A dynamic approach to minimize the impact of misinformation in social networks

Abstract

The application provides a dynamic method for minimizing the influence of false information in a social network, comprising the following steps: 1, constructing a user social network graph and an information set according to a real data set; 2, determining factors influencing the attention value of a user, mainly including the interest degree of the user to the information, the preference similarity between users, the relationship strength between users and the relative influence; 3, formulating an exploration strategy of an information sending path; 4, dynamically matching effective information for different paths; according to the attention space condition of a target user, selecting information to be propagated, and adding a boredom factor to dynamically adjust the information matched by the path in each round; 5, changing the attention of the target user through multiple rounds of information propagation, and finally removing the false information from the attention space of the target user. The application is tested under an IC independent cascade diffusion model, and the effectiveness is proved.

Description

Technical Field

[0001] This invention belongs to the field of social networks and intelligent agent systems, and specifically relates to a dynamic method for minimizing the impact of erroneous information in social networks. Background Technology

[0002] In recent years, with the rapid development of internet technology and social media, the scale and popularity of social networks such as Twitter, Facebook, Weibo, and WeChat have increased significantly. On these social networks, a vast amount of information can spread rapidly and widely among users every day. These characteristics of social networks have attracted more researchers to focus on influence diffusion models within social networks. Influence diffusion models can accelerate and expand the speed and reach of information dissemination by identifying and selecting the most influential users on social networks. However, because users can freely publish information on social networks, many users publish a large amount of false information for various purposes. The widespread dissemination of false information can lead to serious public opinion bias and have a negative impact on society. Today, false information on social networks has become a major social problem.

[0003] Early detection and prevention are crucial to minimizing the impact of misinformation on society. Many researchers have suggested identifying key nodes (i.e., users) and connections within social networks and blocking them when necessary to prevent the spread of misinformation. However, as social networks continue to grow, finding and monitoring key nodes becomes extremely time-consuming and resource-intensive. Furthermore, the dynamic changes within social networks can lead to changes in key nodes and connections, making these detection methods unable to respond promptly to evolving networks. Some researchers utilize influential users to prevent other users from spreading misinformation, thus minimizing its impact. However, the number of influential users in social networks is limited, and their distribution significantly influences the detection and prevention of misinformation.

[0004] In recent years, diverting the attention of target users from misinformation has proven to be a practically effective method for quickly stopping its spread. Researchers have proposed an impact minimization model that can divert the attention of target users by sending them information related to other topics of interest. However, their model only focuses on the impact of different information content on the diversion of target users' attention. In real-world social networks, due to the large scale and complexity of the network structure, the information transmission path also has a significant impact on the effectiveness of diverting target users' attention. Summary of the Invention

[0005] To address the aforementioned unresolved key issues, this invention proposes a decentralized method to minimize the impact of misinformation in social networks. First, the method uses intelligent agents to simulate users in the social network and employs an information transmission path exploration mechanism, enabling users who detect target users holding misinformation to explore efficient and effective information transmission paths in a decentralized manner. Furthermore, a dynamic attention shifting mechanism is proposed, allowing users to dynamically update and select appropriate information and its transmission path based on the target user's level of aversion. The effectiveness of this method is explained under different cost constraints.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A dynamic method for minimizing the impact of misinformation in social networks includes the following steps:

[0008] Step 1: First, construct a user social network graph S and an information set msg based on a real dataset. The information in graph S includes the user's unique identifier id and the user's personal influence inf. i And the strength of the relationship r between users i，j .

[0009] Step 2: Define the user's interest level for different types of information. i and the user's attention space mep i This provides a foundation for subsequent analysis;

[0010] Step 3: Identify the factors influencing user attention scores. Based on sociological knowledge, it is determined that user attention scores for different types of information are primarily influenced by the user's own level of interest in that information. i，k And the influence of the information disseminator on the recipient. i，j The influence of two main factors. Effe i，j Pres i，j Relationship strength r i，j and relative influence Infd i，j The impact;

[0011] Step 4: Develop a strategy for exploring information dissemination paths. Based on the relative influence among users, explore information dissemination paths layer by layer, and calculate the final path's influence on the target user and the probability of successful dissemination. This step ensures that the information dissemination path we choose has a greater influence on the user.

[0012] Step 5: Dynamically match effective information for different paths. Based on the target user's attention space, select the information to be disseminated and add a boredom factor B. i，kThis allows for dynamic adjustment of the path matching information in each round to achieve optimal efficiency.

[0013] Step 6: By spreading information through multiple rounds, change the target user's attention and finally remove the erroneous information from their attention space; test under the IC independent cascade diffusion model to obtain the number of information propagation rounds required to remove the erroneous information from the target user's attention space, and evaluate the effectiveness of the method;

[0014] As a preferred embodiment, the steps for constructing the social network S in step 1 are as follows:

[0015] Step 1-1: In a social network, there are a large number of users, and these users have different relationships. Typically, a social network can be represented by an undirected graph, where nodes and edges represent users and their relationships, respectively. A social network S can be defined as follows:

[0016] S = {U, R}

[0017] Where U={u1, u2,...u i , ...} represents the user set, u i It is the i-th user in S; R = {r 1，2 r 1，3 ,...,r i，j , ...} represents a set of relations, r i，j User u i and u j The degree of the relationship between them. i，j The value of r is between 0 and 1, where 0 and 1 represent the farthest and closest relationships between users, respectively. Since S is represented by an undirected graph, r... i，j =r j，i· .

[0018] Steps 1-2: The message set msg consists of multiple messages, one message msg k It can be defined based on its theme and keywords as follows:

[0019] msg k ={topic k}

[0020] Among them topic k ={kw1, kw2, ..., kw n} is the message msg k The theme kw t It is the t-th keyword in the topic.

[0021] Steps 1-3: The social network S is defined as follows:

[0022] u i={id, pre i ,inf i , nu i mep i}

[0023] Where id is user u i The unique identifier, pre i User u i preferences, inf i User u i Personal influence, nu i User u i The set of neighboring users, and mep i User u i The message pool includes user u i All messages received.

[0024] Preferably, the definition information in step 2 includes:

[0025] Step 2-1: User interest level (pre) i It can be described as follows:

[0026] pre i ={kw1, kw2, ..., kw t , ...}

[0027] Among them, kw t It is the t-th keyword describing user i's preferences.

[0028] Step 2-2: User's attention space (mep) i as follows:

[0029] mep i =(msg1, msg2, ..., msg) k , ...}

[0030] mep i It is the message pool for user i, which includes all the messages it receives.

[0031] As a preferred approach, the factors influencing the user attention value in step 3 are calculated as follows:

[0032] Step 3-1: User u i message k Attention is primarily influenced by the user's own level of interest in the information. i，k and information disseminators u j For u i Impact Effe i，j The influence of two main factors.

[0033] Step 3-2: User i's interest in message k (Inte) i，k By using i's preferences pre i and k's topic k The Jaccard similarity between them is derived, and the calculation method is as follows:

[0034]

[0035] Among them, pre i ={kw1, kw2, ..., kw t , ...} is the set of preferred keywords for user i, topic k ={kw1, kw2, ..., kw n Let} be the set of topic keywords of message k, and take the ratio of the number of elements in the intersection of the two sets to the number of elements in the union of the two sets.

[0036] Step 3-3: Relative personal influence of user j on i (Effe) i The calculation method for j is as follows:

[0037] Effe i，j =(r i，j +Pres i，j +Infd i，j ) / 3,

[0038] Where, r i，j It represents the degree of connection between users i and j, obtained through processing information from the dataset. i，j It is the preference similarity between users i and j, infd i，j It represents the relative personal influence of user j on user i. The calculation methods for both are as follows:

[0039]

[0040]

[0041] Among them, pre i and pre j These are the personal preferences of users i and j, respectively, inf j and inf i λ represents the personal influence values ​​of users j and i, respectively, which is the ratio of the number of elements in the intersection to the number of elements in the union of the two sets. λ is a relative coefficient that takes a value between 0 and 1.

[0042] Preferably, the information transmission path selection strategy in step 4 includes:

[0043] Step 4-1: Set the current user d as the error message detector and iterate through its neighboring users. If the target user t is encountered, calculate the impact effect Effe and the probability P of successful path propagation. l And add this information to the current path.

[0044] The probability P of successful information propagation along the path l The calculation formula is as follows:

[0045]

[0046] Step 4-2: If the target user is not encountered and the length of the current path does not exceed the maximum exploration layer (Max), continue exploring the neighboring users of the neighboring users, gradually building the path. If the path length reaches the limit (Max), the current path will be abandoned.

[0047] Step 4-3: After exploring all possible paths, check if there are any paths in the path set that have the same effect. If so, compare the success probabilities of these paths, keep the path with the highest probability, and remove the path with the lower probability.

[0048] Finally, a filtered set of paths is returned, representing the most effective propagation paths from misinformation detectors to target users. This process ensures the efficiency and effectiveness of information dissemination while taking into account the limitations of information dissemination depth.

[0049] As a preferred embodiment, the dynamic information matching strategy described in step 5 is as follows:

[0050] Step 5-1: Select the information set in the target user's attention space as the information to be disseminated to enhance their attention.

[0051] Step 5-2: Traverse the set of paths. For each selected path, calculate the potential change in the target user's attention due to the propagation of information k through this path l. The calculation formula is as follows:

[0052] Δatt i，k，l =pInte i，k B i，k +qEffe i，j +mP l

[0053] Where p, q, and m are the balance coefficients of each factor, with values ​​between 0 and 1, and B i，k This represents the user i's current level of aversion to information k. Its value decreases as the number of times the user receives this information increases; a smaller value indicates higher aversion and less impact on changes in attention. The calculation is as follows:

[0054] B i，k =1-n×numk

[0055] Where num k This represents the number of times the user receives this message, where n is the scaling factor, based on the integer num. k The range of values ​​is reduced to between 0 and 1.

[0056] By calculation, the information that will have the best propagation effect in the current round of sending is matched for each path.

[0057] Step 5-3: Update the target user's attention to the message.

[0058] This process can continue until the target user's attention level to the error message drops below that of all other information in their attention space, at which point the error message is considered to have faded from the user's attention space, and the impact is eliminated.

[0059] Preferably, step 6 involves selecting the target node and its erroneous information within its attention space under the IC independent cascade diffusion model. Based on the target node and its surrounding neighboring users, a source node is identified. Upon detecting contamination of the target node, this source node begins exploring possible paths for propagating correct information and determines the correct information to be propagated. This continues until the target user's attention value for the erroneous information falls below that of all information in the attention space, at which point the user exits the target's attention space, thus eliminating the impact. The fewer rounds of information propagation required to eliminate the impact are determined; fewer rounds demonstrate the effectiveness of the method.

[0060] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0061] The greatest advantage of this invention compared to previous methods lies in its ability to address, to some extent, the problems arising from the variability of network structures and the excessive scale of networks. Both traditional centralized methods that block user nodes or sever connections between users, and previous methods that leverage the information environment to reduce the impact of misinformation, suffer from limitations imposed by network structure and scale, leading to computational complexity and unstable results. Our proposed method, however, is decentralized, with target detection and behavior initiation occurring within local areas. Based on these factors, the contributions of the proposed method can be summarized as follows: 1. The proposed method's information transmission path exploration mechanism enables users to explore efficient and effective information transmission paths for target users in a decentralized manner; 2. The proposed method's dynamic attention transfer mechanism allows users to dynamically update and select appropriate information and its transmission paths based on the target user's level of aversion; 3. Comparative experiments on real social network datasets demonstrate that the proposed method exhibits high efficiency, high performance, and low cost in minimizing the impact of misinformation. Attached Figure Description

[0062] Figure 1This is a flowchart of the method for minimizing the impact of dynamic error information in this invention;

[0063] Figure 2 This is a partial schematic diagram of social network modeling in this invention;

[0064] Figure 3 Here is an example of the results of the path exploration mechanism;

[0065] Figure 4 Here is an example of the results of the path exploration mechanism;

[0066] Figure 5 Examples of changes in user attention space before information matching and propagation;

[0067] Figure 6 This is an example of how user attention space changes after information matching and propagation.

[0068] Figure 7 This is a comparison chart showing the performance of our method with other path selection methods on a real dataset.

[0069] Figure 8 This is a comparison chart showing the performance of our method and a method that does not consider information matching on a real dataset.

[0070] Figure 9 The images show the results at different costs on different datasets; Detailed Implementation

[0071] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0072] like Figure 1 As shown, the technical route of this invention mainly includes six steps: building a social network model based on user information, relationships between users, and information sets in the social network; the agent explores and selects information propagation paths according to the mechanism; dynamically matching messages; information propagation; observing the attention space of the target node after each round of propagation, and continuing the next round of information matching and propagation if the erroneous information is still attracting attention, until the erroneous information fades out of the user's attention space; and obtaining the number of information propagation rounds required to remove the erroneous information.

[0073] This example uses the Twitter, Wiki Vote, and Epinions datasets. The Twitter dataset was collected from survey participants using the Twitter application. It contains 272,861 nodes and 433,439 edges. The Wiki Vote dataset is historical data extracted from all Wikipedia administrator elections and votes since January 3, 2008, containing 7,115 network nodes and 103,689 edges. The Epinions dataset network is a trust network obtained from the consumer review website Epinions.com. Website members can decide whether to "trust" each other, and it contains 75,879 nodes and 508,837 edges. This invention addresses the problem of minimizing the impact of misinformation in social networks. The following sections will introduce this example from four aspects: social network model establishment, exploration of propagation paths and dynamic information matching, and experimental testing.

[0074] 1) Establishing a social network model

[0075] Step 1-1: In a social network, there are a large number of users, and these users have different relationships. Typically, a social network can be represented by an undirected graph, where nodes and edges represent users and their relationships, respectively. A social network S can be defined as follows:

[0076] S = {U, R}

[0077] Where U={u1, u2,...u i , ...} represents the user set, u i It is the i-th user in S; R = {r 1，2 r 1，3 ,...,r i，j , ...} represents a set of relations, r i，j User u i and u j The degree of the relationship between them. i，j The value is between 0 and 1, where 0 and 1 represent the furthest and closest relationships between users, respectively. This value can be indirectly obtained through network connectivity; in the experimental processing, it is taken as the value of user u. i and u j The ratio of the size of the intersection to the size of the union of neighboring nodes. Since S is represented by an undirected graph, r i，j =r j，i· .

[0078] Steps 1-2: The message set msg consists of multiple messages, one message msg k It can be defined based on its theme and keywords as follows:

[0079] msg k={topic k}

[0080] Among them topic k ={kw1, kw2, ..., kw n} is the message msg k The theme kw t It is the t-th keyword in the topic.

[0081] Steps 1-3: User interest level (preview) i It can be described as follows:

[0082] pre i ={kw1, kw2, ..., kw t , ...}

[0083] Among them, kw t It is the t-th keyword describing user i's preferences.

[0084] Steps 1-4: User's attention space (mep) i as follows:

[0085] mep i ={msg1, msg2, ..., msg k , ...}

[0086] mep i It is the message pool for user i, which includes all the messages it receives.

[0087] Steps 1-5: The social network S is defined as follows:

[0088] u i ={id, pre i ,inf i , nu i mep i}

[0089] Where id is user u i The unique identifier, pre i User u i preferences, inf i User u i Self-influence, nu i User u i The set of neighboring users, and mep i User u i The message pool includes user u i All messages received.

[0090] In our embodiment, the initial modeling partial schematic diagram is as follows: Figure 2As shown, the edges between nodes have weights, which represent the degree of connection between nodes and are related to the probability of successful information transmission between nodes in the model. Each node, in addition to its own attributes, also has its own attention space information. The relationships between nodes are represented by an undirected graph.

[0091] 2) Exploration of transmission paths

[0092] Step 2-1: Set the current user d as the error message detector and iterate through its neighboring users. If the target user t is encountered, calculate the impact effect Effe and the probability P1 of successful path propagation, and add this information to the current path.

[0093] In the path, the relative personal influence (Effe) of information disseminator user j on information receiver i. i The calculation method for j is as follows:

[0094] Effe i，j =(r i，j +Pres i，j +Infd i，j ) / 3,

[0095] Where, r i，j The strength of the connection between users i and j is obtained by processing the dataset centrally, pres i，j It is the preference similarity between users i and j, infd i，j It represents the relative personal influence of user j on user i. The calculation methods for both are as follows:

[0096]

[0097]

[0098] Among them, pre i and pre j These are the personal preferences of users i and j, respectively, inf j and inf i These are the personal influence values ​​of users j and i, respectively, where λ is a relative coefficient, set to 1 in the experiment, ensuring that if (inf j -inf i If j is a positive number, then j has a relatively large influence on i.

[0099] The probability P of successful information propagation along the path l The calculation formula is as follows:

[0100]

[0101] Step 2-2: If the target user is not encountered and the length of the current path does not exceed the maximum exploration layer (Max), continue exploring the neighboring users of the neighboring users, gradually building the path. If the path length reaches the limit (Max), the current path will be abandoned.

[0102] Figure 3 This shows an example of the path exploration results between user u2 and u5.

[0103] Figure 4 This includes detailed path information generated during the exploration process, such as the two adjacent users and their edge values ​​for each path, the calculated impact value of the path (Effe), and the probability P of successful information propagation along the path. l .

[0104] Steps 2-3: After exploring all possible paths, check if there are any paths in the path set that have the same impact. If so, compare the success probabilities of these paths, keep the path with the highest probability, and remove the path with the lowest probability.

[0105] Figure 4 Both path1 and path4 propagate information from user u4 to target user u5 and have the same effect, but the success rate of information propagation for path4 is lower than that for path1, so path4 is discarded.

[0106] Finally, a filtered set of paths is returned, representing the most effective propagation paths from misinformation detectors to target users. This process ensures the efficiency and effectiveness of information dissemination while taking into account the limitations of information dissemination depth.

[0107] 3) Dynamic information matching

[0108] Step 6-1: Select the information set in the target user's attention space as the information to be disseminated to enhance their attention.

[0109] like Figure 5 As shown, in user u5's attention space, the error message is msg2. Users can choose messages with lower attention values ​​than the error message, such as msg1 and msg4, to fade the error message out of the user's attention space by increasing the user's attention to them. This is because the user's attention is limited; for example, as shown in the figure, only the first three messages are of interest to the user.

[0110] Step 6-2: Traverse the set of paths. For each selected path, calculate the potential change in attention of the target user due to the propagation of information k through path 1. The calculation formula is as follows:

[0111] Δatt i，k，l =pInte i，k Bi，k +qEffe i，j +mP l

[0112] Where p, q, and m are the balance coefficients of each factor, and after multiple experiments, their values ​​were determined to be 0.5, 0.2, and 0.3, respectively. B i，k This represents the user i's current level of aversion to information k. Its value decreases as the number of times the user receives this information increases; a smaller value indicates higher aversion and less impact on changes in attention. The calculation is as follows:

[0113] B i，k =1-n×num k

[0114] Where num k This represents the number of times the user receives this message, where n is the scaling factor, based on the integer num. k The range of values ​​is reduced to between 0 and 1.

[0115] By calculation, the information that will have the best propagation effect in the current round of sending is matched for each path.

[0116] Step 6-3: Update the target user's attention to the message.

[0117] like Figure 6 As shown, the mechanism calculates the propagation information that best propagates to different paths. After one round of propagation, the user's attention space changes, and the user's attention value for msg1 is greater than that for the error message msg2.

[0118] This process can continue until the target user's attention level to the error message drops below that of all other information in their attention space, at which point the error message is considered to have faded from the user's attention space, and the impact is eliminated.

[0119] 4) Experimental Testing

[0120] We tested our method on a real dataset.

[0121] like Figure 7 As shown, the X-axis represents the number of rounds of information propagation, while the Y-axis represents the position of the erroneous information in the user's attention space. The higher the position, the more attention it receives from the user, and 0 indicates that it has been removed from the user's attention space. The results show that the path exploration and selection mechanism of the present invention is superior to the path selection results based on other indicators.

[0122] Figure 8This demonstrates the superiority of our dynamic information matching mechanism. Similarly, the X-axis represents the number of rounds of information propagation, while the Y-axis represents the position of the erroneous information in the user's attention space. The higher the position, the more attention it receives from the user, and 0 indicates that it has been removed from the user's attention space. We compared two cases: randomly matching messages to the path and matching messages without considering the user's level of boredom. The results show that the dynamic information matching mechanism of this invention has a certain degree of effectiveness improvement.

[0123] Figure 9 The results demonstrate the effectiveness under different cost constraints across multiple datasets. The X-axis represents different costs, with the maximum number of path exploration layers (Max) used as the cost. This is because adding an extra user to the path practically means persuading another user to spread the information, which incurs a real cost. The Y-axis represents the number of information propagation rounds required to remove erroneous information. The results show that the performance tends to stabilize when the cost is 7.

[0124] Based on the above analysis, our method has certain advantages in minimizing the impact of erroneous information.

[0125] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art should understand that certain modifications and variations can be made to the present invention without departing from the spirit and substance of the present invention, but these modifications and variations should be covered within the scope of protection of the present invention.

Claims

1. A dynamic method for minimizing the impact of misinformation in social networks, characterized in that, Includes the following steps: Step 1: First, construct the user social network graph S and the information set msg based on the real dataset; The information in Figure S includes the user's unique identifier id and the user's personal influence info. i And the strength of the relationship r between users i,j ; Step 2: Define the user's interest level for different types of information. i and the user's attention space mep i This provides a foundation for subsequent analysis; Step 3: Identify the factors influencing user attention scores; using sociological knowledge, determine that user attention scores for different types of information are influenced by the user's own level of interest in that information. i,k And the influence of the information disseminator on the recipient. i,j The influence of two factors; Effe i,j Pres i,j Relationship strength r i,j and relative influence Infd i,j The impact; Step 4: Develop an exploration strategy for information dissemination paths; explore information dissemination paths layer by layer based on the relative influence among users, and calculate the influence value of the final path on the target user and the probability value of successful path propagation to ensure that the selected information dissemination path is the path with the greatest influence on the user. Step 5: Dynamically match effective information for different paths; select the information to be disseminated based on the target user's attention space, and add a boredom factor B. i,k To dynamically adjust the information matched in each round of the path; Step 6: Through multiple rounds of information dissemination, change the target user's attention and ultimately remove the erroneous information from their attention space; The dynamic information matching strategy is as follows: Step 6-1: Select the information set in the target user's attention space as the information to be disseminated to enhance their attention; Step 6-2: Traverse the set of paths. For each selected path, calculate the potential change in attention of the target user due to the propagation of information k through this path l. The calculation formula is as follows: ； Where p, q, and m are the balance coefficients of each factor, with values ​​between 0 and 1. B is the probability of successful path propagation; i，k This represents the user i's current level of aversion to information k. Its value decreases as the number of times the user receives this information increases; a smaller value indicates higher aversion and less impact on changes in attention. The calculation is as follows: ； Where num k This represents the number of times the user receives this message, where n is the scaling factor, based on the integer num. k The range of its values ​​is reduced to between 0 and 1; By calculation, the information that will have the best propagation effect in the current round of transmission is matched for each path; Step 6-3: Update the target user's attention to this information; This process continues until the target user's attention level to the error message drops below that of all other information in their attention space, at which point the error message is considered to have faded from the user's attention space, and its impact is eliminated.

2. The method as described in claim 1, characterized in that, The evaluation method under the IC model is as follows: Step 7-1: Select the target node and the error information in its attention space; Step 7-2: Based on the target node and its surrounding users, determine the source node. Once the source node detects that the target node has been contaminated, it begins to explore possible paths for propagating the correct information and determines the correct information to be propagated. Step 7-3: Begin propagating the correct information until the target user's attention value for the incorrect information is lower than that for all information in the attention space, then exit the target's attention space and the impact is eliminated; The fewer rounds of information dissemination required to eliminate the impact, the better the proof method is.

3. The method as described in claim 1, characterized in that, The steps to construct a social network S are as follows: Step 2-1: Social networks are represented using undirected graphs, where nodes and edges represent users and the relationships between them, respectively; the social network S is defined as follows: ； in Represents a set of users, u i It is the i-th user in S; Let r represent a set of relations. i,j The value of is between 0 and 1, where 0 and 1 represent the farthest and closest relationships between users, respectively; since S is represented by an undirected graph, ; Step 2-2: The message set msg consists of multiple messages, one message msg k Based on its theme and keywords, it is defined as follows: ； in It is a message message k The theme, kw n It is the nth keyword in the topic.

4. The method as described in claim 1, characterized in that, User interest pre i And attention space mep i The structure is as follows: Step 3-1: User interest level (pre) i The description is as follows: ； Among them, pkw t It is the t-th keyword describing user i's preferences; Step 3-2: User's attention space (mep) i as follows: ； mep i It is the information pool of user i, which includes all the messages it receives.

5. The method as described in claim 1, characterized in that, The factors influencing user attention scores are calculated as follows: Step 4-1: User u i message k Attention is influenced by the user's own level of interest in the information. i,k and information disseminators u j For u i Impact Effe i,j The influence of two factors; Step 4-2: User i's interest in information k (Inte) i，k By analyzing user i's interest pre i and k's topic k The Jaccard similarity between them is derived, and the calculation method is as follows: ； in, It is the set of user i's preferred keywords. It is the set of the keywords of information k, and is the ratio of the number of elements in the intersection of the two sets to the number of elements in the union of the two sets; Step 4-3: Effe of the relative personal influence of user j on user i i，j The calculation method is as follows: ;where Pres i,j This is the preference similarity between user i and user j, calculated as follows: ； ； Wherein, λ is a relative coefficient, which takes a value between 0 and 1.

6. The method as described in claim 1, characterized in that, The information sending path selection strategy is as follows: Step 5-1: Set the current user d as the error message detector and traverse its neighboring users; if the target user t is encountered, calculate the path impact effect (Effe) and the probability of successful path propagation. And add this information to the current path; Path propagation success probability The calculation formula is as follows: ； Step 5-2: If the target user is not encountered and the length of the current path does not exceed the maximum exploration layer Max, continue to explore the neighboring users of the neighboring users and gradually build the path; if the path length reaches the limit, that is, it exceeds the maximum exploration layer Max, the current path will be abandoned. Step 5-3: After exploring all possible paths, check if there are any paths in the path set that have the same effect; if so, compare the success probabilities of these paths, keep the path with the highest probability, and remove the path with the lower probability. Finally, the filtered set of paths is returned.

Images