Social network event propagation prediction method based on circle effect
By establishing a composite relationship structure in social networks, combining the circle effect from time sequence and structural perspectives, event propagation prediction is carried out, and the problem of inaccurate prediction caused by the circle effect is solved in the existing methods, and more accurate event propagation is achieved Analyze and forecast.
Patent Information
- Application Number
- CN202410996778.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-12-27
AI Technical Summary
The existing social network event propagation prediction methods do not consider the circle effect, resulting in inaccurate prediction results.
By establishing a composite relationship social network structure, using graph neural network to learn user node embedding representation, and combining the circle effect of timing perspective and structural perspective, integrating the characteristics and structural proximity of different stages, using cross attention network and bias attention The network performs adaptive fusion to generate event propagation representations based on circle effect, and predicts the probability distribution of future participants of the event.
It enhances the accuracy and interpretability of event propagation prediction, can more effectively understand the underlying mechanisms of information propagation in complex social systems, and provides a more accurate assessment of event propagation patterns.
Smart Images
Figure CN119204280A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of social network event propagation prediction methods, and in particular to a social network event propagation prediction method based on circle effect. Background Art
[0002] With the rapid development of online social media, event propagation and information acquisition are becoming increasingly convenient and efficient. Predicting and analyzing event propagation is a crucial task in social networks, aiming to identify potential users who may participate in event propagation. Studying and analyzing event propagation is crucial for understanding how information, behavior, or phenomena spread within social networks. It also holds significant value for practical applications such as understanding public opinion evolution, combating fake news, and online marketing.
[0003] Social networks are inherently dynamic and complex systems, and their event propagation is influenced by a variety of factors beyond simple node connections. In social networks, events may propagate differently over time due to changes in user behavior, attention spans, or external events that influence information dissemination. Social networks exhibit structural clustering, where nodes tend to cluster into tightly knit circles with distinct interaction patterns. Therefore, the circle effect under temporal dynamics is a significant factor influencing the propagation of social network events, and capturing the impact of local user interactions on event propagation is crucial. However, existing social network event propagation prediction methods fail to consider the circle effect, resulting in inaccurate prediction results. Summary of the Invention
[0004] The present invention provides a social network event propagation prediction method based on circle effect, so as to solve the problem that the existing social network event propagation prediction method does not take circle effect into consideration, resulting in inaccurate prediction results.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is:
[0006] A method for predicting the spread of social network events based on the circle effect includes the following steps:
[0007] Step 1: Based on the user nodes and the relationships between user nodes in the social network, a composite relational social network structure G is established;
[0008] Step 2: Obtain the embedding representation H of all user nodes through graph neural network learning based on the composite relational social network structure described in step 1. u ;
[0009] Step 3: Considering the circle effect from the temporal perspective, participants are divided into temporal stages according to their participation time. By integrating the characteristics of related stages, a stage change representation S is obtained that reflects the changes in the event propagation stage. k, and the cross attention network based on the phase change representation S through the temporal perspective k and participants indicated that H k Learn and get the event propagation representation H from the temporal perspective t ;
[0010] Furthermore, considering the circle effect from the structural perspective, the structural proximity of the participants on the social network is calculated to divide the structural relationship, and the structural perspective circle relationship matrix ψ reflecting the structural proximity between the participants of the event propagation is obtained. The biased attention network from the structural perspective is used to construct the circle relationship matrix ψ based on the structural perspective and the participant representation H k Learn and get the event propagation representation H from the structural perspective s ;
[0011] Step 4: Express the event propagation from the temporal perspective obtained in step 3 as H t , event propagation representation from a structural perspective H s Perform adaptive fusion to obtain the event propagation representation H based on the circle effect;
[0012] Step 5: Map the event propagation representation H based on the circle effect to the user feature space, thereby predicting the probability distribution of future participants in the event, which is the event propagation prediction result.
[0013] In the further step 1, there are three types of relationships between user nodes, namely social association relationship, communication influence relationship, and behavior preference relationship.
[0014] Furthermore, a trained prediction model that takes into account the circle effect is used to execute steps 2 to 4; the prediction model includes a composite relational graph neural network based on LightGCN, as well as a cross-attention network from a temporal perspective, a biased attention network from a structural perspective, and an adaptive fusion layer based on a gating mechanism; wherein the composite relational graph neural network is used to execute step 2, the cross-attention network from a temporal perspective and the biased attention network from a structural perspective are used to execute step 3, and the adaptive fusion layer based on a gating network is used to execute step 4.
[0015] Furthermore, the loss function during the training of the prediction model adopts a supervised cross loss function.
[0016] The present invention provides a method for predicting the spread of social network events based on circle effects. Incorporating circle effects into the prediction model of event propagation can enhance the interpretability of the model and maintain consistency with the real social network. It can realize effective and accurate predictive analysis of event propagation on social networks, and then more accurately evaluate and understand the pattern of event propagation, which helps to enhance the understanding of the potential mechanisms driving information propagation in complex social systems.
[0017] Compared with the prior art, the present invention has the following advantages:
[0018] 1. The present invention proposes to characterize the multiple complex relationships between users through a complex relationship social network, and comprehensively uses user relationships such as social associations, communication influence and behavioral preferences, thereby solving the singleness and limitations of user relationship descriptions in existing methods.
[0019] 2. The present invention proposes a lightweight composite relational social network learning method. First, the multiple relationships of users in the social network are lightweight pre-fused, and user representation learning is implemented based on LightGCN. This method can effectively learn the multiple heterogeneous semantics between various relationships between users, avoiding the setting of too many learning modules for different relationships, and thus providing efficient support for the analysis and prediction of subsequent event propagation.
[0020] 3. The present invention proposes an event propagation prediction method based on circle effects, which characterizes and analyzes the influence of circle effects on event propagation participants from a temporal perspective and a structural perspective respectively, and integrates circle effects from different perspectives to achieve a more comprehensive event propagation prediction. This method can help enhance our understanding of the potential mechanisms that drive information propagation in complex social systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 This is an overall framework diagram of the event propagation prediction method according to an embodiment of the present invention;
[0022] Figure 2 4 is a flowchart of an event propagation prediction method according to an embodiment of the present invention. DETAILED DESCRIPTION
[0023] The present invention will be further described below with reference to the accompanying drawings and examples.
[0024] like Figure 1 、 Figure 2 As shown, this embodiment discloses a method for predicting the spread of social network events based on the circle effect, including the following steps:
[0025] Step 1: For a social network, consider users as nodes and the three types of relationships between user nodes—social associations, communication influence, and behavioral preferences—as edges. Based on the user nodes and the relationships between them, a composite relational social network structure G = (V, E, R) is established. Here, V represents the set of user nodes in G; E represents the set of user node relationships; and R represents the set of user relationship types.
[0026] In this embodiment, a user node pair formed by any two user nodes may have one or more of three types of relationships: social association, communication influence, and behavior preference relationship.
[0027] Regarding social association relationships, this embodiment obtains social association relationships between user nodes based on user attention relationships on the social network platform.
[0028] Regarding the propagation influence relationship, this embodiment performs random walks for each user node in the user attention and attention relationship network and records the stay probability. It is considered that there is a propagation influence relationship between the K neighbors with the highest stay probability, thereby obtaining the propagation influence relationship between user nodes.
[0029] Regarding the behavior preference relationship, this embodiment counts the frequency of users appearing together in the historical event propagation, and considers that pairs of user nodes that often appear together have similar behavior preferences, thereby obtaining the behavior preference relationship between user nodes.
[0030] Therefore, constructing a composite relational social network G = (V, E, R) based on these three different types of relationships can more comprehensively and fully reflect the relationships between users in event propagation.
[0031] Step 2: Obtain the embedding representation H of all user nodes through graph neural network learning based on the composite relational social network structure described in step 1. u , the process is as follows:
[0032] (2.1) Composite Relationships In a social network G, multiple different relationships may exist simultaneously between user nodes. These relationships have different roles and properties in event propagation. To maintain the complexity and heterogeneity of user node relationships, this embodiment uses learnable weights to aggregate composite relationships, and can obtain a composite relationship adjacency matrix A, as shown in the following formula:
[0033]
[0034] Among them: |R| represents the number of types of user relationships, r represents a specific relationship, A r Represents the adjacency matrix between user nodes regarding a specific relationship r; β r is the learnable weight of the correspondence.
[0035] (2.2) The graph neural network LightGCN is used to learn the user node embedding based on the composite relational adjacency matrix A. During learning, the embedding matrix H of all users in the composite relational social network G is extracted by stacking L layers of graph encoders in LightGCN. u , as shown in the following formula:
[0036] X(L) =AX (L-1) =A L X (0)
[0037]
[0038] Where: X (0) is the initial embedding of the user node; X (L) is the output of the L-th layer graph encoder; H u is the embedding representation of all users; A is the composite relationship adjacency matrix.
[0039] Step 3: In order to learn the user interactions and influences in event propagation and thus realize the prediction analysis of event propagation, the embedding representation H of all users calculated in step 2 is used. u , and considering the circle effect from the temporal perspective and the structural perspective, we get the event propagation representation from the temporal perspective and the structural perspective. The process is as follows:
[0040] (3.1) The circle effect from the time series perspective is mainly reflected in the stage dynamics of event propagation. This embodiment divides the time propagation into stages by calculating the time intervals of participants, and ultimately obtains multiple stage windows. Based on the gating mechanism, the users within the windows are aggregated to obtain the stage change representation S that reflects the stage changes of event propagation. k , k is the number of participating combinations, and d is the dimension of the embedding representation.
[0041] Furthermore, according to step 2, the representations of all users are obtained, and the representations of participants in a specific event H are retrieved. k For subsequent learning,
[0042] Finally, the temporal perspective crisscross attention network (TCAN) is used to represent S based on the phase change. k and participants indicated that H k The relationship between participants in event propagation is learned to obtain the event propagation representation H from the temporal perspective t , as shown in the following formula:
[0043] Q i =S k W i Q ,K i =H k W i K ,V i =H k W i V
[0044]
[0045] in: They are respectively the learnable weight matrix; the masking matrix Set the upper triangle to -∞ to avoid information leakage; Q i represents the query value of the i-th attention head; K i represents the key value of the i-th attention head; V i Represents the value of the i-th attention head; TCAN i () represents the cross attention head of the i-th temporal perspective; softmax() represents the softmax activation function.
[0046] After that, based on the multi-head attention mechanism and feedforward neural network, we can finally get the event propagation representation H from the temporal perspective t .
[0047] (3.2) The structural perspective of the circle effect is primarily reflected in the local influence of event propagation, that is, users who are closer to each other on the social network have a greater mutual influence. To this end, the structural proximity between event propagation participants can be calculated to obtain the structural perspective circle relationship matrix ψ, which reflects the structural proximity between event propagation participants. For example, using the shortest path distance as the structural proximity, the calculation process of the structural perspective circle relationship matrix is as follows:
[0048] ψ i,j =1 / log(e+dis(v i ,v j ))
[0049] Among them, ψ i,j is the value of the i-th row and j-th column of the relationship matrix ψ, e≈2.718 is the natural base, log() is the natural logarithm function, dis(v i ,v j ) is the participant v i and v j The shortest path distance between them.
[0050] Then, the biased attention network of the structural perspective is used based on the structural perspective circle relationship matrix ψ and the participant representation H k The relationship between participants in event propagation is learned to obtain the event propagation representation H from a structural perspective s , as shown in the following formula:
[0051] Q i =H k W i Q ,K i =H k W iK ,V i =H k W i V
[0052]
[0053] in: It is the circle relationship matrix from the structural perspective; It is a representation of the participants; They are respectively the learnable weight matrix; the masking matrix Set the upper triangle to -∞ to avoid information leakage; SSAN i () represents the biased attention head of the i-th structural perspective; softmax() represents the softmax activation function. The multi-head attention mechanism and feedforward neural network (FFN) are applied here to obtain the event propagation representation H from the temporal perspective. s .
[0054] Step 4: Express the event propagation from the temporal perspective obtained in step 3 as H t , event propagation representation from a structural perspective H s Perform adaptive fusion based on the gating mechanism to obtain event propagation representation based on circle effect As shown in the following formula:
[0055]
[0056] H=α⊙H t +(1-α)⊙H s
[0057] in: represents the learnable weight vector for feature fusion; is a learnable weight matrix.
[0058] Step 5: Use linear mapping to convert the event propagation representation H based on the circle effect into the user feature space, thereby predicting the probability distribution of future participants in the event, which is the event propagation prediction result, as shown in the following formula:
[0059]
[0060] in: It is a mapping matrix used to map the event propagation representation H to the user feature space; Represents the final probability distribution of future participants of the event.
[0061] In this embodiment, steps 2 to 4 are performed by using a trained prediction model that takes into account the circle effect. The prediction model includes a composite graph neural network based on LightGCN, a cross-attention network from a temporal perspective, a biased attention network from a structural perspective, and an adaptive fusion layer based on a gating mechanism; the composite graph neural network is used to perform step 2, the cross-attention network from a temporal perspective and the biased attention network from a structural perspective are used to perform step 3, and the adaptive fusion layer based on a gating mechanism is used to perform step 4.
[0062] When training the prediction model, the user social relationships and event propagation records of real social media platforms are used as the training data set.
[0063] The loss function used in predictive model training is a supervised cross-loss function. During multiple rounds of training, all learnable parameters in steps 2-5 are optimized based on the calculation results of the cross-loss function. The cross-loss function is shown in the following formula:
[0064]
[0065] in: is the cross loss function.
[0066] Θ is all the learnable parameters of the model, including the model parameters involved in steps 2-5.
[0067] Represents the prediction result of the jth participant for event i.
[0068] y i,j represents the true value of the jth participant of event i. If user u i is the jth participant in event propagation, then y i,j =1, otherwise y i,j =0.
[0069] V represents all users on the social network.
[0070] C k A sequence representing all participants in a particular event.
[0071] The preferred embodiments of the present invention are described in detail above with reference to the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner unless there is any contradiction. Such combinations should also be regarded as the contents disclosed in this disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
[0072] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the scope of the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by those skilled in the art should fall within the scope of protection of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.
Claims
1. A method for predicting the spread of social network events based on circle effect, characterized in that: The following steps are involved: Step 1: Establish a composite relational social network structure G based on the user nodes and the relationships between the user nodes in the social network; Step 2: Use graph neural network to learn the composite relational social network structure described in step 1 to obtain the embedding representation H of all user nodes. u ; Step 3: Considering the circle effect from the temporal perspective, participants are divided into temporal stages according to their participation time. By integrating the characteristics of relevant stages, a stage change representation S is obtained that reflects the change in the event propagation stage. k , and represents S based on the stage change through the cross attention network from the temporal perspective k and participants indicated that H k Learn and get the event propagation representation H from the time series perspective t ; In addition, considering the circle effect from the structural perspective, the structural proximity of the participants on the social network is calculated to divide the structural relationship, and the structural perspective circle relationship matrix ψ reflecting the structural proximity between the participants of the event propagation is obtained. The biased attention network from the structural perspective is used to construct the circle relationship matrix ψ based on the structural perspective and the participant representation H k Learn and obtain the event propagation representation H from a structural perspective s ; Step 4: Represent the event propagation from the temporal perspective obtained in step 3 as H t , Event propagation representation from a structural perspective s Perform adaptive fusion to obtain the event propagation representation H based on the circle effect; Step 5: Map the event propagation representation H based on the circle effect to the user feature space, thereby predicting the probability distribution of future participants in the event, which is the event propagation prediction result.
2. A method for predicting the spread of social network events based on circle effect according to claim 1, characterized in that: In step 1, there are three types of relationships between user nodes, namely social association relationships, communication influence relationships, and behavioral preference relationships.
3. The method for predicting the spread of social network events based on circle effect according to claim 1, characterized in that: Use the trained prediction model that takes circle effects into consideration to execute steps 2 to 4; the prediction model includes a composite relationship graph neural network based on LightGCN, as well as a cross-attention network from a temporal perspective, a biased attention network from a structural perspective, and an adaptive fusion layer based on a gating mechanism; the composite relationship graph neural network is used to execute step 2, the cross-attention network from a temporal perspective and the biased attention network from a structural perspective are used to execute step 3, and the adaptive fusion layer based on a gating network is used to execute step 4.
4. The method for predicting the spread of social network events based on circle effect according to claim 3, characterized in that: The loss function during the training of the prediction model adopts a supervised cross loss function.