Social network user influence analysis method and system
By introducing multidimensional indicators and an improved PageRank algorithm, combined with a time decay factor, the shortcomings of the traditional PageRank algorithm in evaluating the influence of social network users are addressed, achieving a more accurate and flexible user influence assessment, which is applied to public opinion monitoring, opinion leader identification, and precision marketing.
Patent Information
- Application Number
- CN202510862283.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-25
- Publication Date
- 2025-09-26
Smart Images
Figure CN120706711A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of social network analysis, and in particular to a method and system for analyzing the influence of social network users. Background Art
[0002] With the rapid development of social networking platforms, user-generated content plays a key role in information dissemination. The influence of users directly determines the dissemination range and effectiveness of their published content. Therefore, accurately assessing user influence has become an important topic in social network research.
[0003] Traditional methods for evaluating user influence rely primarily on the PageRank algorithm, which quantifies the importance of each node by analyzing the relationship network between users. However, the traditional PageRank algorithm for evaluating user influence has the following shortcomings: (1) Single indicator: It only considers the relationship structure between users and ignores other important attributes of users.
[0004] (2) Inflexible weight allocation: There is a lack of dynamic adjustment capabilities for the importance of different indicators, making it difficult to adapt to diverse application scenarios.
[0005] (3) Ignoring time factors: e.g. Figure 1 and Figure 2 As shown in Figure 2, the traditional PageRank algorithm is used to calculate the ranking of web pages, but it is not used to calculate the influence of users in social networks. Therefore, it does not take into account the temporal dynamic characteristics of user behavior, resulting in the evaluation results possibly not reflecting the user's current true influence. Summary of the Invention
[0006] The present invention aims to provide a method and system for analyzing the influence of social network users, so as to solve the technical problem that the traditional PageRank algorithm is inaccurate in evaluating user influence.
[0007] The basic solution provided by the present invention is: a method for analyzing the influence of social network users, comprising the following steps: S1, collects raw data related to each user and user influence from social networks in real time; S2, extracts and calculates key features related to user influence from each user's raw data; key features include user self-influence, user relationship breadth, user authority, and temporal dynamic characteristics; S3, at least based on each user's user self-influence, user relationship breadth and user authority, use the improved PageRank algorithm to iteratively calculate the user influence of each user; S4, based on the time dynamic characteristics of each user, introduces a time decay factor to adjust the user influence of each user.
[0008] The present invention also provides a social network user influence analysis system, which implements the above-mentioned social network user influence analysis method; the system includes: The data processing module is used to collect raw data related to each user and user influence from the social network in real time; The feature extraction module uses a distributed architecture to extract key features related to each user's influence from the raw data. Key features include user self-influence, user relationship breadth, user authority, and temporal dynamic characteristics. The model building module is used to build a user influence calculation model based on the improved PageRank algorithm and calculate the user influence of each user; The result output module is used to output the user influence of each user in a preset manner.
[0009] The working principle and advantages of the present invention are: Compared with existing technologies, this invention introduces multidimensional indicators to combine static attributes (such as the number of users' followers, followers, and historical tweets) with dynamic attributes (dynamic behaviors in users' social networks: likes, comments, reposts, and other indicators). This method extracts key features such as user self-influence, user relationship breadth, user authority, and temporal dynamics from multiple perspectives to analyze user influence. This not only considers the relationship network between users but also enriches the depth and breadth of user influence calculation and analysis based on key features, which is conducive to improving the accuracy and applicability of user influence assessment. On this basis, the traditional PageRank algorithm is optimized to iteratively calculate user influence. In combination with user activity, a time decay factor is introduced to dynamically adjust user influence, ultimately achieving a comprehensive and accurate assessment of user influence. This invention can be widely used in fields such as public opinion monitoring, opinion leader identification, false information prevention and control, and precision marketing, and has important social value and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Figure 1 This is the calculation process diagram of the traditional PageRank algorithm; Figure 2 This is a flow chart of the traditional PageRank algorithm's calculation method for web page influence; Figure 3 A flowchart of a method for analyzing the influence of social network users provided by an embodiment of the present invention; Figure 4 A schematic diagram of a feature extraction process provided by an embodiment of the present invention; Figure 5 A schematic diagram of a processing flow for calculating user influence characteristics and calculations in a social network according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a social network user influence analysis system provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0011] The following is a further detailed description through specific implementation methods: The embodiment is basically as shown in the attached Figure 3 A method for analyzing the influence of social network users includes the following steps: S1, collects raw data related to each user and user influence from social networks in real time; S2, extracts and calculates key features related to user influence from each user's raw data; key features include user self-influence, user relationship breadth, user authority, and temporal dynamic characteristics; S3, at least based on each user's user self-influence, user relationship breadth and user authority, use the improved PageRank algorithm to iteratively calculate the user influence of each user; S4, based on the time dynamic characteristics of each user, introduces a time decay factor to adjust the user influence of each user.
[0012] Specifically: S1 supports data collection from multiple data sources (not limited to mainstream social platforms). Raw data includes the number of original content published and interactive behavior data, the number of followers, followings, and interaction frequency of users, as well as their authentication status, the quality of their past posts, and the number of citations. By introducing multi-dimensional indicators and constructing appropriate key features, we can conduct a comprehensive assessment and analysis of user influence.
[0013] In this example, calculations and processing are performed based on a public user dataset from Kaggle. Data preprocessing includes denoising, formatting, and normalization. Unlike traditional methods, this preprocessing simplifies usernames, removing emojis and symbols like exclamation marks. Furthermore, a number is used to quantify whether a user is authenticated.
[0014] In S2, Figure 4As shown in the figure, the process of extracting user features from user data is demonstrated: user data is divided into "static user data" and "dynamic user data". "Static user data" mainly refers to indicators such as the number of user fans, the number of user followers, and whether the user is authenticated; "dynamic user data" mainly refers to users' likes, comments, and reposts of tweets. The corresponding key user features are extracted from the static and dynamic user data: the breadth of user relationships and user authority. Based on the breadth of user relationships and user authority, the key feature user self-influence is extracted, and then the user influence is obtained iteratively. Finally, a user feature is spliced to obtain the final user feature. Traditional methods often only use static user data, such as the number of users' fans and followers, and often ignore the dynamic behavioral characteristics of users in social networks: liking, commenting, and reposting tweets.
[0015] In S1 and S2, the raw data used to quantify the breadth of user relationships includes the number of users' followers, followings, and interaction frequency. The weight coefficients of each indicator are dynamically adjusted to quantify the breadth of user relationships and to measure the scope of users' connections within social networks. By combining static user data (primarily "number of users' followers" and "number of users' followings") with dynamic user data (primarily "interaction frequency"), this combination of static and dynamic data improves the authenticity and generalizability of the calculation of user relationship breadth compared to traditional methods that only focus on static data. The specific calculation steps are as follows: (1) Count the number of users’ followers to reflect the user’s audience range; (2) Based on the obtained Kaggle public user dataset as the original data; (3) Based on the original data, after traversal and comparison, the maximum number of fans among all users is obtained as the maximum number of fans; (4) In a similar way, obtain the number of followers of the user and the maximum number of followers of all users, that is, the maximum number of followers; (5) Obtain the number of likes, comments, and reposts for a user's tweets under a specific topic, divide them by the maximum value of the corresponding indicator among users, and sum them to obtain the interaction frequency. (The original data includes data from multiple topics, and the specific topic data is obtained here for calculation.) Construct a user relationship matrix, determine the initial weights of the number of users' fans, number of followers, and interaction frequency, and calculate the breadth of user relationships using the following formula.
[0016]
[0017] in, Represents a user the breadth of user relationships; Represents a user Number of fans; Indicates the maximum number of followers a user has in the original data; Represents a user Number of followers; Indicates the maximum number of followers a user has in the original data; Indicates the frequency of interaction; Indicates the weight coefficient of the corresponding indicator.
[0018] Dynamically adjusting the weight coefficients of each indicator includes at least one of manual dynamic adjustment and dynamic adjustment using an attention mechanism. Manual dynamic adjustment involves determining initial weights to calculate the user influence of all users, ranking them, comparing and analyzing the user influence of a portion of the test set in the original data, and making manual dynamic adjustments based on the analysis results. Alternatively, dynamic adjustment can be achieved through programmatic design, such as by using an attention mechanism.
[0019] Among them, comparative analysis can be carried out by comparing the overall results with the actual results. For example, if 100 users are ranked, and 80 users in one experiment are in line with the actual ranking, while in another experiment, 90 users are in line with the actual ranking, then the other experiment is better and the parameters of the other experiment are better. Therefore, the parameters of the other experiment are selected. This process is similar to the dynamic parameter adjustment process in deep learning, and the purpose is to make the results more consistent with the actual values.
[0020] In S1 and S2, the raw data used to quantify user authority includes the user's authentication status and the number of tweets they have posted in history. The weight coefficients of each indicator are dynamically adjusted to quantify user authority, which is used to measure the user's credibility and influence in social networks. Calculating user authority is also particularly important for understanding the user's influence. The higher the user's authority, the higher their influence, meaning that user influence is positively correlated with user authority. The specific calculation steps are as follows: (1) Check the user's authentication status. Authenticated users are assigned a larger weight, and unauthenticated users are assigned a smaller weight. Assuming that the weight of a normal unauthenticated user is 1, the weight of an authenticated user is 1.5. The specific weight value is adaptively determined according to the actual computing scenario requirements. It should be noted that the weight here is defined differently from the weight coefficient in the weighted summation, which is 1.
[0021] (2) Get the number of tweets posted by the user in history. For example, if the total number of tweets posted by user A in history is 15, then the number of tweets posted in history is 15.
[0022] (3) Obtain the maximum number of tweets sent in history by all users in the original data. For example, if the total number of tweets sent in history by user B in the original data is 1000, which is greater than the total number of tweets sent in history by all other users in the original data, then the maximum number of tweets sent in history by the user in the original data is 1000.
[0023] (4) Dynamically adjust the weight coefficient of each indicator to ensure that the authority evaluation results can accurately reflect the user's credibility and influence. The dynamic adjustment method is the same as above.
[0024] Determine the initial weights of the user's authentication status and historical tweets, and calculate the user's authority using the following formula:
[0025]
[0026] in, Represents a user User authority; Represents a user The value corresponding to whether the account has been authenticated; Represents a user The number of tweets sent in history; Indicates the maximum number of tweets sent by the user in the original data; a and b represent the weight coefficients of the corresponding indicators. In this embodiment: If not authenticated, Take 1, otherwise take 1.5.
[0027] In the process of calculating user authority, the present invention combines the number of tweets sent by the user in history to judge the user's authority. Considering that the greater the number of historical tweets sent by the user, the more active the user is, and the user with low long-term activity has lower authority, the present invention adopts the method of using the number of historical tweets sent by the user to assist in measuring the user's authority, rather than the traditional method of using "whether the user is authenticated". Compared with the traditional method, the evaluation and analysis results are more accurate.
[0028] like Figure 5 As shown, in S1 and S2, the user relationship breadth and user authority are calculated, the weight coefficients of user relationship breadth and user authority are dynamically adjusted, and the user's self-influence is quantified to measure the user's own content creation ability and dissemination effect. The details are as follows: (1) Combining the above calculation process, calculate the user authority and user relationship breadth; (2) Dynamically adjust the weight coefficients of user relationship breadth and user authority, and perform weighted summation of user authority and user relationship breadth to obtain the final user self-influence; Determine the initial weights of user authority and user relationship breadth, and iteratively calculate each user's self-influence using the following formula:
[0029] in, Represents a user User self-influence; Represents a user the breadth of user relationships; Represents a user User authority; Represents the weight coefficient. Among them, the weight coefficient The dynamic adjustment method is the same as above.
[0030] like Figure 4 and Figure 5 As shown in Figure 3, in S3, the user influence of each user is calculated based on the improved PageRank algorithm, and the influence score of each user is updated through the iterative formula until convergence. The final score reflects the user influence. It is calculated by the following formula:
[0031] in, Represents a user The final PageRank value of the user User influence; Represents a user User self-influence; Indicates a link to a user The user collection of represents the final PageRank value of user j in the user set, that is, the user influence of user j; represents the number of external relationships of user j in the user set, which can be the number of followers; d represents the damping coefficient, which is used to simulate the probability of random jumps.
[0032] Based on the improved PageRank algorithm, two indicators, user relationship breadth and user authority, are introduced to quantify user self-influence. Then, the user-link user set information is combined, and a damping coefficient is set to balance the impact of random jumps and relationship transmission. The iterative calculation is performed until the user influence score converges, and finally the user influence is obtained.
[0033] In S4, the raw data used for the time dynamics feature includes the user's active time, which is used to determine whether the user has been inactive for a long time. A time decay factor is introduced to reduce the influence score of users who have been inactive for a long time, so that the final user influence reflects the user's current status in real time.
[0034] Based on the number of tweets sent and the user's historical behavior, the most recent time a user took an action on a tweet, such as liking, commenting, or retweeting, is defined as the most recent active time, and the last time a user took an action, such as liking, commenting, or retweeting, is defined as the last active time. If the difference between the most recent active time and the last active time exceeds one month, the user is considered inactive for a long time. For users with a long inactivity period, their user influence score is calculated by multiplying the calculated user influence score by a time decay factor. The time decay factor can be selected between 0.5 and 0.7, with 0.6 being the preferred value.
[0035] The process further includes S5, performing statistical processing on the user influence of each user, wherein the statistical processing includes ranking the user influence of each user, and classifying and grading the users according to the ranking results.
[0036] Specifically, users are sorted from high to low according to their calculated user influence values, and high and low thresholds are set (for example, the high threshold is set to 2 and the low threshold is set to 0.5. Of course, the specific values are adaptively set according to the actual computing scenario requirements), and classified as follows: (1) High-influence users: Users whose comprehensive influence calculation value is higher than the high threshold, usually opinion leaders or core communicators; (2) Medium-influence users: users whose comprehensive influence calculation value is in the middle range (below the high threshold and above the low threshold), usually ordinary active users; (3) Low-influence users: Users whose comprehensive influence calculation value is lower than the low threshold, usually ordinary users or inactive users.
[0037] The role of user classification and grading, for example, is key user mining: screening out highly influential users through algorithms, identifying opinion leaders or core communicators in social networks; brand cooperation: helping companies accurately locate opinion leaders suitable for cooperation, improving brand exposure and user trust; community operation: finding core users who can drive community activity in specific fields or interest groups; it has important social value and economic benefits.
[0038] Statistical processing also includes forming user influence distribution, marking user classification and grading results, and analyzing the dynamic changes in user influence of each user in different time periods; The process further includes S6, visually displaying and / or exporting the statistical processing results of the user influence of each user.
[0039] like Figure 6 As shown, this embodiment further provides a social network user influence analysis system, which executes the above-mentioned social network user influence analysis method. The system includes: The data processing module is used to collect raw data related to each user and their influence from social networks in real time. The data collection module supports access to multiple data sources, including but not limited to mainstream social platforms such as Twitter, Facebook, and Weibo, and can automatically adapt to the data formats of different platforms.
[0040] The feature extraction module is used to extract key features related to each user and user influence from the raw data using a distributed architecture. Key features include user self-influence, user relationship breadth, user authority, and temporal dynamic characteristics. The distributed architecture supports efficient processing of large-scale data.
[0041] The model building module is used to build a user influence calculation model based on the improved PageRank algorithm, calculate the user influence of each user, and introduce a time decay factor according to the time dynamic characteristics of each user to adjust the user influence of each user; the model building module provides a configurable parameter interface, allowing users to adjust model parameters according to specific needs.
[0042] The prediction output module is used to output each user's user influence in a preset manner. The user influence data calculated using this method is used as the prediction output. The prediction output module supports multi-dimensional statistical processing and analysis of each user's user influence, including: statistical analysis of user influence distribution; identification and labeling of high-influence users; analysis of dynamic changes in user influence over different time periods; and the export and display of user classification and grading results. Furthermore, the prediction output module enables visual display, presenting user influence data and statistical analysis results in charts.
[0043] It is understandable that the above system can fully execute the above method with the same effect, which will not be described again here.
[0044] This embodiment provides a method and system for analyzing user influence on social networks. By introducing multidimensional indicators, the present invention combines static attributes (such as the number of users' followers, followings, and historical tweets) with dynamic attributes (such as user behaviors within social networks, such as likes, comments, and reposts). This method extracts key features from multiple perspectives, including user self-influence, user relationship breadth, user authority, and temporal dynamics, to analyze user influence. This method not only considers the relationship network between users but also enriches the depth and breadth of user influence calculation and analysis based on these key features, thereby improving the accuracy and applicability of user influence assessment. Furthermore, it optimizes the traditional PageRank algorithm for iteratively calculating user influence and, based on user activity, introduces a time decay factor to dynamically adjust user influence, ultimately achieving a comprehensive and accurate assessment of user influence. The present invention can be widely applied in fields such as public opinion monitoring, identifying opinion leaders, preventing and controlling false information, and precision marketing, and possesses significant social value and economic benefits.
[0045] The above is only an embodiment of the present invention. Common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. Ordinary technicians in the relevant field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all existing technologies in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the relevant field can improve and implement this scheme in combination with their own abilities under the guidance of this application. Some typical well-known structures or well-known methods should not become obstacles for ordinary technicians in the relevant field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent.
Claims
1. A method for analyzing the influence of social network users, characterized in that: The steps include: S1, collects raw data related to each user and user influence from social networks in real time; S2, extracts and calculates key features related to user influence from each user's raw data; key features include user self-influence, user relationship breadth, user authority, and temporal dynamic characteristics; S3, at least based on each user's user self-influence, user relationship breadth and user authority, use the improved PageRank algorithm to iteratively calculate the user influence of each user; S4, based on the time dynamic characteristics of each user, introduces a time decay factor to adjust the user influence of each user.
2. A social network user influence analysis method according to claim 1, characterized in that: In S1 and S2, the breadth of user relationships and user authority are calculated, and combined with the dynamic adjustment of the weight coefficients of each indicator, the user's self-influence is quantitatively calculated to measure the user's own content creation ability and dissemination effect.
3. A social network user influence analysis method according to claim 1, characterized in that: In S1 and S2, the original data used to quantitatively calculate the breadth of user relationships include the number of users' fans, number of followers, and interaction frequency. Combined with the dynamic adjustment of the weight coefficients of each indicator, the breadth of user relationships is quantitatively calculated to measure the scope of users' connections in social networks.
4. A social network user influence analysis method according to claim 1, characterized in that: In S1 and S2, the original data for quantitatively calculating user authority includes the user's authentication status and the number of tweets posted in history. Combined with the dynamic adjustment of the weight coefficients of each indicator, the user authority is quantitatively calculated to measure the user's credibility and influence in social networks.
5. A method for analyzing the influence of social network users according to any one of claims 2, 3 or 4, characterized in that: The method of dynamically adjusting the weight coefficient of each indicator includes at least one of a manual dynamic adjustment method and a dynamic adjustment method introducing an attention mechanism; the manual dynamic adjustment method is to determine the initial weight calculation to obtain the user influence of all users and sort them, compare and analyze the user influence of some test sets in the original data, and perform manual dynamic adjustment based on the analysis results.
6. A social network user influence analysis method according to claim 1, characterized in that: In S3, the user influence of each user is calculated based on the improved PageRank algorithm, which is calculated by the following formula: in, Represents a user The final PageRank value of the user User influence; Represents a user User self-influence; Indicates a link to a user The user collection of represents the final PageRank value of user j in the user set, that is, the user influence of user j; represents the number of external relations of user j in the user set; d represents the damping coefficient, which is used to simulate the probability of random jumps.
7. A social network user influence analysis method according to claim 1, characterized in that: In S4, the raw data used for the time dynamics feature includes the user's active time, which is used to determine whether the user has been inactive for a long time. A time decay factor is introduced to reduce the influence score of users who have been inactive for a long time, so that the final user influence reflects the user's current status in real time.
8. A method for analyzing the influence of social network users according to claim 1, characterized in that: The process further includes S5, performing statistical processing on the user influence of each user, wherein the statistical processing includes ranking the user influence of each user, and classifying and grading the users according to the ranking results.
9. A method for analyzing the influence of social network users according to claim 8, characterized in that: Statistical processing also includes forming user influence distribution, marking user classification and grading results, and analyzing the dynamic changes in user influence of each user in different time periods; The process further includes S6, visually displaying and / or exporting the statistical processing results of the user influence of each user.
10. A social network user influence analysis system, characterized in that: A method for analyzing the influence of social network users according to any one of claims 1 to 9 is implemented, wherein the system comprises: The data processing module is used to collect raw data related to each user and user influence from the social network in real time; The feature extraction module uses a distributed architecture to extract key features related to each user's influence from the raw data. Key features include user self-influence, user relationship breadth, user authority, and temporal dynamic characteristics. The model building module is used to build a user influence calculation model based on the improved PageRank algorithm, calculate the user influence of each user, and introduce a time decay factor according to the time dynamic characteristics of each user to adjust the user influence of each user; The prediction output module is used to output the user influence of each user in a preset manner.