Social media key user identification system based on information non-uniform propagation characteristics

Abstract

The invention discloses a social media key user identification system based on information non-uniform propagation characteristics, and relates to the technical field of key user identification, and the system comprises a dynamic module which is used for calling an information propagation model through a weighted directed graph of N time windows, simulating information propagation, obtaining initial influence scores of a user in the N time windows, and obtaining initial influence scores of the user in the N time windows; and a cross-layer module which constructs a multi-layer network, quantifies the influence of the user in different layers in the multi-layer network based on the dynamic influence score, obtains a cross-layer influence score, and performs weighted average on the initial influence score of the user in N time windows to obtain a dynamic influence score. The recognition module is used for calculating comprehensive scores according to the initial influence scores, the dynamic influence scores and the cross-layer influence scores of the users, and the comprehensive scores are arranged in a descending order to recognize key users; through the key user identification method, the social platform can automatically identify key users really having transmission force and guiding force.

Description

Technical Field

[0001] This invention relates to the field of key user identification technology, and in particular to a key user identification system for social media based on the characteristics of non-uniform information propagation. Background Technology

[0002] With the rapid development of information, social media platforms have become an indispensable part of people's daily lives. Social media platforms have not only changed the way people communicate, but also greatly influenced the dissemination patterns of information. Against this backdrop, the study of user behavior and information dissemination characteristics on social media has gradually become an important research field. Previous social network analysis mainly focused on basic indicators such as node degree and centrality, aiming to identify users with high influence in the network. In recent years, researchers have begun to try to simulate the diffusion process of information in social networks by constructing social network graph models and combining them with different information dissemination models, so as to more accurately assess the influence of users. Therefore, how to accurately identify key users has become an important research direction.

[0003] Nevertheless, existing key user identification methods still have room for improvement. First, most key user identification methods fail to fully consider the impact of time factors on changes in user influence, resulting in key users that are often identified lacking dynamism and timeliness. Second, when constructing social network graphs, they usually adopt a single type of interaction behavior, ignoring the differences between different interaction behaviors and their different contributions to the effectiveness of information dissemination. Summary of the Invention

[0004] This invention provides a social media key user identification system based on the characteristics of non-uniform information propagation, which solves the problems of lack of dynamism and timeliness in identifying key users.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides a social media key user identification system based on the characteristics of non-uniform information propagation, comprising: The data acquisition module collects user interaction data from social media platforms and preprocesses it to obtain a user interaction dataset. The module is divided into N time windows based on the user interaction dataset, and a weighted directed graph is constructed for the N time windows; The dynamic module calls the information propagation model through a weighted directed graph of N time windows to simulate information propagation, obtain the user's initial influence score in N time windows, and then calculates the weighted average of the user's initial influence scores in N time windows to obtain the dynamic influence score. The cross-layer module constructs a multi-layer network and, based on dynamic influence scoring, quantifies the user's influence at different levels within the multi-layer network to obtain a cross-layer influence score. The identification module calculates a comprehensive score based on a user's initial influence score, dynamic influence score, and cross-level influence score. The comprehensive scores are sorted in descending order to identify key users.

[0006] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the step of collecting user interaction data from social media platforms and preprocessing it to obtain a user interaction dataset specifically involves... We obtain user interaction data through the official API interfaces provided by social media platforms, and then clean and standardize it. The cleaned and standardized user interaction data is labeled with time tags, source user ID, target user ID, and interaction behavior type to obtain a user interaction dataset. The types of interactive behaviors include liking, commenting, and sharing.

[0007] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, wherein: the division of the user interaction dataset into N time windows specifically involves, Set the total analysis period for the user interaction dataset, and select the time window length based on the analysis requirements and the characteristics of the user interaction dataset; Based on the length of the time window, the total analysis period is evenly divided into N time windows.

[0008] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, wherein: the construction of a weighted directed graph using N time windows is specifically as follows: For N time windows, extract the user interaction dataset within the time window, create a node for each user appearing in the interaction dataset, and when there is interaction between users, create unweighted directed edges between the corresponding user nodes to connect them, thus obtaining the social network graph. For each pair of directly connected user nodes, the interaction frequency and intensity are statistically analyzed to update the weight values ​​of the directed edges in the social network graph, resulting in a weighted directed graph with N time windows.

[0009] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the step of calling the information propagation model through a weighted directed graph of N time windows to simulate information propagation and obtain the user's preliminary influence score in the N time windows is as follows: Based on a weighted directed graph with N time windows, the user node state within the weighted directed graph is set to the susceptible state. At the same time, a range of highly active user nodes is defined, and W highly active user nodes are randomly selected as the infected state before being imported into the information propagation model to simulate information propagation and obtain the user's preliminary influence score in each time window.

[0010] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the step of weighted averaging of the user's initial influence scores over N time windows to obtain a dynamic influence score specifically involves: Set the attenuation coefficient and calculate the weights of N time windows based on the attenuation coefficient; The user's dynamic influence score is obtained by weighting and averaging the weights of N time windows and the initial influence score.

[0011] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the construction of the multi-layer network specifically involves: The user interaction dataset is divided into a set of likes, a set of comments, and a set of shares. Weighted directed graphs are then built for each of the sets of likes, comments, and shares to obtain the like layer, comment layer, and share layer, respectively. The like, comment, and share layers are saved as independent layers, forming a unified multi-layered network.

[0012] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the method further includes: quantifying a user's influence at different levels within a multi-layered network based on dynamic influence scoring to obtain a cross-layer influence score, specifically... The user's dynamic influence score is imported into the like layer, comment layer, and share layer, and the user's influence score at each layer is calculated by combining the weight coefficients of the like layer, comment layer, and share layer. The influence scores at each level are aggregated to obtain the user's cross-level influence score.

[0013] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the step of calculating a comprehensive score based on a user's initial influence score, dynamic influence score, and cross-level influence score, and then sorting the comprehensive scores in descending order to identify key users, specifically involves... Create a table, fill in the user ID and the corresponding initial influence score, dynamic influence score and cross-level influence score, and set the importance weight; For each user, a weighted average is calculated in a table based on the set importance weights to obtain the user's overall score.

[0014] As a preferred embodiment of the social media key user identification system based on the non-uniform information propagation characteristics described in this invention, the comprehensive scores are arranged in descending order to identify key users. Specifically, Sort all user records from highest to lowest based on their overall score, and set a threshold for the overall score to measure key users. Users whose overall score is higher than the threshold are considered key users.

[0015] The social media key user identification system provided by this invention, based on the non-uniform information propagation characteristics, divides time into multiple continuous windows and constructs a weighted directed graph. This not only captures the complex interaction patterns and intensity changes among users but also fully considers the impact of time factors on user influence, thus more realistically reflecting the dynamics of information propagation in social networks. Furthermore, it uses a method of simulating information propagation to quantify the user's initial influence score, revealing the user's propagation role and potential influence range in different time periods, enhancing the timeliness and accuracy of the identification results. Finally, it constructs a multi-layer network and quantifies the user's influence at different behavioral levels such as likes, comments, and reposts, providing a comprehensive and accurate user influence assessment framework, making the identified key users more representative and practical. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 A flowchart for identifying key users on social media based on the characteristics of non-uniform information propagation.

[0018] Figure 2 A schematic diagram for constructing a weighted directed graph.

[0019] Figure 3 A flowchart for collecting user interaction data.

[0020] Figure 4 A flowchart for calculating user dynamic influence scores. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] Reference Figures 1-4 This is one embodiment of the present invention, which provides a social media key user identification system based on the characteristics of non-uniform information propagation, including the following steps: Data Acquisition Module: Collects user interaction data from social media platforms and preprocesses it to obtain a user interaction dataset.

[0023] Specifically, the steps include the following: Log in to the target social media platform, such as Twitter, Weibo, or Facebook. Based on the social media platform's API documentation, determine the API endpoint URLs for different interaction types. For example, likes, comments, and reposts each correspond to different API paths. Generate a Bearer Token using the OAuth 2.0 protocol and add it to the HTTP request header to ensure the security and legitimacy of the request. Define common query parameters, such as time ranges and pagination markers, to facilitate batch retrieval of user interaction data.

[0024] Select the target user set U={u1,u2,...,u...} for which interaction data needs to be collected. n For each user u∈U, the API interfaces for three interaction types—likes, comments, and shares—are called sequentially to retrieve user interaction data for one period (e.g., 30 days). If the returned user interaction data exceeds the maximum limit for a single request, a pagination flag is used to continue requesting the remaining user interaction data until all user interaction data has been retrieved. The timestamp, status code, success or failure status of each API request is recorded for easy tracking and debugging.

[0025] It should also be noted that the maximum limit for a single request refers to the upper limit of the amount of data returned by the social media platform API interface in a single request, which needs to be set according to the API specifications of the social media platform being connected.

[0026] Remove records from user interaction data that lack necessary fields such as source_user_id, target_user_id, and timestamp, and standardize the timestamp format to ISO8601 standard, converting it to UTC+0 time. At the same time, use the unique identifier combination of tweet_id + user_id to identify and delete duplicate user ID records, ensuring that the user ID field conforms to the format requirements specified by the platform.

[0027] A table structure is created, and the cleaned and standardized user interaction data, along with the corresponding source user ID, target user ID, and timestamp, are filled into the table. The data is then exported as a comma-separated CSV file, resulting in a user interaction dataset. This user interaction data collection method improves data quality and reliability, and enhances the depth and accuracy of data analysis. The automated data collection and processing workflow reduces manual intervention and improves work efficiency, making it particularly suitable for processing large-scale datasets. Furthermore, its excellent scalability easily adapts to the integration of new interaction types and multi-platform data access, laying the foundation for future feature expansion and upgrades.

[0028] Partitioning Module: Divide the user interaction dataset into N time windows and construct a weighted directed graph for each of the N time windows.

[0029] Specifically, the steps include the following: Determine the total analysis period based on business needs. For example, to analyze the changing trends of user interaction behavior over a month, set the total analysis period to 30 days. Review the user interaction dataset to understand the daily frequency of user interactions. If the data shows a large number of interactions every day with significant fluctuations, choose a shorter time window (e.g., 5 days) to capture these short-term fluctuations. This time window length ensures that there is enough data for meaningful analysis within each time window, while also reflecting the dynamics of information dissemination in the short term.

[0030] Based on the total analysis period (30 days) and the set time window length of 5 days, the total number of windows is 30 / 5, or 6 time windows. For each time window from 1 to 6, the user interaction dataset with timestamps falling within the time window is selected. For example, the first window is the user interaction dataset from day 1 to day 5, the second window is the user interaction dataset from day 6 to day 10, and so on. Each time window has an independent user interaction dataset (source user ID, target user ID, interaction type).

[0031] Iterate through the user interaction dataset for each time window, adding all source user IDs and target user IDs appearing in each time window to different lists (one list per time window). Use deduplication features in programming languages, such as Python's `set()`, to remove duplicate user IDs, resulting in the user set for each time window. For each user ID in the user set, create a corresponding user node in the graph database's data structure, recording the relationship between each user node and its corresponding user ID. Based on the source user ID and target user ID within the time window, create a directed edge between the corresponding user nodes, resulting in a time window social network graph. The directed edges have no weight; they only indicate the existence of an interaction relationship. Store the unweighted social network graph constructed for each time window in a suitable storage medium, such as a local file or a cloud database.

[0032] Calculate the number of interactions and cumulative interaction intensity between the source user and the target user. Use a subjective weighting method to assign values ​​to the interaction type, such as like = 1, comment = 2, share = 3. Use the interaction type as the weight of the number of interactions, and sum the interaction numbers in a weighted manner to obtain the cumulative interaction intensity.

[0033] A weighted value is calculated based on the number of interactions and the cumulative interaction intensity between the source and target users. Rules can be set for this calculation; for example, if the interaction frequency between the source and target users is 5 times and the average interaction intensity is 2, then the weighted value can be set to 7. Note that the weighted values ​​for the source and target users differ for each time window.

[0034] By using weighted values ​​of source and target users, the directed edges in the social network graph within the time window are updated. Specifically, for directed edges between source and target users, a weight attribute is set to the corresponding weighted value, resulting in a weighted directed graph for the time window. By transforming user interaction data into weighted directed edges, complex interaction patterns between users can be captured more accurately. This includes not only simple connections (who follows whom) but also deeper levels of interaction frequency and intensity, such as the specific manifestations of likes, comments, and reposts. This refined processing can identify which connections play a key role in information dissemination, thereby providing a better understanding of the information flow mechanism of the entire social network.

[0035] Dynamic module: By calling the information propagation model through a weighted directed graph of N time windows, the information propagation is simulated to obtain the user's initial influence score in N time windows. The user's initial influence score in N time windows is then weighted and averaged to obtain the dynamic influence score.

[0036] Specifically, the steps include the following: Based on the weighted directed graph of the time window, the initial state of user nodes in the weighted directed graph is set to "susceptible state," indicating that these users have not yet been affected by any information. A high-activity user standard is defined. Specifically, within the time window, the number of interactions between all users and other users in the weighted directed graph is counted. An activity threshold is set based on the historical interaction count using a percentile method. For example, the historical interaction counts are sorted in ascending order, and the 95th percentile of the historical interaction count is selected as the activity threshold. Users whose interaction count exceeds the activity threshold are defined as high-activity users. Several (e.g., 30) high-activity users are randomly selected as seed nodes, and the state of the selected seed nodes is updated to "infected state" as the starting point for information propagation. This is used to initialize the weighted directed graph for each time window.

[0037] Each time window, after initialization, is used to simulate information propagation using an Independent Concatenation Model (ICM+). In this propagation, each directed edge has a corresponding information propagation probability, i.e., the probability of information propagation between each pair of user nodes. This probability is calculated based on the weights of the directed edges. ; in, Represents user node User Node The probability of propagation, Indicates the probability of information propagation. Denotes the base of the natural logarithm. Represents user node User Node The weights of the directed edges (from the weighted directed graph). Indicates the weight of a directed edge; The simulated information propagation process is executed. For each infected seed node, its neighboring nodes are checked. If a neighboring node is currently in a "susceptible state," its state is converted to "infected state" based on the propagation probability. This is done by generating a random number between 0 and 1 and comparing it with the propagation probability. If a neighboring node is successfully converted to an "infected state," it is added to the next round of propagation queue. The simulated information propagation process stops when the propagation queue is empty, meaning there are no new susceptible nodes that can be converted to an "infected state."

[0038] Throughout the simulation, the number and duration of each infected node's infection of susceptible nodes were recorded, and the user's initial influence score within the time window was calculated, expressed by the formula: ; in, This indicates the user's initial influence rating within the time window. This indicates the number of susceptible nodes that have been infected and become infected nodes. This indicates the time when a susceptible node becomes an infected node; By transforming the structural advantages of weighted directed graphs into a quantitative expression of information dissemination capabilities, the originally abstract concept of "influence" acquires an operational and reproducible mathematical meaning. Secondly, by simulating the real information dissemination process, it not only considers the connection relationships between users but also delves into the temporal characteristics and hierarchical effects in the dissemination path, making the scoring results closer to the actual dissemination patterns.

[0039] Setting a decay factor emphasizes that recent interactions have a greater impact on dynamic influence scores. For example, setting a decay factor... α =0.9, which means that the influence of the previous time window is reduced by 90% of that of the current window.

[0040] Based on the attenuation coefficient, the weight of each time window is calculated and normalized, for example, expressed by the formula: ; ; in, Indicates the first The weight of each time window, Indicates the attenuation factor. Indicates the total number of time windows. The normalized first The weight of each time window, This represents the sum of the weights of all time windows. This indicates that the subscript has no specific meaning; The user's dynamic influence score is calculated based on the user's initial influence score for each time window and the normalized weights for each time window. Specifically, ; in, This represents a user's dynamic influence rating. Indicates the user's position in the first month. Preliminary influence score within a time window The normalized first The weight of each time window, Indicates the total number of time windows; By introducing a time decay coefficient to weight and average the initial influence scores of users across different time windows, the significant role of recent interactions in a user's dynamic influence is effectively highlighted, making influence assessment more timely and relevant. The weights for each time window are calculated using an exponential decay method, and weight normalization ensures the fairness and comparability of the scoring system, ultimately yielding a dynamic influence score that comprehensively considers time factors. This approach not only enhances sensitivity to trends in user behavior but also helps to more accurately identify users with key dissemination capabilities within specific time periods, thereby improving the accuracy and practicality of overall influence analysis.

[0041] Cross-layer module: Constructs a multi-layer network and, based on dynamic influence scoring, quantifies the user's influence at different levels within the multi-layer network to obtain a cross-layer influence score.

[0042] Specifically, the steps include the following: The user interaction dataset was divided into a set of likes, a set of comments, and a set of shares. Weighted directed graphs were then constructed for each of these sets, resulting in like, comment, and share layers. Based on business requirements, the weights for the like layer (0.3), comment layer (0.3), and share layer (0.4) were set. These weight values ​​demonstrated high accuracy and stability in tests on social media platform data.

[0043] A unified multi-layered network framework is established with three layer labels: likes, comments, and shares. The like, comment, and share layers are added to the framework according to their corresponding labels, ensuring that different layers share the same user nodes. User nodes in the like, comment, and share layers are scanned; if some users only appear in one layer, isolated nodes are added to the other two layers. A global user index table is built to record the unique identifier of each user in each layer, ensuring that the same user can be located across layers during subsequent analysis. The integrated multi-layered network is exported as JSON format. JSON is a lightweight data exchange format, easy to read and write, and easy for machines to parse and generate; each layer can also be used independently for subsequent behavioral-level influence analysis.

[0044] Extract user activity levels in the like, comment, and share layers. For example, if a user has 50 likes, 30 comments, and 20 shares, the corresponding activity level in the like layer is 50, the activity level in the comment layer is 30, and the activity level in the share layer is 20, with the corresponding activity ratio being 50%:30%:20%.

[0045] A user's dynamic influence score is distributed across the like, comment, and share layers based on the proportion of activity in their respective layers, multiplied by the corresponding behavioral weight. For example, if a user's dynamic influence score is 75, their activity level in the like layer is 50%, and their like weight is 0.3, then the user's influence score in the like layer would be: 75 × 30% × 0.3 = 6.75. This process is repeated for the comment and share layers. For each user, the sum of their influence scores in the like, comment, and share layers is their cross-layer influence score.

[0046] By constructing a multi-layered network consisting of like, comment, and share layers, this study achieves a refined breakdown of the sources of user influence for the first time. Unlike previous methods that relied solely on a single graph model and static scoring, this approach integrates multi-dimensional behavioral features and dynamic evolutionary trends, making influence assessment more scientific, accurate, and practical. Furthermore, the inclusion of a user indexing mechanism and an isolated node completion strategy ensures the consistency and integrity of nodes across all layers in the multi-layered network, providing a solid data foundation for cross-layer influence calculation.

[0047] Identification module: Calculates a comprehensive score based on the user's initial influence score, dynamic influence score, and cross-level influence score. The comprehensive scores are sorted in descending order to identify key users.

[0048] Specifically, the steps include the following: Prepare a spreadsheet and write the column headings in the first row: "User ID," "Initial Influence Score," "Dynamic Influence Score," and "Cross-Level Influence Score." Determine the weight of each score; for example, set the weight of the Initial Influence Score to 0.2, the Dynamic Influence Score to 0.4, and the Cross-Level Influence Score to 0.4. The weighting is not fixed and can be adjusted according to the specific application scenario. For example, in some scenarios, if more emphasis is placed on the user's long-term influence, the weight of the Dynamic Influence Score can be increased; conversely, if more emphasis is placed on short-term explosive growth, the weight of the Initial Influence Score can be increased.

[0049] A comprehensive score is calculated based on each user's initial influence score, dynamic influence score, and cross-level influence score. ; in, This represents the user's overall rating. This represents a user's dynamic influence rating. This indicates the user's initial influence rating within the time window. This represents a user's cross-level influence score; Add a new column called "Overall Rating" to the spreadsheet. Enter the overall rating for each user into the "Overall Rating" column and sort them in descending order, that is, reorganize the user list according to the overall rating from high to low.

[0050] Set a comprehensive score threshold based on the proportion of key users to be identified. For example, to identify the top 10% of key users, use the comprehensive score of users ranked at the 90th percentile as the comprehensive score threshold. Review the sorted user list and identify all users whose comprehensive scores are higher than the comprehensive score threshold; these are the key users. This achieves integrated and quantitative evaluation of user influence, breaking through the limitations of single-dimensional assessment. By integrating short-term performance, long-term trends, and multi-behavioral levels of influence, it constructs a more comprehensive, accurate, and operational user value measurement system, enabling social platforms to automatically identify truly influential and guiding key users.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A social media key user identification system based on the characteristics of non-uniform information propagation, characterized in that: include, The data acquisition module collects user interaction data from social media platforms and preprocesses it to obtain a user interaction dataset. The module is divided into N time windows based on the user interaction dataset, and a weighted directed graph is constructed for the N time windows; The dynamic module calls the information propagation model through a weighted directed graph of N time windows to simulate information propagation, obtain the user's initial influence score in N time windows, and then calculates the weighted average of the user's initial influence scores in N time windows to obtain the dynamic influence score. The cross-layer module constructs a multi-layer network and, based on dynamic influence scoring, quantifies the user's influence at different levels within the multi-layer network to obtain a cross-layer influence score. The identification module calculates a comprehensive score based on a user's initial influence score, dynamic influence score, and cross-level influence score. The comprehensive scores are sorted in descending order to identify key users.

2. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The process involves collecting user interaction data from social media platforms and preprocessing it to obtain a user interaction dataset. Specifically, this includes... We obtain user interaction data through the official API interfaces provided by social media platforms, and then clean and standardize it. The cleaned and standardized user interaction data is labeled with time tags, source user ID, target user ID, and interaction behavior type to obtain a user interaction dataset. The types of interactive behaviors include liking, commenting, and sharing.

3. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The process of dividing the user interaction dataset into N time windows is specifically as follows: Set the total analysis period for the user interaction dataset, and select the time window length based on the analysis requirements and the characteristics of the user interaction dataset; Based on the length of the time window, the total analysis period is evenly divided into N time windows.

4. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The construction of a weighted directed graph using N time windows is specifically as follows: For N time windows, extract the user interaction dataset within the time window, create a node for each user appearing in the interaction dataset, and when there is interaction between users, create unweighted directed edges between the corresponding user nodes to connect them, thus obtaining the social network graph. For each pair of directly connected user nodes, the frequency and intensity of interaction are counted, and the weight values ​​of the directed edges in the social network graph are updated to obtain a weighted directed graph with N time windows.

5. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The method involves calling an information propagation model using a weighted directed graph with N time windows to simulate information propagation and obtain a preliminary influence score for the user across the N time windows. Specifically, Based on a weighted directed graph with N time windows, the user node state within the weighted directed graph is set to the susceptible state. At the same time, a range of highly active user nodes is defined, and W highly active user nodes are randomly selected as the infected state before being imported into the information propagation model to simulate information propagation and obtain the user's preliminary influence score in each time window.

6. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The process involves weighting and averaging the user's initial influence scores across N time windows to obtain a dynamic influence score. Specifically, this involves... Set the attenuation coefficient and calculate the weights of N time windows based on the attenuation coefficient; The user's dynamic influence score is obtained by weighting and averaging the weights of N time windows and the initial influence score.

7. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The construction of the multi-layer network specifically involves... The user interaction dataset is divided into a set of likes, a set of comments, and a set of shares. Weighted directed graphs are then built for each of the sets of likes, comments, and shares to obtain the like layer, comment layer, and share layer, respectively. The like, comment, and share layers are saved as independent layers, forming a unified multi-layered network.

8. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: Based on dynamic influence scoring, the system quantifies a user's influence at different levels within a multi-layered network, yielding a cross-layer influence score. Specifically... The user's dynamic influence score is imported into the like layer, comment layer, and share layer, and the user's influence score at each layer is calculated by combining the weight coefficients of the like layer, comment layer, and share layer. The influence scores at each level are aggregated to obtain the user's cross-level influence score.

9. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The process involves calculating a comprehensive score based on a user's initial influence score, dynamic influence score, and cross-level influence score. These comprehensive scores are then sorted in descending order to identify key users. Create a table, fill in the user ID and the corresponding initial influence score, dynamic influence score and cross-level influence score, and set the importance weight; For each user, a weighted average is calculated in a table based on the set importance weights to obtain the user's overall score.

10. The social media key user identification system based on the non-uniform information propagation characteristics as described in claim 1, characterized in that: The comprehensive scores are sorted in descending order to identify key users. Specifically... Sort all user records from highest to lowest based on their overall score, and set a threshold for the overall score to measure key users. Users whose overall score is higher than the threshold are considered key users.

Images