An influence maximization method driven by user behavior in an evolving social network

Abstract

This invention discloses a method for maximizing user-behavior-driven influence in evolutionary social networks. The method includes: Step 1, acquiring user behavior data in the social network and establishing an evolutionary social network model that evolves over time; Step 2, calculating the information propagation probability between users based on user behavior logs to obtain the edge weights in the evolutionary social network; Step 3, improving the independent cascading propagation model to establish a user-behavior-driven independent cascading propagation model; Step 4, using an improved backsampling method to query the k most influential nodes in the evolutionary social network at different time points, i.e., the seed node set; Step 5, verifying the effectiveness of the user-behavior-driven independent cascading propagation model. This invention solves the problem that existing technologies do not consider the impact of user behavior on information propagation and cannot accurately query the most influential users in evolutionary social networks.

Description

Technical Field

[0001] This invention relates to the field of social network technology, and in particular to a method for maximizing user behavior-driven influence in evolutionary social networks. Background Technology

[0002] With the development of internet technology, more and more virtual social networks have emerged, such as large social networking sites like Facebook and interpersonal networks formed through mobile communication. Compared with traditional social networks, these newly emerging social networks have strong dynamic characteristics and are called evolving social networks. An evolving social network is a network that changes continuously over time. Throughout the evolution process, the number of users and the relationships between them constantly change. The influence between users does not remain constant but changes with user behavior. Therefore, focusing on evolutionary characteristics is more reasonable for modeling and studying social networks.

[0003] Social influence refers to the phenomenon that a person's emotions, opinions, or behaviors affect others. For example, people share and spread their thoughts, news, and other information on social networks, thereby influencing other users on the network. Social networks, as a medium, play a fundamental role in how individuals in society influence each other and spread information and opinions.

[0004] The Influence Maximization (IM) problem aims to identify k users as seed nodes in a network, maximizing the impact of information on other users through these k users within a specific propagation model (such as independent cascading propagation or linear threshold propagation). Currently, influence maximization has wide-ranging applications, including marketing, personalized recommendations, expert user discovery, and content ranking. Therefore, analyzing social network data to dissect social phenomena and measure user influence has become a major research focus.

[0005] Currently, there are many approximate algorithms for solving the problem of maximizing influence in social networks.

[0006] While the issue of maximizing influence has received widespread attention, existing work primarily focuses on static graphs, where the topology and node relationships of social networks remain unchanged during dissemination. However, in reality, social networks for information dissemination are constantly evolving. According to the China Internet Development Statistics Report released by CNNIC in September 2020, as of June 2020, my country had 931 million instant messaging users, accounting for 99.0% of all internet users, and 888 million online video users, accounting for 94.5% of all internet users. The user base of social networks is continuously growing, and the structure of social networks is constantly evolving. Existing research often overlooks the impact of user behavior on information dissemination. Some information diffusion processes can last for days or even months. During this period, changes in user behavior, such as the addition and departure of new users, and the posting or deletion of information, all affect the dissemination of information.

[0007] Chinese patent CN112446634A discloses "a method and system for detecting nodes that maximize influence in a social network". Its technical solution does not consider the temporal nature of the network when establishing the network structure. When the network structure changes over time, the seed nodes detected by the method will lose their effectiveness. When calculating the expected value of the node's influence, it does not consider the situation that the expected value of influence will change over time.

[0008] Chinese patent CN112380456A discloses a dynamic influence maximization method based on cohesive entropy. However, this method uses overlapping communities to dynamically select seed nodes, without truly considering the characteristics of network structure changes over time or the impact of changes in user behavior on information dissemination. Therefore, the calculated seed nodes are not effective for evolving social networks.

[0009] Therefore, current solutions to the influence maximization problem do not consider the impact of user behavior on information dissemination. However, changes in user behavior alter the probability of information dissemination, thus affecting information propagation within the network. Furthermore, the scale and social relationships of users constantly change during network evolution. This dynamic characteristic is difficult to observe in a static graph, hindering current influence maximization solutions from identifying the most influential users. Therefore, accurately identifying the most influential users on evolving social networks is a technical problem that needs to be solved. Summary of the Invention

[0010] Therefore, the purpose of this invention is to propose a user behavior-driven method for maximizing influence in evolutionary social networks, in order to solve the problem that existing technologies do not consider the impact of user behavior on information dissemination and cannot accurately query the most influential users on evolutionary social networks.

[0011] This invention provides a method for maximizing user behavior-driven influence in evolutionary social networks, comprising:

[0012] Step 1: Obtain user behavior data in social networks and establish an evolutionary social network model that evolves over time. Through this evolutionary social network model, information can be disseminated in the ever-evolving social network.

[0013] Step 2: Calculate the information propagation probability between users based on user behavior logs to obtain the edge weights in the evolutionary social network;

[0014] Step 3: Improve the independent cascading propagation model and establish a user behavior-driven independent cascading propagation model;

[0015] Step 4: Based on the improved backsampling method, query the k most influential nodes in the evolving social network at different time points, i.e., the seed node set;

[0016] Step 5: Design a node similarity algorithm to compare the similarity between the seed node set and the affected nodes, and verify the effectiveness of the user behavior-driven independent cascading propagation model.

[0017] The aforementioned method for maximizing influence driven by user behavior in evolutionary social networks, specifically step 1 includes:

[0018] Obtain user behavior data from social networks and build an evolutionary social network model G that evolves over time. m =(V m E m W m ), where m represents time, V m Let E represent the set of user nodes at time m. m Let W represent the set of edges at time m. m Let m be the set of weights on the edges at time m.

[0019] The above-mentioned method for maximizing influence driven by user behavior in evolutionary social networks, specifically includes step 2:

[0020] Abstract the relational action tuple (u, a, t) from the behavior log. k ), (u, a, t k ) indicates that user u is in t k Action a was executed at time 1, causing A to... u A represents the action performed by user u in the dataset. v A represents the action performed by user v in the dataset. u ∩A vLet represent the actions performed by both users u and v. Analyze the behavior logs of all nodes in the evolutionary social network model, and calculate the duration window τ of a node's influence on its neighbors based on each node's actions with its neighbors. u，v :

[0021]

[0022] Where A represents the set of actions to be performed, t u (a) indicates that user u performed action a at time t. v (a) indicates that user v performed action a at time t;

[0023] Calculate the probability p of information propagation between users using a duration window. u，v :

[0024]

[0025] in, This indicates that Δt = t exists. v -t u Δt represents the time t during which user v performs action a. v The time t when user u performs action a u The time difference, prop, represents the number of identical actions performed by users u and v within the time difference Δt.

[0026] The above-mentioned method for maximizing influence driven by user behavior in evolutionary social networks, specifically includes step 3:

[0027] Step 3.1: Given a time point, calculate the duration window of each user's influence on their neighbors at that time point. Calculate the information propagation probability using the formula for information propagation probability. Whether any node v is activated is determined by the influence value of a neighbor on it. For a given initial set of active nodes, the cascading propagation process is performed as follows:

[0028] Step 3.2: At the start of propagation, given an initial set S of active nodes, the active nodes will remain active during the subsequent cascading process;

[0029] Step 3.3: At time t, the newly activated node u influences its neighboring node v with a success probability of p. u，v If v has multiple neighboring nodes that are all recently activated nodes, then these nodes will attempt to activate node v in any order.

[0030] Step 3.4: If node v is successfully activated, then at time t+1, node v becomes active and will affect its adjacent inactive nodes; otherwise, the state of node v does not change at time t+1.

[0031] Step 3.5: Repeat the above process continuously. When there are no nodes that can be activated, the propagation will terminate.

[0032] The above-mentioned method for maximizing influence driven by user behavior in evolutionary social networks, specifically includes step 4:

[0033] Step 4.1, randomly select a node v∈V, where V represents the set of user nodes;

[0034] Step 4.2: Backsample the set R of nodes that can activate v. For each node u∈V, the probability of it appearing in R is equal to the probability that u can activate v. Let R be the back reachable set of v.

[0035] Step 4.3: Repeat steps 4.1 and 4.2 above to generate multiple reverse reachable sets. Find the node that appears most frequently and add it to the seed node set. Remove the node that appears most frequently from the reverse reachable set. Repeat this process k times and output the seed set with the largest influence, which is of size k.

[0036] The above-mentioned method for maximizing influence driven by user behavior in evolutionary social networks includes, in step 5, the following steps:

[0037] By modeling nodes as vectors based on the actions they perform, the similarity between the vectors of seed nodes and affected nodes is compared to determine whether the affected nodes are truly impacted, thereby evaluating the effectiveness of seed users.

[0038] Compared to existing technologies, the method provided by this invention establishes an evolutionary social network model that evolves over time based on user behavior data acquired in social networks. This model enables information to propagate within the continuously evolving social network. Furthermore, it improves upon this model by employing a backsampling method, enabling a more accurate and effective query for the k most influential users in the network. This avoids biases caused by randomly selecting information propagation probabilities, resulting in a more efficient and accurate calculation of user influence and a more authentic and effective set of users retrieved. This invention considers the impact of user behavior on information propagation, as well as the continuous changes in user scale and social relationships during network evolution, enabling a more accurate retrieval of the most influential users on an evolving social network. Attached Figure Description

[0039] The above and / or additional aspects and advantages of the embodiments of the present invention will become apparent and readily understood from the description of the embodiments in conjunction with the following drawings, wherein:

[0040] Figure 1 This is a schematic diagram illustrating the logical principle of a user behavior-driven influence maximization method in an evolutionary social network according to an embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram of information transmission in an evolutionary network;

[0042] Figure 3 This is an exemplary diagram illustrating the calculation of message propagation probability;

[0043] Figure 4 This is a comparison chart of the influence of the model of this invention on computing nodes with other models under four different datasets;

[0044] Figure 5 This is a schematic diagram illustrating the change over time between seed sets of different sizes and the similarity of affected nodes under the BDIC and IC models. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0046] For ease of subsequent description, we will first briefly explain the relevant concepts of evolutionary social networks and independent cascading propagation models.

[0047] Evolutionary social networks are continuously evolving networks. This invention uses time wheels to evolve forward, with time intervals from 0 to T being the first round, and the m-th round being a time interval from (m-1)T to mT. G is used. m =(V m E m W m Let (u, v) represent the social network at time m. Assume all edges are directed, where (u, v) represents the edge from user (also called node) u to user v. Influence between users can only spread from u to v along the edge (u, v). If (u, v) exists, it means u is a neighbor of v, and at each time m, the most influential set of seed nodes can be identified.

[0048] Independent Cascade (IC) models describe how information propagates in social networks. Each node in a social network has two states: active and inactive. Each node can only transition from an inactive state to an active state and remain active until the propagation ends. Each node can be activated by its neighboring nodes, and whether any node v is activated is determined by the probability p of information propagated to it by a neighbor. uv The decision was made. uvGenerally, it is randomly selected and p uv ∈[0,1). Information begins to propagate from the initial active node and continues until there are no more nodes that can be activated, at which point the propagation terminates.

[0049] Please combine Figure 1 An embodiment of the present invention proposes a method for maximizing user behavior-driven influence in evolutionary social networks, comprising steps 1 to 5:

[0050] Step 1: Obtain user behavior data in social networks and establish an evolutionary social network model that evolves over time. This evolutionary social network model enables information to spread within the continuously evolving social network.

[0051] Step 1 specifically includes:

[0052] Obtain user behavior data from social networks and build an evolutionary social network model G that evolves over time. m =(V m E m W m ), where m represents time, V m Let E represent the set of user nodes at time m. m Let W represent the set of edges at time m. m Let m be the set of weights on the edges at time m.

[0053] Figure 2 This is a schematic diagram of information transmission in an evolutionary network. The arrows represent the direction of message propagation. As time t progresses, the point that receives the message continues to propagate the message through the edges. At the same time, the edges that propagate the message also change continuously until there are no more edges to propagate the message, at which point the process stops.

[0054] Step 2: Calculate the information propagation probability between users based on user behavior logs to obtain the edge weights in the evolutionary social network.

[0055] This approach calculates the probability of information propagation between users based on their behavior logs, rather than simply selecting probabilities randomly. This allows for the determination of edge weights in the evolutionary social network, thereby improving the independent cascading model and establishing a Behavior Driven Independent Cascading (BDIC) propagation model. This model can then be used to query the group of users with the greatest influence in the evolutionary social network.

[0056] Step 2 specifically includes:

[0057] Abstract the relational action tuple (u, a, t) from the behavior log. k ), (u, a, t k ) indicates that user u is in tk Action a was executed at time 1, causing A to... u A represents the action performed by user u in the dataset. v A represents the action performed by user v in the dataset. u ∩A v Let represent the actions performed by both users u and v. Analyze the behavior logs of all nodes in the evolutionary social network model, and calculate the duration window τ of a node's influence on its neighbors based on each node's actions with its neighbors. u，v :

[0058]

[0059] Where A represents the set of actions to be performed, t u (a) indicates that user u performed action a at time t. v (a) indicates that user v performed action a at time t;

[0060] Calculate the probability p of information propagation between users using a duration window. u，v :

[0061]

[0062] in, This indicates that Δt = t exists. v -t u Δt represents the time t during which user v performs action a. v The time t when user u performs action a u The time difference, prop, represents the number of identical actions performed by users u and v within the time difference Δt. The numerator in this formula represents finding the number of actions 'a' (possibly more than one) performed by both users u and v, where the time difference Δt is less than τ. u，v of.

[0063] Figure 3 Here is an exemplary diagram illustrating the calculation of message propagation probability. (a) shows the action logs of three users P, Q, and R, containing three actions a1, a2, and a3; (b), (c), and (d) are the propagation graphs PG for the three actions, where the numbers on the edges represent the time required for the actions to propagate. All edges in the propagation graph PG are directed. A model influence matrix (IM) is established, where IM[i,j] = (p...). i，j , τ i，j For example, IM[P, R] = (1 / 2, 10) means that user P performs two actions, a1 and a2, within a window of time 10, and user R performs an action a1 that is the same as user P within the window. Therefore, the time interval τ P，R When the value is 10, the probability P P，R It is 1 / 2.

[0064] Step 3: Improve the independent cascading propagation model and establish a user behavior-driven independent cascading propagation model.

[0065] Step 3 specifically includes:

[0066] Step 3.1: Given a time point, calculate the duration window of each user's influence on their neighbors at that time point. Calculate the information propagation probability using the formula for information propagation probability. Whether any node v is activated is determined by the influence value of a neighbor on it. For a given initial set of active nodes, the cascading propagation process is performed as follows:

[0067] Step 3.2: At the start of propagation, given an initial set S of active nodes, the active nodes will remain active during the subsequent cascading process;

[0068] Step 3.3: At time t, the newly activated node u influences its neighboring node v with a success probability of p. u，v If v has multiple neighboring nodes that are all recently activated nodes, then these nodes will attempt to activate node v in any order.

[0069] Step 3.4: If node v is successfully activated, then at time t+1, node v becomes active and will affect its adjacent inactive nodes; otherwise, the state of node v does not change at time t+1.

[0070] Step 3.5: Repeat the above process continuously. When there are no nodes that can be activated, the propagation will terminate.

[0071] Step 4: Based on the improved backsampling method, query the k most influential nodes in the evolving social network at different time points, i.e., the seed node set.

[0072] Step 4 specifically includes:

[0073] Step 4.1, randomly select a node v∈V, where V represents the set of user nodes;

[0074] Step 4.2: Backsample the set R of nodes that can activate v. For each node u∈V, the probability of it appearing in R is equal to the probability that u can activate v. Let R be the back reachable set of v.

[0075] Step 4.3: Repeat steps 4.1 and 4.2 above to generate multiple (e.g., 1000) reverse reachable sets. Find the node that appears most frequently and add it to the seed node set. Remove the node that appears most frequently from the reverse reachable set. Repeat this process k times and output the seed set with the largest influence, which is of size k.

[0076] Step 5: Design a node similarity algorithm to compare the similarity between the seed node set and the affected nodes, and verify the effectiveness of the user behavior-driven independent cascading propagation model.

[0077] Step 5 specifically includes:

[0078] By modeling nodes as vectors based on their actions, the similarity between seed node vectors and affected node vectors is compared to determine whether the affected nodes are truly impacted, thereby evaluating the effectiveness of seed users. It should be noted that this embodiment uses the cosine of two user behavior vectors as a validity metric when verifying the effectiveness of the seed set. In practice, Chebyshev distance or Jaccard similarity coefficient can also be used to measure the effectiveness of the seed set.

[0079] The above method establishes a user behavior-driven information propagation model (BDIC) in evolutionary social networks. By analyzing user behavior, it calculates the probability of information propagation between users, basing the calculation on facts and thus better reflecting reality than other information propagation models. Based on the BDIC model, an improved backsampling method is used to query the k most influential nodes in the network. This method also designs a validity algorithm to calculate the validity of the seed set.

[0080] Figure 4 The graph shows a comparison of the influence of the model of this invention on computing nodes with other models under four different datasets. The four datasets are the Digg social network dataset, the Sina Weibo social network dataset, dataset data3 which is generated based on the Digg dataset using a random distribution, and dataset data4 which is generated based on the Weibo dataset using a normal distribution. Specific information about the datasets is shown in Table 1.

[0081] Table 1

[0082]

[0083] contrast Figure 4 As shown in (a), (b), (c), and (d), the seed node influence derived by the BDIC model is greater than that of the seed node under the IC model, regardless of whether the dataset is real or synthetic. This indicates that the BDIC model is more effective than the IC model in finding the most influential nodes, and the resulting node set has a wider influence.

[0084] Figure 5This diagram illustrates the changes in similarity between seed sets of different sizes and affected nodes over time under the BDIC and IC models. Based on user behavior analysis of each dataset, this embodiment sets the window sizes for datasets 1, 2, 3, and 4 to 4.6 days, 14.3 days, 8.1 days, and 8.1 days, respectively. Furthermore, for ease of observation, [the diagram shows...]. Figure 5 The horizontal axis time is set to an integer. Figure 5 The comparison shows that, over time, in dataset 1, the average similarity between seed nodes and their neighbors under BDIC is 15% higher than that under IC. In dataset 3, the similarity between seed nodes under BDIC is 20% higher than that under IC. In datasets 2 and 4, the average similarity between seed nodes under BDIC is 10% higher than that under IC.

[0085] Therefore, regardless of whether the seed set size is 5 or 15, the similarity of the seed node set obtained by BDIC is on average 10%-20% higher than that of the seed node set under the IC model. This indicates that the seed node set found by the BDIC model proposed in this invention is more realistic than that of the IC model. This result conforms to the actual message propagation behavior: when a user is influenced by other nodes, the user's behavior tends to be similar to that of the nodes that influenced him, proving that the seed nodes found by the BDIC model method are more in line with the actual scenario.

[0086] In summary, based on the user behavior-driven influence maximization method in the evolutionary social network provided in this embodiment, an evolutionary social network model that evolves over time is established based on the acquired user behavior data in the social network. This model enables information to spread within the continuously evolving social network. Furthermore, an improved backsampling method is implemented, allowing for a more accurate and effective query of the k most influential users in the network. This avoids biases caused by randomly selecting information propagation probabilities, making the calculation of user influence more efficient and accurate, and resulting in a more realistic and effective set of users. This invention considers the impact of user behavior on information propagation, as well as the continuous changes in user scale and social relationships during network evolution, enabling a more accurate query of the most influential users on the evolutionary social network.

[0087] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0088] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for maximizing user-behavior-driven influence in evolutionary social networks, characterized in that, include: Step 1: Obtain user behavior data in social networks and establish an evolutionary social network model that evolves over time. Through this evolutionary social network model, information can be disseminated in the ever-evolving social network. Step 2: Calculate the information propagation probability between users based on user behavior logs to obtain the edge weights in the evolutionary social network; Step 3: Improve the independent cascading propagation model and establish a user behavior-driven independent cascading propagation model; Step 4: Based on the improved backsampling method, query the k most influential nodes in the evolving social network at different time points, i.e., the seed node set; Step 5: Design a node similarity algorithm to compare the similarity between the seed node set and the affected nodes, and verify the effectiveness of the user behavior-driven independent cascading propagation model. Step 1 specifically includes: Obtain user behavior data from social networks and build an evolutionary social network model G that evolves over time. m =(V m E m W m ), where m represents time, V m Let E represent the set of user nodes at time m. m Let W represent the set of edges at time m. m Let m be the set of edge weights at time m; Step 2 specifically includes: Abstract the relational action tuple (u, a, t) from the behavior log. k ), (u,a,t k ) indicates that user u is in t k Action a was executed at time 1, causing A to... u A represents the action performed by user u in the dataset. v A represents the action performed by user v in the dataset. u ∩A v Let represent the actions performed by both users u and v. Analyze the behavior logs of all nodes in the evolutionary social network model, and calculate the duration window τ of a node's influence on its neighbors based on each node's actions with its neighbors. u,v : Where A represents the set of actions to be performed, t u (a) indicates that user u performed action a at time t. v (a) indicates that user v performed action a at time t; Calculate the probability p of information propagation between users using a duration window. u,v : in, This indicates that Δt = t exists. v -t u Δt represents the time t during which user v performs action a. v The time t when user u performs action a u The time difference, prop represents the number of identical actions performed by users u and v within the time difference Δt; Step 3 specifically includes: Step 3.1: Given a time point, calculate the duration window of each user's influence on their neighbors at that time point. Calculate the information propagation probability using the formula for information propagation probability. Whether any node v is activated is determined by the influence value of a neighbor on it. For a given initial set of active nodes, the cascading propagation process is performed as follows: Step 3.2: At the start of propagation, given an initial set S of active nodes, the active nodes will remain active during the subsequent cascading process; Step 3.3: At time t, the newly activated node u influences its neighboring node v with a success probability of p. u,v If v has multiple neighboring nodes that are all recently activated nodes, then these nodes will attempt to activate node v in any order. Step 3.4: If node v is successfully activated, then at time t+1, node v becomes active and will affect its adjacent inactive nodes; otherwise, the state of node v does not change at time t+1. Step 3.5: Repeat the above process continuously. When no node is activated, the propagation will terminate. Step 4 specifically includes: Step 4.1, randomly select a node v∈V, where V represents the set of user nodes; Step 4.2: Backsample the set R of nodes that can activate v. For each node u∈V, the probability of it appearing in R is equal to the probability that u can activate v. Let R be the back reachable set of v. Step 4.3: Repeat steps 4.1 and 4.2 above to generate multiple reverse reachable sets. Find the node that appears most frequently and add it to the seed node set. Remove the node that appears most frequently from the reverse reachable set. Repeat this process k times and output the seed set with the largest influence, which is of size k.

2. The method for maximizing user behavior-driven influence in evolutionary social networks according to claim 1, characterized in that, Step 5 specifically includes: By modeling nodes as vectors based on the actions they perform, the similarity between the vectors of seed nodes and affected nodes is compared to determine whether the affected nodes are truly impacted, thereby evaluating the effectiveness of seed users.

Images