A method and device for information diffusion prediction based on dynamic interaction relationship perception

Abstract

The application discloses a kind of information diffusion prediction method and device based on dynamic interaction relationship perception, belong to the direction of information diffusion prediction in artificial intelligence technical field.It is used to solve the problem that the existing method is difficult to accurately represent the dynamic relationship and potential preference of users in the information propagation environment, resulting in poor prediction performance.Using graph neural differential equation to model the potential features of users in social relationship graph and information diffusion graph in continuous time can effectively perceive and represent the dynamic interaction relationship and potential preference change of users in the global information propagation environment, solving the problems of unreasonable time granularity selection and difficulty in naturally capturing the continuous evolution of user preferences and behaviors over time in the existing graph representation-based prediction methods, effectively improving the accuracy of information diffusion prediction.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of artificial intelligence, and particularly relates to an information diffusion prediction method and device based on dynamic interaction relationship perception. BACKGROUND

[0002] The rise of social media and wireless technology has completely changed the pattern of online information sharing. Unlike traditional communication methods, digital platforms give users the ability to quickly spread content, triggering a ripple effect often referred to as "information cascades." These information cascades describe situations where a large number of individuals make similar decisions in succession, which can influence public discourse, set trends, and even affect real-world behavior. Therefore, how to effectively perceive the situation of the information cascade process and then carry out situation prediction and risk warning has become an important issue in network space governance. In actual research and application, the focus of information diffusion prediction is to perceive and predict future diffusion dynamics based on observable social interaction records, especially to predict information propagation paths at the user level and identify the next user that may be affected in a timely manner, which is of great significance for information recommendation, information interception, risk assessment and other applications.

[0003] Before the development of deep learning technology, the research on information diffusion prediction methods mainly focused on generative diffusion models and feature engineering. Methods based on feature engineering use carefully designed features as input and use matrix decomposition, tensor decomposition, and other techniques to obtain cascade prediction results. Methods based on generative diffusion models follow pre-set propagation rules to let information spread from active users to inactive users through social relationships, thereby simulating the real propagation process. For example, some studies use the heat diffusion equation to model information diffusion and embed users in a low-dimensional space; other studies use independent cascade models to embed users in a hidden space in social networks to obtain more reliable diffusion probabilities. However, these methods usually rely on strong assumptions and fixed parameters, limiting their generalization ability and expression ability. In addition, simplifying the information diffusion mechanism often reduces prediction accuracy and limits the ability to model real dynamics.

[0004] In recent years, deep learning-based methods have become mainstream in the field of information diffusion prediction due to their powerful representation capabilities. Existing mainstream methods can be broadly categorized into sequence representation-based methods and graph representation-based methods. Sequence representation-based methods primarily utilize models such as Recurrent Neural Networks (RNNs), Convolutional Neural Networks (CNNs), and Transformers, combined with the unique characteristics of information cascading, to enhance the representational capabilities of cascaded system patterns. In contrast, graph representation-based methods focus on learning high-quality user representations from graph structures. By constructing various types of graph structures such as social graphs and diffusion graphs, different user role attributes can be effectively captured. Some methods further divide the cascade into different time intervals, create multiple snapshots of the dynamic graph or hypergraph for each interval, and capture the user's dynamic preferences and spatiotemporal characteristics through fusion mechanisms.

[0005] However, existing graph-based information diffusion prediction methods suffer from two main drawbacks: 1) Temporal discretization bias: Traditional sequence graph or hypergraph methods require dividing time into discrete intervals, while real-world cascading occurs in continuous time. Therefore, this method of discretizing time intervals leads to unreasonable selection of temporal granularity, thus affecting prediction accuracy. 2) Lack of continuity in user node representations: Sequence graph and hypergraph methods typically construct independent snapshots for each time interval, and the node representations of different snapshots are spliced ​​together using a gating fusion module. However, this "snapshot splicing" method cannot model the feature transition process between adjacent snapshots, resulting in a "jumping" evolution of user preferences, failing to naturally reflect the continuous changes in user behavior over time in real-world propagation. In summary, existing methods struggle to accurately characterize the dynamic relationships and potential preferences of users in the information propagation environment, leading to poor prediction performance. Summary of the Invention

[0006] To address the problems existing in the prior art, embodiments of the present invention provide an information diffusion prediction method and apparatus based on dynamic interaction relationship perception, which is used to solve the problem that existing methods are unable to accurately characterize the dynamic relationships and potential preferences of users in the information dissemination environment, thus resulting in poor prediction performance.

[0007] This invention provides an information diffusion prediction method based on dynamic interaction relationship perception, comprising:

[0008] Obtain the cascaded dataset and static social relationship graph, and construct a static user-cascaded bipartite graph, a static information diffusion graph, a static information diffusion hypergraph, and a new heterogeneous graph;

[0009] The initial node representation of the new heterogeneous graph is obtained through a graph convolutional network (GCN). The graph neural differential equations for the evolution of the new heterogeneous nodes over time are constructed to determine the edge weights of the new heterogeneous dynamic graph and the node representations of the new heterogeneous graph at different times.

[0010] The participation time is normalized and evenly divided to obtain multiple intervals and key time points. Based on the node representation of the new heterogeneous graph, the node representation of the new heterogeneous graph corresponding to each key time point is determined. The first participation time included in the first concatenation is normalized to obtain the first time. Based on the intervals and the first time, the final dynamic node representation of the first user is determined.

[0011] The user's final dynamic node representation and the user's static node representation are fused through an attention-gated fusion mechanism to obtain the user's comprehensive node representation; the user's comprehensive node representation is then fused through multi-head attention to obtain the cascaded final features; the cascaded final features are then used by a multilayer perceptron to predict the probability of each user participating in the propagation in the future.

[0012] This invention provides an information diffusion prediction device based on dynamic interaction relationship perception, comprising:

[0013] The building unit is used to acquire cascaded datasets and static social relationship graphs, and to construct static user-cascaded bipartite graphs, static information diffusion graphs, static information diffusion hypergraphs, and new heterogeneous graphs.

[0014] The first determining unit is used to obtain the initial node representation of the new heterogeneous graph through the graph convolutional network (GCN), construct the graph neural differential equation of the evolution of the new heterogeneous nodes over time, and determine the edge weights of the new heterogeneous dynamic graph and the node representation of the new heterogeneous graph at different times.

[0015] The second determining unit is used to normalize the participation time and divide it equally to obtain multiple intervals and key time points, and determine the node representation of the new heterogeneous graph corresponding to each key time point based on the node representation of the new heterogeneous graph; normalize the first participation time included in the first concatenation to obtain the first time, and determine the final dynamic node representation of the first user based on the intervals and the first time.

[0016] The prediction unit is used to fuse the user's final dynamic node representation and the user's static node representation through an attention-gated fusion mechanism to obtain the user's comprehensive node representation; the user's comprehensive node representation is fused through multi-head attention to obtain cascaded final features, and the cascaded final features are used to predict the probability of each user participating in propagation in the future through a multilayer perceptron.

[0017] This invention provides a computer device, which includes a memory and a processor. The memory stores a computer program, and when the computer program is executed by the processor, the processor performs any of the above-described information diffusion prediction methods based on dynamic interaction relationship perception.

[0018] This invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform any of the above-described information diffusion prediction methods based on dynamic interaction relationship awareness.

[0019] This invention provides a method and apparatus for predicting information diffusion based on dynamic interaction relationship perception. From a modeling perspective, this method uses graph neural differential equations to characterize the dynamic interaction relationships and potential preferences between users over continuous time, thereby improving the accuracy of diffusion prediction and the ability to perceive risk situations. First, it collects real information dissemination event data from social networks and users' social relationship networks to construct a user-cascaded bipartite graph, an information diffusion relationship graph, and an information diffusion hypergraph. Second, it initializes the node representations of users and information cascades using a gating mechanism, and uses graph neural differential equations to characterize the node states of users during information dissemination to achieve dynamic interaction and global situational awareness of information diffusion. Combined with the user's temporal information within the information cascade, it retrieves the dynamic node representation at a specific time and achieves real-time perception and analysis of the diffusion situation. Then, it uses graph convolutional networks and hypergraph convolutional networks to obtain the user's static node representation, and fuses the dynamic and static node representations through an attention gating fusion mechanism. Finally, combining the user node representation and temporal information, it uses a multi-head attention mechanism to obtain the features of the information cascade, and employs a multilayer perceptron to predict the likelihood of future user dissemination. During the training phase, the cross-entropy loss function is used to optimize the model parameters. During the testing phase, the trained model is used to predict users who may participate in events. The method provided in this invention utilizes graph neural differential equations to model the latent features of users in continuous-time social relationship graphs and information diffusion graphs. This effectively perceives and characterizes the dynamic interaction relationships and potential preference changes of users in the global information dissemination environment. It solves the problems of existing graph representation-based prediction methods, such as unreasonable selection of time granularity and difficulty in naturally capturing the continuous evolution of user preferences and behaviors over time, effectively improving the accuracy of information diffusion prediction. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of an information diffusion prediction method based on dynamic interaction relationship perception provided in an embodiment of the present invention;

[0022] Figure 2 A schematic diagram illustrating how to obtain the final dynamic node representation of a user by utilizing the user's time retrieval in a cascade, as provided in an embodiment of the present invention.

[0023] Figure 3A This invention provides a schematic diagram illustrating the relationship between training time and performance metrics for an Android dataset.

[0024] Figure 3B This invention provides a schematic diagram illustrating the relationship between training time and performance metrics on the Douban dataset for embodiments of the invention.

[0025] Figure 4 This is a schematic diagram of an information diffusion prediction device based on dynamic interaction relationship perception provided in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0027] 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.

[0028] Dynamic graph structures are essentially "time-dependent graph structures," meaning that the graph's components (nodes, edges, edge weights) change over time. Updated according to changes, the formula is usually expressed in terms of... Or superscript with time parameter , Clearly stated, in contrast to "static graph structures (such as static social relationship graphs)". Initial bipartite graph "In contrast, the nodes and edges of a static graph are fixed in the long term and do not change over time."

[0029] Figure 1 This is a schematic diagram of a method for predicting information diffusion based on dynamic interaction relationship awareness, provided in an embodiment of the present invention. Figure 1As shown, the method includes the following steps:

[0030] Step 101: Obtain the cascaded data set and static social relationship graph, and construct the static user-cascaded bipartite graph, static information diffusion graph, static information diffusion hypergraph, and new heterogeneous graph;

[0031] Step 102: Obtain the initial node representation of the new heterogeneous graph through the graph convolutional network (GCN), construct the dynamic edge-coupled graph neural differential equation of the new heterogeneous nodes evolving over time, and determine the edge weights of the new heterogeneous dynamic graph and the node representation of the new heterogeneous graph at different times.

[0032] Step 103: Normalize the participation time and divide it equally to obtain multiple intervals and key time points. Determine the node representation of the new heterogeneous graph corresponding to each key time point based on the node representation of the new heterogeneous graph. Normalize the first participation time included in the first concatenation to obtain the first time. Determine the final dynamic node representation of the first user based on the intervals and the first time.

[0033] Step 104: The user's final dynamic node representation and the user's static node representation are fused through an attention gating fusion mechanism to obtain the user's comprehensive node representation; the user's comprehensive node representation is fused through multi-head attention to obtain the cascaded final features; the cascaded final features are used to predict the probability of each user participating in propagation in the future through a multilayer perceptron.

[0034] In step 101, it is necessary to collect cascaded data sets and static social relationship graphs on social networks.

[0035] Specifically, cascaded data sets include Each cascade records the complete trajectory of information propagation, including users, participation time, and propagation order; that is, each cascade is a combined record of "users, participation time, and propagation order," fully reflecting the user participation sequence and timeline from the starting point of information propagation to its subsequent spread. In this embodiment of the invention, a cascade can be represented as:

[0036] (1)

[0037] in, Indicates the first One user who participated in this cascading propagation, Indicates the order of propagation. This indicates the total number of users participating in the propagation within the cascade (propagation length). Indicates the first The original participation time of the k-th user in each cascade is the discrete user participation time. This represents a set of users, which includes the IDs of all users. , This indicates a single cascade. It should be noted that, unless otherwise specified, the "participation time" mentioned later may refer to the original participation time or the participation time after normalization or other processing, and the specific interpretation should be based on the operational scenario.

[0038] In this embodiment of the invention, cascading is an abbreviation for "information cascading," which refers to the complete trajectory of information propagation in a social network, including users, participation time, and propagation order; user is an abbreviation for "users participating in information cascading propagation," that is, individuals participating in the information cascading propagation process.

[0039] Furthermore, a static social relationship graph is a structured representation of the social connections between users, specifically including user sets and social relationship edges. Specifically, a static social relationship graph is represented as follows:

[0040] (2)

[0041] in, This represents a static social relationship graph. This represents a set of users, which includes the IDs of all users. Represents a set of social relationship edges. Indicates user and users There are social relationships, such as friends or following each other.

[0042] It should be noted that static social relationship graphs represent... It is an abstract representation of the entire network, in which This represents the overall structure of the "static social relationship graph," with the superscript 's' indicating the "social" attribute, and is related to the subsequent static information diffusion graph. Static User-Cascaded Bipartite Graph The formation of distinctions (the superscript d indicates diffusion, and b indicates dichotomy).

[0043] In practical applications, the data included in cascaded datasets and static social relationship graphs are discrete and cannot be directly used for feature learning in models such as GCN. To address this issue, this invention proposes constructing static user-cascaded bipartite graphs, static information diffusion graphs, and static information diffusion hypergraphs based on the aforementioned data. This transforms unstructured data into a structured representation of "nodes and edges," capturing the association patterns between users and cascades, and between users themselves.

[0044] Specifically, a static user-cascaded bipartite graph is constructed based on the cascaded data set. Specifically, for each cascade... Iterate through all the users it contains. In all users With each cascade Establish an edge between them to form a static user-cascaded bipartite graph, as shown below:

[0045] (3)

[0046] in, This represents a static user-cascaded bipartite graph. This represents the set of nodes in a static user-cascaded bipartite graph, which includes two types of nodes: one type is the IDs of all users, such as... Another type is the numbering of all cascades, such as... ; Represents the user-cascaded binary edge set. Indicates user Participated in the Cascade The process of its spread.

[0047] Specifically, a static information diffusion graph is constructed based on the cascaded data set. Specifically, the propagation sequence of each cascade (e.g., ...) is traversed. ), for adjacent user pairs in the propagation sequence An edge is then established between the two to form a static information diffusion graph, as shown below:

[0048] (4)

[0049] in, This represents a static information diffusion diagram. Represents a set of users. Represents the set of edges in a static information diffusion graph. Indicates user and users At least in one cascade, there exists a relationship of "adjacent propagation order" (i.e., After spreading, Then spread, or After spreading, Then it propagates (i.e., bidirectional adjacency). It should be noted that if the same pair of users are adjacent in multiple cascades, only one edge needs to be retained.

[0050] Furthermore, a static information diffusion hypergraph is constructed based on the cascaded data set. Specifically, each cascaded... Corresponding to a super edge This will include all users in the cascade (i.e. Included Both are related to the superedge The connections form a static information diffusion hypergraph, i.e., a "single hyperedge - multiple users" association, as shown below:

[0051] (5)

[0052] in, This represents a static information diffusion hypergraph. Represents a set of users. Indicates information diffusion hyperedge, Indicates a cascading message Super-edge It is cascaded A concrete representation in a static information diffusion hypergraph, namely each cascade A unique corresponding hyperedge in the static information diffusion hypergraph The set of nodes on the superedge is all the users participating in the cascade.

[0053] It should be noted that the novel heterogeneous graph provided in the embodiments of the present invention includes a dynamic novel heterogeneous diffusion graph, a dynamic novel heterogeneous social graph, and a static novel heterogeneous relationship graph.

[0054] Furthermore, based on static user-cascaded bipartite graphs, static information diffusion graphs, and static information diffusion hypergraphs, dynamic new heterogeneous diffusion graphs, dynamic new heterogeneous social graphs, and static new heterogeneous relationship graphs can also be constructed, as shown below:

[0055] (6)

[0056] (7)

[0057] (8)

[0058] in, This represents a dynamic new heterogeneous diffusion diagram. This represents a static user-cascaded bipartite graph. This represents a static information diffusion diagram. This represents a dynamic, new, heterogeneous social graph. This represents a static social relationship graph. This represents a static novel heterogeneous relationship graph. It should be noted that the novel heterogeneous graphs in this embodiment include dynamic novel heterogeneous diffusion graphs, dynamic novel heterogeneous social graphs, and static novel heterogeneous relationship graphs.

[0059] Furthermore, a gating mechanism is used to initialize users and cascades. The gating mechanism is a function used to filter and optimize the original features of nodes. Its core function is to selectively highlight key features and suppress secondary features from the original node representation containing users and cascades.

[0060] The gating mechanism processes the raw representations containing all users, obtaining a set of initial representations focusing on static social attributes and a set focusing on dynamic preference nodes through two different gating functions. Specifically, the gating mechanism uses a "filtering factor" ( (Function output) on the original features The initial representation (including all users and cascading representations) is weighted using the following formula:

[0061] (9)

[0062] (10)

[0063] (11)

[0064] in, This represents the set of node representations that, after being processed by a gating function, focus on a specific attribute; specifically, it represents the initial set of node representations. The specific attributes obtained after gating and filtering, among which, These represent different attributes, such as static social attribute s and dynamic preference attribute o; This represents a general abstract representation of all gated functions. The trainable weight matrix representing the gating mechanism. This represents the trainable bias parameters of the gating mechanism. This represents the Hadamard product operation on matrices. Represents the normalized exponential function, Static social attribute nodes represent sets. The dynamic preference node represents a set. This represents the set of initial nodes that contain all users and cascades.

[0065] In practical applications, Its core function is to generate a "gated mask" with values ​​between 0 and 1 using the sigmoid normalized exponential function, and then combine it with the initial representation. Perform the Hadamarda step to achieve the following: The selection of different attribute features involves retaining important features and suppressing irrelevant features.

[0066] Furthermore, the set of nodes representing static social attributes By using a gating function to filter static and dynamic preference features The results focus on features related to social behavior, such as user social activity and user interest relevance. The initial representation set of dynamic preference nodes... Gating functions used to filter dynamic interactive attribute features The focus is on characteristics related to the dissemination behavior, such as the frequency of user dissemination and the speed of cascading dissemination.

[0067] In step 102, the dynamic new heterogeneous diffusion graph is formed by merging the static user-cascaded bipartite graph and the static information diffusion graph, that is:

[0068] (6)

[0069] in, This represents a dynamic new heterogeneous diffusion diagram. This represents a static user-cascaded bipartite graph. This represents a static information diffusion diagram, which, in practical applications, Provides user-cascaded binary edges for "user-cascaded" It reflects the user's participation relationship with the cascading system. Provides static information diffusion graph edges for "user-to-user" connections. This reflects the direct propagation relationships between users. The merged dynamic new heterogeneous diffusion diagram The node set is users and cascades, and the edge set is... It contains two relation types.

[0070] Furthermore, a gating function representing the dynamic diffusion relationship is used. Initial representation set of dynamic preference nodes Features related to diffusion relationships are obtained through screening; then GCN is used in dynamic new heterogeneous diffusion maps. The neighbor features at each stage are aggregated to obtain the node representation at the initial time step of the GCN. This serves as the starting state for subsequent dynamic evolution. It should be noted that the dynamic novel heterogeneous diffusion graph employs dynamically edge-coupled graph neural differential equations, while the dynamic novel heterogeneous social graph employs highly scalable graph neural differential equations.

[0071] (12-1)

[0072] in, This represents the set of nodes in the dynamic new heterogeneous diffusion graph at the initial moment. Indicates GCN operation, This represents a gating function for dynamic diffusion relationships, used to filter features related to diffusion. This represents the initial set of dynamic preference nodes. This represents a dynamic new heterogeneous diffusion diagram.

[0073] Furthermore, GCN's update rules are as follows:

[0074] (13)

[0075] in, Represents a node The updated representation after GCN integrates nodes. Characteristics of itself and its neighboring nodes This represents a non-linear activation function used to perform non-linear transformations on the aggregated features, enhancing the model's expressive power. Representing graph structure Any node in the middle, Represents a node In graph structure The set of neighboring nodes in the middle, Represents the edges in any graph The weights are used to adjust the neighboring nodes. right The degree of influence is 1 under the default conditions. Representing neighboring nodes The expression, Indicates a node All neighboring nodes Sum the corresponding edge weights. As a normalization factor, it avoids inconsistencies in feature scales caused by differences in the number of neighbors. This represents the general trainable weight matrix of the GCN, used to perform a linear transformation on the representations of neighboring nodes. Representing neighboring nodes The original representation of .

[0076] It should be noted that graph structure Representing an arbitrary graph structure, in this embodiment of the invention, it can represent a dynamic novel heterogeneous diffusion graph. Dynamic new heterogeneous social graph Static new heterogeneous relationship diagram Any one of them is a node. and The neighbor relationships provide a topological basis. Representing network structure Middle The weight is 1 under the default conditions.

[0077] Furthermore, a graph neural differential equation for the dynamic edge coupling of nodes in the new heterogeneous diffusion graph is constructed to determine the edge weights of the nodes in the new heterogeneous diffusion graph and the node representations of the new heterogeneous diffusion graph at different times.

[0078] Specifically, the dynamic edge-coupled graph neural differential equation is a process of "defining the rules" of dynamic evolution, used to describe the time-varying rules of nodes in a dynamic new heterogeneous diffusion graph. The core of this equation is the definition. rate of change, express The set of node representations of the dynamic new heterogeneous diffusion graph at any given time is driven by two parts, specifically:

[0079] Structures that partially depend on dynamic novel heterogeneous diffusion maps Specifically, the node feature update amount under the current dynamic new heterogeneous diffusion graph can be calculated through the graph convolutional network (GCN). The difference between this update amount and the current feature is the driver of this change, which reflects the influence of neighbor relationships on the feature.

[0080] Another part depends on the node features themselves. Specifically, the update amount of the feature itself is obtained through nonlinear transformation of the feature, and the difference between this update and the current feature is the driving force of this change. The entire dynamically edge-coupled graph neural differential equation combines the dynamic changes of the dynamic new heterogeneous diffusion graph with the evolution of the feature itself, forming a complete time evolution rule.

[0081] In this embodiment of the invention, the graph neural differential equation with dynamic edge coupling can be expressed by the following formula:

[0082] (14)

[0083] (14-1)

[0084] (14-2)

[0085] in, express Time node representation set Time derivative, The weight parameters representing the balance between the two evolution functions control the relative importance of the influence of the graph structure and the influence of the node's own characteristics. This represents the evolution function associated with the dynamic graph structure, and outputs the influence of the graph structure on node evolution. This represents the evolution function related to the node's own characteristics, and outputs the influence of the node's own attributes on the evolution. express The set of node representations for a dynamic, new heterogeneous diffusion graph at any given time. express Dynamic new heterogeneous diffusion diagram at time t, This represents the evolution function associated with the dynamic graph structure. This represents the evolution function related to the node's own characteristics. Indicates based on Time node representation and animated GIFs The graph convolution operation outputs a node representation that incorporates neighbor information. express The set of node representations for a dynamic, new heterogeneous diffusion graph at any given time. express Dynamic new heterogeneous diffusion diagram at time t, This represents the trainable weight matrix in the differential equation of the dynamic novel heterogeneous diffusion graph, used to characterize the internal correlations of node features. This represents the bias vector in the differential equation of the dynamic new heterogeneous diffusion graph, used to adjust the offset of the internal correlation of node features, and can be used to train the weight matrix. Synergistic effect on feature transformation, express The transpose of .

[0086] It should be noted that, The instantaneous evolution rate of node features is characterized by a larger value, indicating that the node features change more drastically at that moment. It is the core indicator for describing the continuous evolution of node representation. Used to adjust the relative importance of "the influence of graph structure on evolution" and "the influence of node characteristics on evolution". The larger the value, the stronger the constraint that the graph structure places on the evolution of nodes.

[0087] Furthermore, regarding the dynamic novel heterogeneous diffusion map For each edge in the array (connecting users or users to cascades), calculate its position in the array. Time weight First, determine the first appearance time of the nodes at both ends of the edge: the default first appearance time of a user node is 0 (always active), and the first appearance time of a cascade node is the participation time of its first user; only when the first appearance times of both nodes are no earlier than 0... Time (i.e.) (Only when all nodes have participated in the propagation) are edges considered "active," at which point the weight is determined by the weight of the two nodes. The features at each time point are concatenated and then calculated using a function; otherwise, the weight is 0. In this way, The dynamic new heterogeneous diffusion map at any given time can dynamically filter out The propagation relationship that is effective at all times is the structural dependency term in the evolution equation. Provides accurate topology.

[0088] Specifically, any edge in the dynamic novel heterogeneous diffusion graph The corresponding weight Determined by the following formula:

[0089] (15-1)

[0090] in, express Edge of the dynamic new heterogeneous diffusion graph at any time The weight, Let represent the normalized exponential function, and the formula is: The output is mapped to the [0,1] interval, representing the strength of the edge weights. This represents the function for calculating the dynamic diffusion edge weights. The input is the concatenated node representations, and the output is the median value of the edge weights. Indicates that the node and nodes exist Representation of time and splicing operation, Representing neighboring nodes exist Representation in the dynamic neo-heterogeneous diffusion diagram at time t. Represents a node exist Representation in the dynamic neo-heterogeneous diffusion diagram at time t. Represents a node The time when it first appears in the entire information dissemination environment, if For the user, then ,like For cascading, then This indicates the time when the first user in the cascade participated in the propagation. Indicates filtering For node pairs that have appeared at or before a certain time, only edges of such node pairs are assigned weights; otherwise, the edge weight is 0.

[0091] Furthermore, the Ordinary Differential Equation Solver (ODESolver) is used, with the nodes of the dynamic new heterogeneous diffusion diagram at the initial time as the representation. Starting with the previously defined dynamic edge-coupled graph neural differential equation, we can solve for the results at different time points. Node representation of dynamic novel heterogeneous diffusion graph The details are as follows:

[0092] (16-1)

[0093] in, express The set of node representations for a dynamic, new heterogeneous diffusion graph at any given time. This refers to a tool for numerically solving ordinary differential equations. This represents the set of nodes in the dynamic new heterogeneous diffusion graph at the initial moment. This represents the target time point, i.e., the specific moment represented by the node that needs to be solved.

[0094] Furthermore, dynamic new heterogeneous social graphs include: Provides user-cascaded binary edges for "user-cascaded" It reflects the user's participation relationship with the cascading system. It provides a "user-to-user" social relationship edge. The merged dynamic new heterogeneous social graph reflects users' inherent social connections. The node set is users and cascades, and the edge set is... .

[0095] Furthermore, a gating function representing a dynamic new heterogeneous social graph is used. Initial representation set of dynamic preference nodes The system filters features related to social interaction, such as users' long-term social preferences and social activity levels. These filtered features are then input into the GCN (Generative Network Context) based on a dynamic, novel, heterogeneous social graph. The topological structure is used to aggregate the social neighbor features of each node, ultimately obtaining the node representation at the initial time step. This serves as the initial state for the evolution of dynamic, novel, and heterogeneous social graphs.

[0096] (12-2)

[0097] in, This represents the set of node representations in the dynamic new heterogeneous social graph at the initial moment. This represents a graph convolutional network operation. The input is node features and graph structure, and the output is a node representation after aggregating neighbor information. The gate function represents a dynamic, novel, heterogeneous social graph. This represents the initial set of dynamic preference nodes. This represents a dynamic, new, heterogeneous social graph, which is composed of a static user-cascaded bipartite graph. With static social relationship graph The resulting dynamic new heterogeneous social graph (i.e.) ).

[0098] For input feature vectors and topology of The update rules for its node representation are shown in formula (13), and will not be elaborated here.

[0099] Furthermore, a highly scalable graph neural differential equation is constructed to represent the evolution of nodes in a dynamic new heterogeneous social graph over time, and the edge weights of the nodes in the dynamic new heterogeneous social graph and the node representations of the dynamic new heterogeneous social graph at different time points are determined.

[0100] Specifically, the highly scalable graph neural differential equation is a process of "defining the rules" for the dynamic evolution of social relationships, used to describe the rules governing the changes in node representations over time. The core of this equation is the definition. rate of change, express The set of node representations in a constantly evolving, heterogeneous social graph is driven by both "user feature similarity" and "social structure dependency." Specifically, the rate of change of a single user's features equals the product of its "feature difference" with all other users multiplied by the sum of "similarity weight" and "structural weight." This reflects the dynamic similarity of user characteristics; the more similar the characteristics, the stronger the mutual influence. This reflects the structural constraints of social networks; the closer the social distance, the more significant the impact.

[0101] Specifically, the highly scalable graph neural differential equation is shown below:

[0102] (17)

[0103] (17-1)

[0104] (17-2)

[0105] in, express User nodes at any time Representation in dynamic neo-heterogeneous social graphs The time derivative of represents its characteristic instantaneous rate of change. express User nodes at any time Representation in dynamic new heterogeneous social graphs This indicates the total number of users and cascading nodes. express Time Node and The normalized weights of the social representation cosine similarity reflect the nodes With nodes Dynamic similarity, Represents static structural weights. , Represents a node In motion The number of neighbors in the structure reflects static structural constraints. Representing neighboring nodes With the current node The term represents the difference. express User nodes at any time and The original similarity, with values ​​ranging from [0,2], express User nodes at any time With all user nodes The sum of the original similarities, This indicates the user set excluding user nodes. Any user index outside of, , express A set of node representations in a constantly evolving and heterogeneous social graph, containing the social characteristics of all users. Represents user node exist Representation of dynamic new heterogeneous social graphs in time The result after L2 norm normalization This represents the L2 norm.

[0106] Dynamic New Heterogeneous Social Graph The calculation of edge weights is a "structural selection" process in the dynamic evolution of social relationships, targeting dynamic and heterogeneous social graphs. Each social edge in the process is determined based on the node's "first appearance time". Whether the edge is active at any given time – an edge is only retained (with a weight of 1) if both users at both ends of the edge have participated in the propagation (first appearance time ≤ t), otherwise the weight is 0 (the edge is inactive).

[0107] Specifically, any edge in a dynamic new heterogeneous social graph The corresponding weight Determined by the following formula:

[0108] (15-2)

[0109] in, express The edge of a dynamic and heterogeneous social graph The weight, , Represents a node In dynamic new heterogeneous social graphs The number of neighbors in a dynamic new heterogeneous social graph The edges in the algorithm are filtered based on the occurrence time of each node.

[0110] Furthermore, a numerical solution method for ordinary differential equations (ODESolver) is used, with the nodes of the dynamic new heterogeneous social graph at the initial time as the basis. Starting with the previously defined highly scalable graph neural differential equation, we can solve for the results at different time points. Node representation of dynamic new heterogeneous social graph The details are as follows:

[0111] (16-2)

[0112] in, express A set of node representations for a dynamic, heterogeneous social graph. This refers to a tool for numerically solving ordinary differential equations. This represents the set of node representations in the dynamic new heterogeneous social graph at the initial moment. This represents the target time point, i.e., the specific moment represented by the node that needs to be solved.

[0113] In step 103, all participating times included in the cascaded data set obtained in step 101 are... Normalized to the interval [0, 1], the normalized time is obtained. This method can eliminate the time scale differences between different cascades, making the time features comparable.

[0114] Furthermore, the normalized time interval is then divided into R equal intervals to obtain R+1 key time points. Further, based on the dynamic novel heterogeneous diffusion graph, the node representations of the dynamic novel heterogeneous diffusion graph at different times corresponding to the R+1 key time points are determined. Based on the dynamic novel heterogeneous social graph, the node representations of the dynamic novel heterogeneous social graph at different times corresponding to R+1 key time points are determined. .in, and The "dynamic novel heterogeneous diffusion graph" and "dynamic novel heterogeneous social graph" obtained by solving the graph neural differential equation in the above steps are at time... The node representation aims to discretize continuous time, making it easier to retrieve features corresponding to any point in time.

[0115] Furthermore, arbitrarily select a cascade from the cascaded data set, which is referred to as the first cascade. Normalize the first participation time included in the first cascade to obtain the first time. Determine an interval that matches the first time from R intervals, which is referred to as the first interval. Sequentially determine the start and end time points of the first interval, and then sequentially determine the candidate feature set of the first user corresponding to the start and end time points of the first interval.

[0116] It should be noted that the aforementioned first participation time refers to the original participation time of the user in the first cascade (as shown in formula (1) in step 101). The first time obtained after normalization belongs to the processed participation time, which is consistent with the meaning of "participation time" in the general expression. Here, the candidate feature set of the first user corresponds to the node representations of the dynamic new heterogeneous social graph and the dynamic new heterogeneous diffusion graph at different times for the first user at the start time of the first interval; and the node representations of the dynamic new heterogeneous social graph and the dynamic new heterogeneous diffusion graph at different times for the first user at the end time of the first interval. Specifically, the candidate feature set of the first user is as follows:

[0117] (18)

[0118] in, Indicates user A set of candidate features for multiple scenarios within the given time interval. Indicates the first in the cascade One user, This represents users in a dynamic, heterogeneous social graph. The representation of the starting time of the interval; This represents users in a dynamic, heterogeneous social graph. The representation at the end of the interval; Represents users in a dynamic new heterogeneous diffusion graph The representation of the starting time of the interval; Represents users in a dynamic new heterogeneous diffusion graph The representation at the end of the interval; Indicates user The index of the interval containing the participation time. This represents the hyperparameter, indicating the number of equal divisions the normalized time interval [0,1] is into. This represents the start time of the r-th interval. This represents the end time of the r-th interval.

[0119] It should be noted that the candidate feature set is determined in this step using a cascade as an example. Here, the selected cascade is referred to as the first cascade. However, in actual applications, all cascades included in the cascade data set in step 101 need to be processed. Therefore, when explaining other steps, the term "first" will no longer be added before the term "first" in step 103.

[0120] Because the influence weights of features at different time points within a given interval on user behavior vary (e.g., the end time point may be closer to the user's actual participation time than the start time point), an attention mechanism is needed to automatically learn and fuse these weights. Specifically, for the candidate feature set... Each feature in By calculating their importance weights and then weighted summation, the user's score is obtained. The final dynamic node representation The specific formula is as follows:

[0121] (19)

[0122] (19-1)

[0123] in, Indicates user The final dynamic node representation is given by The weighted sum of the elements in the equation is obtained. Indicates user Attention weights within a specific time interval Indicates user exist Given any representation in the set, Indicates user The set of representations of the start and end times of a time interval. This represents the natural exponential function, used to amplify the differences in importance between different representations. This is represented as a temperature coefficient and is used to control the sharpness of attention weighting. The smaller the value, the more concentrated the weight distribution. This represents the trainable weight matrix in the attention mechanism. express Any node in the set is represented.

[0124] exist Figure 2 In the process, the time interval [0,1] is divided into four segments. The normalized time for user A participating in the cascade is 0.4, therefore its corresponding r=2 in this cascade. Subsequently, the start and end embedding representations of user nodes under different roles are retrieved, and the corresponding time points are marked with pentagrams of different colors. Finally, the final dynamic node representation of the user is obtained through a fusion operation.

[0125] Furthermore, a static new heterogeneous relationship graph is obtained based on the static user-cascaded bipartite graph, the static information diffusion graph, and the static social relationship graph, as shown in formula (8).

[0126] In practical applications, static new heterogeneous relationship graphs contain three types of edges: user-cascade participation relationship, user-user propagation adjacency relationship, and user-user social relationship, which can comprehensively capture the multi-dimensional structural associations of users in the "social + propagation" scenario.

[0127] The set of static node representations of users in a static novel heterogeneous relation graph is obtained through GCN, and the set of static node representations of users in a static information diffusion hypergraph is obtained through HGCN.

[0128] Furthermore, by aggregating the set of node representations of static social attributes through a 2-layer GCN, a set of static node representations of users in the static new heterogeneous relationship graph is obtained, as follows:

[0129] (20-1)

[0130] in, The static node representation set represents a new heterogeneous relation graph. This represents the operation of a graph convolutional network. Static social attribute nodes represent sets. This represents a static new heterogeneous relationship diagram.

[0131] In practical applications, the first-layer GCN captures the local neighbor relationships of nodes, while the second-layer GCN further aggregates structural information from a greater distance, achieving the fusion of "local-global" structural features.

[0132] Furthermore, the set of static node representations of the user in the static information diffusion hypergraph is obtained through HGCN (Hypergraph Convolutional Network). Specifically, the set of static node representations of the user in the static information diffusion hypergraph is as follows:

[0133] (20-2)

[0134] in, The set of nodes representing a static information diffusion hypergraph. Represents a static information diffusion hypergraph The node degree matrix, Represents a static information diffusion hypergraph The node-hyperedge incidence matrix, The hyperedge degree matrix represents the hypergraph of static information diffusion. This represents a node representation that focuses on the attributes of a static social relationship graph. This represents the trainable weight matrix of HGCN.

[0135] The two static nodes mentioned above represent sets. , Reflecting users' long-term structural attributes and dynamic characteristics It reflects the user's real-time status at a specific time, and combining the two can yield a more comprehensive node representation.

[0136] In step 104, the user's final dynamic node representation and the user's static node representation are fused through an attention-gated fusion mechanism to obtain the user's comprehensive node representation.

[0137] Specifically, based on the set of static node representations of users in the static new heterogeneous relationship graph, the static node representations of users in the static new heterogeneous relationship graph and the final dynamic node representations of users in the dynamic new heterogeneous social graph and the dynamic new heterogeneous diffusion graph are obtained. The importance weights of the user's static node representation in the static novel heterogeneous relationship graph and the user's final dynamic node representation in the dynamic novel heterogeneous social graph and the dynamic novel heterogeneous diffusion graph are calculated through an attention-gated fusion mechanism. The weighted sum is then used to obtain the user's comprehensive node representation, specifically:

[0138] (twenty one)

[0139] (twenty two)

[0140] in, Indicates merged users The static node representation of the static novel heterogeneous relationship graph, the final dynamic node representation fused from the dynamic novel heterogeneous social graph and the dynamic novel heterogeneous diffusion graph. This represents the node to be evaluated. Indicates user In the static new heterogeneous relationship diagram Static node representation in , Indicates user In the dynamic new heterogeneous diffusion diagram and dynamic new heterogeneous social graphs The final dynamic node representation of the search and fusion process. This represents the attention weights for static / dynamic node representations. The superscript represents the trainable parameters of attention fusion. This corresponds to Static-Dynamic Fusion. This represents the trainable weight matrix for static-dynamic fusion attention. Indicates cascading The Middle One user, This represents the natural exponential function, used to amplify the differences in importance between different representations. This represents the sum of the exponential scores of the user's static node representation in the static neo-heterogeneous relationship graph and the user's final dynamic node representation in the dynamic neo-heterogeneous social graph and the dynamic neo-heterogeneous diffusion graph, achieving weight normalization. General The static node representation of users within the set in the static new heterogeneous relationship graph and the final dynamic node representation of users fused in the dynamic new heterogeneous social graph and the dynamic new heterogeneous diffusion graph.

[0141] The comprehensive node representation of a user determined by the above steps is the fusion of the static node representation of each user in the static new heterogeneous relationship graph and the user's final dynamic node representation, which includes the user's long-term structural relationships and real-time propagation behavior.

[0142] In this embodiment of the invention, the user's time information includes two types: a discrete form of propagation order and a continuous form of propagation time interval. The propagation order represents the propagation sequence number in the user cascade. The propagation time interval represents the time difference between a user's participation and that of the previous user.

[0143] In this embodiment of the invention, discrete order and continuous time interval are transformed into computable feature vectors through an embedding mechanism, namely discrete order embedding that reflects the user's positional attributes in the propagation sequence and propagation time interval embedding that captures the time rhythm of propagation.

[0144] Specifically, for cascading The participation time series of each user The discrete order embedding and the propagation time interval λ are expressed by the following formulas:

[0145] (23-1)

[0146] (23-2)

[0147] in, Indicates cascading The Middle Each user propagation order embedding vector This represents a multilayer perceptron. Indicates the cascaded c-th One-hot vector for each user Indicates cascading The Middle The propagation time interval embedding vector for each user Indicates cascading The Middle The corrected normalized time for each user is calculated as follows:

[0148] (23-3)

[0149] in, , Indicates the first The user and the first The time difference of participation of each user The maximum threshold representing the propagation time interval. The minimum threshold representing the propagation time interval. This represents the discretized granularity of the control propagation time interval after normalization, and its value is a positive integer. This represents the floor function, which converts continuous normalized time differences into discrete integer labels.

[0150] It should be noted that the comprehensive node representation of the user obtained in the above steps only integrates the static node representation and the dynamic node representation. However, it determines the user's temporal association information in the propagation sequence. In this embodiment of the invention, the temporal association information is the key to determining whether the user will participate in subsequent propagation. Therefore, it is necessary to concatenate the user's comprehensive node representation with the aforementioned determined discrete order embedding and continuous temporal embedding to form the user's complete features as shown below:

[0151] (24-1)

[0152] (24-2)

[0153] in, Indicates cascading The Middle Complete characteristics of a user Indicates cascading The Middle A comprehensive node representation for each user Indicates cascading The basic feature set, Each element integrates the user's comprehensive node representation, discrete-order embedding, and continuous-time embedding. Indicates cascading The length of the propagation.

[0154] Furthermore, the user's complete features are linked to the user's static node representation in the static information diffusion hypergraph. By concatenating the features, we obtain the following feature representation of the static information diffusion hypergraph enhancement:

[0155] (25-1)

[0156] (25-2)

[0157] in, Indicates cascading The Middle Static information diffusion hypergraph enhanced feature representation for each user Indicates cascading The Middle Complete characteristics of a user Indicates cascading The Middle A user is represented by a static node in the static information diffusion hypergraph. Indicates cascading The static information diffusion hypergraph enhancement feature set.

[0158] It should be noted that the set of static node representations of users in the static information diffusion hypergraph obtained according to formula (20-2) in step 103 is... Then, the set of static node representations of the user in the static information diffusion hypergraph can be used. Obtain the static node representation of the user in the static information diffusion hypergraph.

[0159] Furthermore, the cascaded features are aggregated using a multi-head attention mechanism, as follows:

[0160] (26-1)

[0161] (26-2)

[0162] (26-3)

[0163] in, This represents the result of attention calculation. Represents a query, key, and value matrix. express Activation function express Dimensions , , Indicates the first Learnable parameters of head attention Indicates the first The output feature vector of head attention Indicates the first The query matrix for head attention. Indicates the first The key matrix of head attention, Indicates the first The value matrix of head attention, The trainable output weight matrix representing multi-head attention. This indicates a cascade resulting from multi-head attention fusion. Overall characteristics Indicates the first The output feature vector of head attention Indicates the number of attention heads. Represents a matrix mask.

[0164] Furthermore, the multi-head attention output cascades the overall features, which, after layer normalization and residual network optimization, yield the final cascaded features:

[0165] (26-4)

[0166] (26-5)

[0167] in, This indicates a cascade resulting from multi-head attention fusion. The first layer of normalized overall features, This indicates a cascade resulting from multi-head attention fusion. The final characteristic, Presentation layer normalization operation.

[0168] The multilayer perceptron is used to predict the likelihood of each user participating in the propagation in the future, specifically as follows:

[0169] (27)

[0170] in, This represents the trainable weight matrix of the prediction layer in a multilayer perceptron. This represents the trainable bias parameters of the prediction layer in a multilayer perceptron. Indicates cascading The Middle The order is number The probability of a user. Indicates cascading The final feature set is obtained through multi-head attention fusion.

[0171] In this embodiment of the invention, the model parameters are further optimized using a cross-entropy loss function during the training phase, specifically as follows:

[0172] (28)

[0173] in, To label the real dissemination data, 1 represents user It is the first Participating in the cascade , This represents all the trainable parameters mentioned above, and the model parameters are trained through gradient backpropagation and a parameter optimizer.

[0174] Table 1 shows the scale of the dataset used in this embodiment:

[0175] Table 1 Test Dataset Size

[0176]

[0177] Table 2 shows the comparison of the Hits@10 metric between the (DIIDP, Dynamic Interaction-aware Information Diffusion Prediction) method provided by this embodiment of the invention and other comparative methods on all datasets. Bold text indicates the highest accuracy, and underlined text indicates the second highest accuracy. The experimental settings are as follows: all cascaded data were randomly divided into training set (80%), validation set (10%), and test set (10%). The maximum cascade length on all datasets was set to 200. The model was trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 16 for the training set. The dimension of user embedding was set to 64, and the dimension of temporal embedding was set to 8. The ordinary differential equation solver used the Euler method (a first-order numerical method) with a step size of 0.05. The number of attention heads was set to 10.

[0178] Table 2. Accuracy of the Hits@10 metric across all datasets

[0179]

[0180] Table 3 shows the comparison of the method (DIIDP) provided by this embodiment of the invention with other comparative methods on all datasets in terms of MAP@10. Bold text indicates the highest accuracy, and underlined text indicates the second highest accuracy. The experimental settings are as follows: all cascaded data were randomly divided into training set (80%), validation set (10%), and test set (10%). The maximum cascade length on all datasets was set to 200. The model was trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 16 for the training set. The dimension of user embeddings was set to 64, and the dimension of temporal embeddings was set to 8. The ordinary differential equation solver used the Euler method (a first-order numerical method) with a step size of 0.05. The number of attention heads was set to 10.

[0181] Table 3 shows the accuracy of the MAP@10 metric across all datasets.

[0182]

[0183] Table 4 shows the ablation experiments of the method provided in this embodiment on the Christianity dataset and the Android dataset. The experimental settings are as follows: all cascaded data are randomly divided into training set (80%), validation set (10%), and test set (10%). The maximum cascade length on all datasets is set to 200. The model is trained using the Adam optimizer with a learning rate of 0.001 and a batch size of 16 for the training set. The dimension of user embedding is set to 64, and the dimension of temporal embedding is set to 8. The ordinary differential equation solver uses the Euler method (a first-order numerical method) with a step size of 0.05. The number of attention heads is set to 10.

[0184] Table 4 Ablation experiments of the method of this invention patent

[0185]

[0186] Method DIIDP-DO (Dynamic Interaction-aware Information Diffusion Prediction-Diffusion ODEs, removing the graph differential equation representation module of dynamic new heterogeneous diffusion graphs) removes the graph differential equation representation module on information diffusion networks. Method DIIDP-SO (Dynamic Interaction-aware Information Diffusion Prediction-Social ODEs, removing the graph differential equation representation module of dynamic new heterogeneous social graphs) removes the graph differential equation module on social networks. Method DIIDP-O (Dynamic Interaction-aware Information Diffusion Prediction-ODEs, simultaneously removing the graph differential equation representation module of dynamic new heterogeneous social graphs and dynamic new heterogeneous diffusion graphs) removes the graph differential equation module on both diffusion networks and social networks. Method DIIDP-H (Dynamic Interaction-aware Information Diffusion Prediction-Hypergraph, removing the diffusion hypergraph representation module) removes the HGCN module. Method DIIDP-E (Dynamic Interaction-aware Information Diffusion Prediction-Enhanced) removes the graph differential equation representation module of HGCN. The removal of the static new heterogeneous graph representation module (removing the GCN module on the enhanced graph) is shown in Table 4. As can be seen from Table 4, the complete model DIIDP achieved the best performance in all metrics (Hits@10 and MAP@10) on both datasets (Christianity and Android), indicating that the synergistic effect of each graph representation module can significantly improve the overall model performance. Removing the graph differential equation module on the diffusion network (DIIDP-DO) significantly reduced the model performance, indicating that the information from the diffusion network makes a significant contribution to the prediction results. Removing the graph differential equation module on the social network (DIIDP-SO) also led to a performance decrease, especially on the Android dataset where MAP@10 was close to that of the complete model, but Hits@10 decreased, indicating that social network modeling is also indispensable and complementary to the diffusion network. When both the diffusion and social network graph differential equation modules were removed simultaneously (DIIDP-O), the model performance further decreased, demonstrating the enhancing effect of combining the two on information modeling. Furthermore, removing the super GCN module (DIIDP-H) and the GCN module on the augmented graph (DIIDP-E) also leads to a decrease in model performance, indicating that the hypergraph structure and augmented graph structure help capture richer node relationships and structural information, thereby improving the model's generalization ability.In summary, each graph structure module has a positive effect on model performance, and the absence of any module will lead to a decline in performance. This verifies the effectiveness and necessity of the multi-graph fusion modeling strategy, among which the graph differential equation module of diffusion networks and social networks is particularly critical.

[0187] Figure 3A and Figure 3B The chart shows the average training time of different models per epoch on the Android and Douban datasets. The blue bars correspond to the left Y-axis, representing the average training time per epoch; the red dashed lines correspond to the right Y-axis, representing the key performance indicators of the algorithm. Models are sorted from left to right according to performance. Overall, although the training time of DIIDP provided in this embodiment of the invention is slightly increased, it achieves competitive performance within an acceptable training time, demonstrating a good balance between efficiency and effectiveness.

[0188] Based on the same inventive concept, this invention provides an information diffusion prediction device based on dynamic interaction relationship perception. Since the principle of this device in solving the technical problem is similar to that of an information diffusion prediction method based on dynamic interaction relationship perception, the implementation of this device can refer to the implementation of the method, and the repeated parts will not be described again.

[0189] like Figure 4 As shown, the device includes a construction unit 401, a first determination unit 402, a second determination unit 403, and a prediction unit 404.

[0190] Construction unit 401 is used to obtain cascaded data sets and static social relationship graphs, and to construct static user-cascaded bipartite graphs, static information diffusion graphs, static information diffusion hypergraphs, and new heterogeneous graphs.

[0191] The first determining unit 402 is used to obtain the initial node representation of the new heterogeneous graph through the graph convolutional network GCN, construct the graph neural differential equation of the evolution of the new heterogeneous nodes over time, and determine the edge weights and node representations of the new heterogeneous graph at different times.

[0192] The second determining unit 403 is used to normalize the participation time and divide it equally to obtain multiple intervals and key time points, and determine the node representation of the new heterogeneous graph corresponding to each key time point based on the node representation of the new heterogeneous graph; normalize the first participation time included in the first concatenation to obtain the first time, and determine the final dynamic node representation of the first user based on the interval and the first time.

[0193] The prediction unit 404 is used to fuse the user's final dynamic node representation and the user's static node representation through an attention-gated fusion mechanism to obtain the user's comprehensive node representation; the user's comprehensive node representation is fused through multi-head attention to obtain cascaded final features, and the cascaded final features are used to predict the probability of each user participating in propagation in the future through a multilayer perceptron.

[0194] It should be understood that the units included in the above-described information diffusion prediction device based on dynamic interaction relationship perception are merely a logical division based on the functions implemented by the device. In practical applications, the above-described units can be superimposed or split. Furthermore, the functions implemented by the information diffusion prediction device based on dynamic interaction relationship perception provided in this embodiment correspond one-to-one with the information diffusion prediction method based on dynamic interaction relationship perception provided in the above-described embodiment. The more detailed processing flow implemented by this device has been described in detail in the first embodiment of the method described above, and will not be described in detail here.

[0195] Another embodiment of the present invention provides a computer device, the computer device including: a processor and a scene database; the scene database is used to store computer program code, the computer program code including computer instructions; when the processor executes the computer instructions, the electronic device executes each step of the information diffusion prediction method based on dynamic interaction relationship perception shown in the above method embodiment.

[0196] Another embodiment of the present invention provides a computer-readable storage medium storing computer instructions that, when executed on a computer device, cause the computer device to perform the various steps of the information diffusion prediction method based on dynamic interaction relationship awareness as shown in the above method embodiment.

[0197] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

[0198] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An information diffusion prediction method based on dynamic interaction relationship perception, characterized in that, include: Obtain the cascaded dataset and static social relationship graph, and construct a static user-cascaded bipartite graph, a static information diffusion graph, a static information diffusion hypergraph, and a new heterogeneous graph; The initial node representation of the new heterogeneous graph is obtained through a graph convolutional network (GCN). The graph neural differential equations for the evolution of the new heterogeneous nodes over time are constructed to determine the edge weights of the new heterogeneous dynamic graph and the node representations of the new heterogeneous graph at different times. The participation time is normalized and evenly divided to obtain multiple intervals and key time points. Based on the node representation of the new heterogeneous graph, the node representation of the new heterogeneous graph corresponding to each key time point is determined. The first participation time of the normalized first cascade is obtained as the first time, and the final dynamic node representation of the first user is determined based on the interval and the first time. The user's final dynamic node representation and the user's static node representation are fused through an attention-gated fusion mechanism to obtain the user's comprehensive node representation; The user's integrated node representation is obtained by multi-head attention fusion to obtain the cascaded final feature, and the cascaded final feature is used to predict the probability of each user participating in the propagation in the future through a multilayer perceptron.

2. The method as described in claim 1, characterized in that, The novel heterogeneous graph includes a dynamic novel heterogeneous diffusion graph, the initial node representation of the novel heterogeneous graph includes the initial node representation of the dynamic novel heterogeneous diffusion graph, and the dynamic novel heterogeneous diffusion graph corresponds to a dynamically edge-coupled graph neural differential equation. The process involves obtaining the initial node representation of the new heterogeneous graph through a graph convolutional network (GCN), constructing a dynamic edge-coupled graph neural differential equation for the evolution of the new heterogeneous nodes over time, and determining the edge weights and node representations of the new heterogeneous graph at different time points. Specifically, this includes: The initial representation set is obtained through gating functions and dynamic preference nodes. From GCN and the dynamic new heterogeneous diffusion graph, we obtain the initial node representation of the dynamic new heterogeneous diffusion graph: The node update rules are as follows: We construct the dynamic edge-coupled graph neural differential equations for the time-varying node evolution of a novel heterogeneous diffusion graph, and determine the edge weights of the nodes in the graph and the node representations at different time points. in, This represents the set of nodes in the dynamic new heterogeneous diffusion graph at the initial moment. Indicates GCN operation, This represents the gate function for dynamic diffusion relationships. This represents the initial set of dynamic preference nodes. This represents a dynamic new heterogeneous diffusion diagram. Represents a node The representation after GCN update. Represents a non-linear activation function. Let G represent any node in the graph structure G. Represents a node In graph structure The set of neighboring nodes in the middle, Representing graph structure Middle The weight, This represents the general trainable weight matrix of GCN. Representing neighboring nodes The original representation, express Time node representation set Time derivative, The weight parameters represent the balance between the two evolution functions. express The set of node representations for a dynamic, new heterogeneous diffusion graph at any given time. express Dynamic new heterogeneous diffusion diagram at time t, This represents the evolution function associated with the dynamic graph structure. This represents the evolution function related to the node's own characteristics. express Edge of the dynamic new heterogeneous diffusion graph at any time The weight, Represents the normalized exponential function, The nonlinear transformation function representing the node embedding features. Indicates that the node and nodes exist Representation of time and splicing operation, Representing neighboring nodes exist The novel heterogeneous dynamic diffusion representation at time, Represents a node exist The novel heterogeneous dynamic diffusion representation at time, Represents a node The time when it first appeared in the entire information dissemination environment, Indicates filtering The time and the node pairs that have appeared before it. express The set of node representations for a dynamic, new heterogeneous diffusion graph at any given time. This refers to a tool for numerically solving ordinary differential equations. Indicates the target time point.

3. The method as described in claim 1, characterized in that, The novel heterogeneous graph includes a dynamic novel heterogeneous social graph, the initial node representation of the novel heterogeneous graph includes the initial node representation of the dynamic novel heterogeneous social graph, and the dynamic novel heterogeneous social graph corresponds to a highly scalable graph neural differential equation. The process involves obtaining the initial node representation of the new heterogeneous graph through a graph convolutional network (GCN), constructing a highly scalable graph neural differential equation for the evolution of the new heterogeneous nodes over time, and determining the edge weights and node representations of the new heterogeneous graph at different time points. Specifically, this includes: The initial representation set is obtained through gating functions and dynamic preference nodes. Using GCN and dynamic novel heterogeneous social graphs, we obtain the initial node representation of the dynamic novel heterogeneous social graph: The node update rules are as follows: We construct highly scalable graph neural differential equations for the temporal evolution of nodes in dynamic novel heterogeneous social graphs, and determine the edge weights of nodes in the dynamic novel heterogeneous social graphs and the node representations of the dynamic novel heterogeneous social graphs at different time points: in, This represents the set of node representations of the dynamic new heterogeneous social graph at the initial moment. This represents the operation of a graph convolutional network. The gate function represents a dynamic, novel, heterogeneous social graph. This represents the initial set of dynamic preference nodes. This represents a dynamic, new, heterogeneous social graph. Represents a node The representation after GCN update. Represents a non-linear activation function. Let G represent any node in the graph structure G. Represents a node In graph structure The set of neighboring nodes in the middle, Representing graph structure Middle The weight, This represents the general trainable weight matrix of GCN. Representing neighboring nodes The original representation, express User nodes at any time Representation in dynamic neo-heterogeneous social graphs The time derivative of represents its characteristic instantaneous rate of change. This indicates the total number of users and cascading nodes. express Time Node and The normalized weights of the social representation cosine similarity. express Nodes in a constantly evolving and heterogeneous social graph and Static structure weights, Represents a node exist Representation of dynamic, heterogeneous social graphs in real time. Representing neighboring nodes exist Representation of dynamic, heterogeneous social graphs in real time. express The edge of a dynamic and heterogeneous social graph The weight, Represents a node The time when it first appeared in the entire information dissemination environment, Indicates filtering The time and the node pairs that have appeared before it. express A set of node representations for a dynamic, heterogeneous social graph. This refers to a numerical solution tool for ordinary differential equations. This represents the set of node representations of the dynamic new heterogeneous social graph at the initial moment. Indicates the target time point.

4. The method as described in claim 1, characterized in that, The normalized participation time is divided equally to obtain multiple intervals and key time points. Based on the node representation of the new heterogeneous graph, the node representation of the new heterogeneous graph corresponding to each key time point is determined. The normalized first cascade includes the first participation time to obtain the first time. Based on the interval and the first time, the final dynamic node representation of the first user is determined, specifically including: The participation time included in the cascade is normalized to the interval [0,1], and the normalized time interval is divided into R intervals and R+1 key time points. Based on the dynamic new heterogeneous diffusion graph, the node representations of the dynamic new heterogeneous diffusion graph at different times corresponding to R+1 key time points are determined; based on the dynamic new heterogeneous social graph, the node representations of the dynamic new heterogeneous social graph at different times corresponding to R+1 key time points are determined. The first participation time included in the first cascade selected from the cascaded data set is normalized to obtain the first time. The first interval matching the first time is determined from the R intervals. The candidate feature set of the first user corresponding to the start time point and the end time point of the first interval and the final dynamic node representation of the first user are determined in sequence. The candidate feature set for the first user is shown below: The final dynamic node representation of the first user is shown below: in, Indicates user The set of subsequent features from multiple scenarios within the given time interval. Indicates the first in the cascade One user, This represents users in a dynamic, heterogeneous social graph. The representation of the starting time of the interval; This represents users in a dynamic, heterogeneous social graph. The representation at the end of the interval; Represents users in a dynamic new heterogeneous diffusion graph The representation of the starting time of the interval; Represents users in a dynamic new heterogeneous diffusion graph The representation at the end of the interval; Indicates user The index of the interval containing the participation time. Indicates hyperparameters, This represents the start time of the r-th interval. This represents the end time of the r-th interval. Indicates user The final dynamic node representation, Indicates user Attention weights within a specific time interval Indicates user exist Given any representation in the set, This represents the natural exponential function. This is represented as a temperature coefficient and is used to control the sharpness of attention weighting. This represents the trainable weight matrix in the attention mechanism. express Any node in the set represents, It is an abbreviation for MultiScenario, indicating that the feature set includes features from two types of scenarios: dynamic new heterogeneous social graphs and dynamic new heterogeneous diffusion graphs.

5. The method as described in claim 1, characterized in that, Before fusing the user's final dynamic node representation and the user's static node representation through the attention-gated fusion mechanism to obtain the user's comprehensive node representation, the process also includes: The static new heterogeneous relation graph included in the new heterogeneous graph is obtained based on the static user-cascaded bipartite graph, static information diffusion graph, and static social relation graph. Based on the node representation set of static social attributes, GCN, and hypergraph convolutional network HGCN, the static node representation set of users in the static new heterogeneous relation graph and the static node representation set of users in the static information diffusion hypergraph are obtained respectively. Based on the set of static node representations of users in the static new heterogeneous relationship graph, we obtain the static node representations of users in the static new heterogeneous relationship graph, as well as the final dynamic node representations in the dynamic new heterogeneous social graph and the dynamic new heterogeneous diffusion graph.

6. The method as described in claim 1, characterized in that, Before obtaining the initial node representation of the new heterogeneous graph through a graph convolutional network (GCN), the method further includes: The user and cascading nodes are initialized based on a gating mechanism, resulting in a set of static social attribute node representations. and the initial representation set of dynamic preference nodes The cascaded data set consists of multiple cascades, each including users, participation time, and information propagation order. The gating mechanism is as follows: in, This represents a set of node representations that, after being processed by a gating function, focus on a specific attribute. This represents a general abstract representation of all gated functions. The trainable weight matrix representing the gating mechanism. This represents the trainable bias parameters of the gating mechanism. This represents the Hadamard product operation on matrices. Represents the normalized exponential function, This represents the set of initial nodes that contain all users and cascades.

7. The method as described in claim 1, characterized in that, The user's comprehensive node representation is obtained by multi-head attention-focused fusion to obtain the cascaded final features. The cascaded final features are used by a multilayer perceptron to predict the probability of each user participating in propagation in the future, specifically including: The propagation order and propagation time interval are converted into discrete order embeddings reflecting propagation location attributes and time interval embeddings capturing the propagation time rhythm, respectively, through mapping. The comprehensive node representation of the user, the discrete order embedding, and the time interval embedding are concatenated to obtain the user's complete features. The complete features of the user and the static information diffusion hypergraph enhanced feature representation are fused through multi-head attention to obtain the cascaded overall features and the cascaded final features. The cascaded final features are used to predict the probability of each user participating in propagation in the future through a multilayer perceptron.

8. An information diffusion prediction device based on dynamic interaction relationship perception, characterized in that, include: Building blocks are used to acquire cascaded datasets and static social relationship graphs. Construct a static user-cascaded bipartite graph Static information diffusion diagram Static information diffusion hypergraph And new heterogeneous graphs; The first determining unit is used to obtain the initial node representation of the new heterogeneous graph through the graph convolutional network (GCN), construct the graph neural differential equation of the evolution of the new heterogeneous nodes over time, and determine the edge weights of the new heterogeneous dynamic graph and the node representation of the new heterogeneous graph at different times. The second determining unit is used to normalize the participation time and divide it equally to obtain multiple intervals and key time points, and to determine the node representation of the new heterogeneous graph corresponding to each key time point based on the node representation of the new heterogeneous graph. The first participation time of the normalized first cascade is obtained as the first time, and the final dynamic node representation of the first user is determined based on the interval and the first time. The prediction unit is used to fuse the user's final dynamic node representation and the user's static node representation through an attention-gated fusion mechanism to obtain the user's comprehensive node representation; The user's integrated node representation is obtained by multi-head attention fusion to obtain the cascaded final feature, and the cascaded final feature is used to predict the probability of each user participating in the propagation in the future through a multilayer perceptron.

9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the information diffusion prediction method based on dynamic interaction relationship awareness as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, causes the processor to perform the information diffusion prediction method based on dynamic interaction relationship awareness as described in any one of claims 1-7.

Images