graph structure data based on the frequency of SMS messages sent by users corresponding to the nodes of the directed graph structure; and determining the edge attributes of the corresponding directed edges in the graph structure data based on the frequency of SMS messages sent from the starting user to the ending user corresponding to the directed edge of the directed graph structure and the deviation information between the user information of the starting user and the ending user.

[0104] Optionally, the memory 1120 is also configured to provide the processor 1110 with instructions to process the following steps: determining deviation information between the user information of the starting user and the ending user according to the following operations, wherein the deviation information is used to quantify the preference difference between the two in a preset interest element dimension: obtaining the preference degree vector of the starting user and the ending user; calculating the preference difference between the starting user and the ending user on each preset interest element according to the preference degree vector; constructing a deviation information vector according to the preference difference; and determining the deviation information between the user information of the starting user and the ending user according to the deviation information vector.

[0105] Optionally, the operation of inputting graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data includes: using the message passing layer of the graph neural network model to pass messages to the node attributes of the graph structure data to determine the node features of the graph structure features; and using the message passing layer to pass messages to the edge attributes of the graph structure data to determine the edge features of the graph structure features.

[0106] Optionally, based on the edge features of the graph structure features, a pre-trained neural network model is used to predict whether the starting user corresponding to the edge feature will forward the target SMS to the ending user. This includes: inputting the edge features of the graph structure features into the pre-trained neural network model to determine the probability value of the starting user corresponding to the edge feature forwarding the target SMS to the ending user; and determining that the starting user corresponding to the edge feature forwards the target SMS to the ending user if the probability value is greater than a preset probability threshold.

[0107] Therefore, this application achieves the technical effects of accurately screening target users to reduce invalid pushes, saving distribution resources, avoiding repeated pushes to users who can obtain messages through spontaneous forwarding to improve user experience, reducing unnecessary push volume to control mass sending costs, and ensuring that information reaches the users who really need it to optimize information transmission efficiency. It also solves the technical problems of invalid push waste, repeated push affecting user experience, high mass sending costs and unbalanced information transmission efficiency in the existing short message distribution technology.

[0108] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0109] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0110] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0111] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0112] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0113] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or 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 the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0114] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A short message prediction and distribution method based on graph structure, characterized in that, include: Identify the set of users associated with the content category of the target SMS message; The user set is divided into multiple user subsets based on the frequency of short message communication between any two users in the user set; Construct graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of SMS messages sent by users in the corresponding user subset and the frequency of SMS messages sent between users; The graph structure data are input into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data. Based on the edge features of the graph structure features, a pre-trained neural network model is used to predict whether the starting user corresponding to the edge feature will forward the target short message to the ending user. The ending user is the user pointed to by the directed edge in the graph structure data, the starting user is the user that initiates the directed edge pointing to the ending user, and the directed edge is used to indicate that the starting user has forwarded the short message to the ending user. Based on the prediction results, users who do not need to receive the target SMS are removed from the user set, and the target users of the target SMS are determined. as well as The target SMS message is distributed to the target user, wherein The method further includes determining deviation information between the user information of the starting user and the ending user by performing the following operations, wherein the deviation information is used to quantify the preference difference between the two in a preset interest element dimension: obtaining the preference degree vector of the starting user and the ending user; calculating the preference difference between the starting user and the ending user on each preset interest element based on the preference degree vector; constructing a deviation information vector based on the preference difference; and determining the deviation information between the user information of the starting user and the ending user based on the deviation information vector.

2. The method according to claim 1, characterized in that, The operation of constructing graph structure data corresponding to each user subset includes: Construct a directed graph structure corresponding to each user subset, wherein the nodes in the directed graph structure correspond to users in the corresponding user subset, and the directed edges in the directed graph structure are used to indicate that the frequency at which the starting user (corresponding to the starting point) sends short messages to the ending user (corresponding to the ending point) meets a preset condition; and Construct graph structure data corresponding to each directed graph structure, wherein the node attributes of the nodes in the graph structure data represent the frequency of SMS messages sent by the corresponding users, and the edge attributes of the directed edges in the graph structure data represent the similarity between the starting user corresponding to the starting point and the ending user corresponding to the ending point, and the frequency of SMS messages sent by the starting user to the ending user.

3. The method according to claim 2, characterized in that, The operation of constructing the directed graph structure corresponding to each user subset includes: Assigning users of the user subset to nodes of the corresponding directed graph structure; and Based on the frequency with which the first user in the user subset sends short messages to the second user, determine whether to construct a directed edge that starts from the first user and ends at the second user.

4. The method according to claim 2, characterized in that, The operations for constructing graph structure data corresponding to each directed graph structure include: Based on the frequency of SMS messages sent by users corresponding to nodes in the directed graph structure, the node attributes of the corresponding nodes in the graph structure data are determined; and Based on the frequency of short messages sent from the starting user to the ending user corresponding to the directed edge of the directed graph structure, and the deviation information between the user information of the starting user and the ending user, the edge attributes of the corresponding directed edges in the graph structure data are determined.

5. The method according to claim 1, characterized in that, The operation of inputting the graph structure data into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data includes: Using the message passing layer of the graph neural network model, message passing is performed on the node attributes of the graph structure data to determine the node features of the graph structure features; and Using the message passing layer, message passing is performed on the edge attributes of the graph structure data to determine the edge features of the graph structure.

6. The method according to claim 1, characterized in that, Based on the edge features of the graph structure, and using a pre-trained neural network model, the operation of whether the starting user corresponding to the edge feature forwards the target short message to the ending user is predicted, including: The edge features of the graph structure are input into a pre-trained neural network model to determine the probability value of the starting user forwarding the target SMS message to the ending user corresponding to the edge features; and If the probability value is greater than a preset probability threshold, it is determined that the starting user corresponding to the edge feature forwards the target short message to the ending user.

7. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, the method described in any one of claims 1 to 6 is performed by a processor.

8. A short message prediction and distribution device based on a graph structure, characterized in that, include: The set determination module is used to determine the set of users associated with the content category of the target SMS message; The partitioning module is used to divide the user set into multiple user subsets based on the frequency of short message communication between any two users in the user set; The construction module is used to construct graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of SMS messages sent by users in the corresponding user subset and the frequency of SMS messages sent between users; The graph structure feature determination module is used to input the graph structure data into the pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data. The prediction module is used to predict, based on the edge features of the graph structure features and using a pre-trained neural network model, whether the starting user corresponding to the edge features will forward the target short message to the ending user. The ending user is the user pointed to by the directed edge in the graph structure data, the starting user is the user that initiates the directed edge pointing to the ending user, and the directed edge is used to indicate that the starting user has forwarded the short message to the ending user. The target user determination module is used to delete users who do not need to receive the target SMS from the user set based on the prediction results, and to determine the target user of the target SMS; The distribution module is used to distribute the target short message to the target user; as well as The deviation information determination module is used to determine the deviation information between the user information of the starting user and the ending user according to the following operations, wherein the deviation information is used to quantify the preference difference between the two in the dimension of preset interest elements: obtaining the preference degree vector of the starting user and the ending user; calculating the preference difference between the starting user and the ending user on each preset interest element according to the preference degree vector; constructing a deviation information vector according to the preference difference; and determining the deviation information between the user information of the starting user and the ending user according to the deviation information vector.

9. A short message prediction and distribution device based on a graph structure, characterized in that, include: processor; as well as A memory, connected to the processor, for providing the processor with instructions to perform the following processing steps: Identify the set of users associated with the content category of the target SMS message; The user set is divided into multiple user subsets based on the frequency of short message communication between any two users in the user set; Construct graph structure data corresponding to each user subset, wherein the graph structure data is used to indicate the frequency of SMS messages sent by users in the corresponding user subset and the frequency of SMS messages sent between users; The graph structure data are input into a pre-trained graph neural network model to determine the graph structure features corresponding to each graph structure data. Based on the edge features of the graph structure features, a pre-trained neural network model is used to predict whether the starting user corresponding to the edge feature will forward the target short message to the ending user. The ending user is the user pointed to by the directed edge in the graph structure data, the starting user is the user that initiates the directed edge pointing to the ending user, and the directed edge is used to indicate that the starting user has forwarded the short message to the ending user. Based on the prediction results, users who do not need to receive the target SMS are removed from the user set, and the target users of the target SMS are determined. as well as Distributing the target short message to the target user, the method further includes determining deviation information between the user information of the starting user and the ending user by performing the following operations, wherein the deviation information is used to quantify the preference difference between the two in a preset interest element dimension: obtaining the preference degree vector of the starting user and the ending user; calculating the preference difference between the starting user and the ending user on each preset interest element based on the preference degree vector; constructing a deviation information vector based on the preference difference; and determining the deviation information between the user information of the starting user and the ending user based on the deviation information vector.

Images