Campus social network interactive information exchange platform and method

Abstract

The invention discloses a campus social network interaction information exchange platform and method, and relates to the technical field of campus social networks, and the method comprises the following steps: carrying out the information collection of each campus interaction post, collecting the interaction content of all accounts in the post, and guaranteeing the capture of all related comments; for the same post, feature extraction is carried out on all comment information from the same account, after feature extraction is completed, the extracted features are deeply analyzed, and the processed and analyzed features are input into a pre-trained machine learning model. According to the method, through comprehensive collection of comment information and feature extraction, a machine learning model and multi-post comprehensive analysis are combined, and the potential robot accounts are accurately classified and managed. High-risk, general and low-risk hierarchical management strategies are implemented, the false information interception capability is improved, misjudgment is reduced, platform authenticity and user experience are guaranteed, and technical support is provided for healthy development of the campus social network.

Description

Technical Field

[0001] The present invention relates to the technical field of campus social network, and in particular to a campus social network interactive information exchange platform and method. Background Art

[0002] The campus social network interactive information exchange platform is an online communication tool specifically for campus groups. By establishing a virtual campus social space, students, teachers and the school can interact and share information in it. The platform integrates multiple functions such as dynamic publishing, instant messaging, resource sharing, interest groups, etc., which not only meets the needs of campus social interaction, but also supports students to organize activities, academic discussions, learning resource sharing and campus notification push. Through this platform, users can easily obtain campus activity information, participate in academic and social interactions, promote information flow and collaborative communication on campus, and enhance the convenience and interactivity of overall campus life.

[0003] When campus social network interactive information exchange platforms conduct online interactive information exchange, there may be a large number of robot accounts that are not easy to detect on the platform, participating in interactions or publishing false information through automated means. This behavior may cause false public opinion and even affect students' academic and life decisions, leading to information confusion and disrupting normal campus order. Therefore, how to promptly detect robot accounts on campus social network interactive information exchange platforms and take corresponding measures to deal with them is a problem that needs to be solved urgently.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a campus social network interactive information exchange platform and method, which ensures the integrity analysis of account behavior by comprehensively collecting information and extracting features from the comment content of each post, and after classifying the accounts using a pre-trained machine learning model, further improves the assessment accuracy of potential robot account behavior patterns through comprehensive analysis of multiple posts. Finally, by dividing potential robot accounts into high-risk, general and low-risk accounts and implementing targeted management measures, not only the ability to accurately intercept high-risk accounts is improved, the spread of false information is avoided, but also the misjudgment and interference of normal users can be effectively reduced, so as to solve the problems in the above-mentioned background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a campus social network interactive information exchange method, comprising the following steps:

[0007] Collect information on every campus interactive post, collect the interactive content of all accounts in the post, and ensure that all relevant comments are captured;

[0008] For the same post, all comments from the same account are feature extracted. After feature extraction, the extracted features are deeply analyzed and the processed and analyzed features are input into the pre-trained machine learning model to predict the interactive comments of all accounts.

[0009] Based on the prediction results of the machine learning model, all accounts in the post are divided into normal accounts and potential robot accounts;

[0010] For accounts marked as potential robot accounts, obtain the interactive comments of the account under multiple posts, establish a comprehensive analysis set, and obtain more comprehensive data to more accurately evaluate the account's behavior pattern;

[0011] Conduct an in-depth analysis of the comprehensive analysis set of potential robot accounts. Based on the results of the comprehensive analysis, the potential robot accounts are further divided into high-risk accounts, general accounts, and low-risk accounts. Different types of measures are taken for different types of accounts.

[0012] Preferably, information collection is achieved by real-time or batch crawling, which means that different data collection methods are used to obtain all account interaction content in the post from the campus interactive platform;

[0013] Real-time crawling means starting the data collection program immediately after a post is published, monitoring and crawling new comments and interactive content in real time;

[0014] Batch crawling refers to the centralized crawling of multiple posts within a specified time interval, and the unified collection of interaction data within the specified period.

[0015] Preferably, for the same post, feature extraction is performed on all comment information from the same account, wherein the extracted features include the proportion of the same account using highly similar content in multiple comments and the comment frequency of the same account in a short period of time. After analyzing the proportion of the same account using highly similar content in multiple comments and the comment frequency of the same account in a short period of time, a duplicate content rate factor and an active frequency anomaly factor are generated respectively. The duplicate content rate factor is used to quantify the degree to which the same account uses highly similar content in multiple comments, and the active frequency anomaly factor is used to quantify the degree of abnormality in the comment frequency of the same account in a short period of time.

[0016] Preferably, the duplicate content rate factor and the active frequency anomaly factor generated after processing and analyzing the features extracted from the same account are input into a pre-trained machine learning model, and a latent robot coefficient is generated based on the machine learning model, and the interactive comment situation of all accounts is predicted by the latent robot coefficient.

[0017] Preferably, the potential robot coefficient generated when predicting the interactive comments of the account through the machine learning model is compared and analyzed with the pre-set potential robot coefficient reference threshold, and all the accounts in the post are divided. The specific division steps are as follows:

[0018] If the potential robot coefficient is greater than or equal to the potential robot coefficient reference threshold, the account is classified as a potential robot account;

[0019] If the potential robot coefficient is less than the potential robot coefficient reference threshold, the account is classified as a normal account.

[0020] Preferably, for the same post, the specific steps of generating a duplicate content rate factor after analyzing the proportion of highly similar content used by the same account in multiple comments are as follows:

[0021] First, for all comments of the same account, the content similarity between each pair of comments is calculated to form a comment content similarity matrix. Assume that the account has a total of several comments under a certain post. Use the text embedding model in natural language processing technology to convert each comment into a high-dimensional vector. For any two comments C i and C j , calculate the cosine similarity, the calculation expression is:

[0022] ,

[0023] in, and Comments C i and C j The vector representation of and Represents vectors and vector The Euclidean norm of Sim(C i , C j ) indicates comment C i and C j The similarity of comments, i and j represent the index of the comment, and the value range is [-1, 1]. By calculating the pairwise similarity of all comments, a symmetric matrix S is constructed. ij , whose element S ij = sin(C i , C j ), indicating comment Ci and C The similarity between j;

[0024] Based on the similarity matrix S ij , calculate the proportion of duplicate content in the account's comment set, introduce a similarity threshold θ, when Sim(C i , C j )≥θ, depending on the comment C i and C j The formula for defining the proportion of duplicate content in highly similar review pairs is:

[0025] ,

[0026] in, is an indicator function, which takes the value 1 when the condition is met, otherwise it takes the value 0. n is the total number of comments made by the account under the post, and R x is the proportion of duplicate content, with a value range of [0, 1], indicating the proportion of highly similar content in the comment set;

[0027] In obtaining the duplicate content ratio R x After that, the duplicate content rate factor is generated through nonlinear transformation to further amplify the risk of high duplicate content accounts. The generation expression of the duplicate content rate factor is:

[0028] ,

[0029] In the formula, I RG is the duplicate content rate factor, ∈ is the minimum value to avoid the denominator being zero, Through R x Perform non-linear amplification so that accounts with high repetitive content have higher index values. By combining the number of comments and the proportion of duplicate content, high similarities in large-scale comments are dynamically amplified.

[0030] Preferably, for the same post, the specific steps of analyzing the comment frequency of the same account in a short period of time to generate the active frequency abnormality factor are as follows:

[0031] Set a fixed monitoring time window ΔT to count the comment behavior of the same account within the time period. Suppose that within the monitoring time window ΔT, the account has posted a total of y comments, record the timestamp t1, t2, …, ty of each comment, and calculate the time interval Δt between adjacent comments u =t u+1 -t u (u=1,2,…,y-1), next, define the comment interval fluctuation coefficient φ, which is used to measure the fluctuation degree of the time interval. The calculation expression of the comment interval fluctuation coefficient is:

[0032] ,

[0033] Among them, t u+1 and t u are the timestamps of the u+1th and uth moments, Δt u represents the time interval between the u+1th and uth moments. Similarly, Δt u+1 Represents the time interval between the u+2th and u+1th moments;

[0034] In the same monitoring time window ΔT, the frequency surge rate λ is defined to measure the sudden increase in the frequency of account comments. At the beginning of the monitoring time window, the number of comments in each time segment c is counted. k , k = 1, 2, ..., m, where m = ΔT / δT, calculate the frequency surge rate, the calculation expression is:

[0035] ,

[0036] Among them, c k and c k+1 denote the number of comments in the kth and k+1th time segments respectively, and m denotes the total number of time segments in the monitoring time window;

[0037] Combine the comment interval fluctuation coefficient φ and the frequency surge rate λ to generate the active frequency anomaly factor η act , the generated expression is:

[0038] η act =α·φ+β·λ,

[0039] Where α and β are the adjustable weight parameters of the comment interval fluctuation coefficient φ and the frequency surge rate λ, respectively, which are used to balance the contribution of the comment interval fluctuation coefficient φ and the frequency surge rate λ. The active frequency anomaly factor η act Quantifies the abnormality of the account's comment frequency in a short period of time.

[0040] Preferably, for an account marked as a potential robot account, the interactive comment data of the account in multiple posts are collected to establish a comprehensive analysis set, and the post set is set as C, C = {P p}={P1,P2,…,P q}, where P p represents the pth post. For each post, the potential robot coefficient of the post is calculated by the pre-trained machine learning model, and then the weighted average of the potential robot coefficients of multiple posts of the account is performed. The formula is as follows:

[0041] ,

[0042] Among them, w pis the weight of the pth post, PBC multi It represents the comprehensive potential robot coefficient after weighted average, which reflects the comprehensive potential robot behavior degree of the account in all posts. represents the potential robot coefficient of the p-th post;

[0043] Based on the comprehensive analysis set, the comprehensive behavior deviation index is calculated to evaluate the consistency and abnormality of the account's behavior. The potential robot coefficient of the account in multiple posts is compared with the potential robot coefficient reference threshold. The behavior deviation is defined as:

[0044] ,

[0045] in, represents the behavioral deviation of the pth post, Potential threshold It represents the reference threshold of the potential robot coefficient. The comprehensive behavior deviation index is calculated by the behavior deviation. The formula of the comprehensive behavior deviation index is as follows:

[0046] ,

[0047] In the formula, CBAI μ Represents the comprehensive behavioral deviation index.

[0048] Preferably, the comprehensive behavior deviation index is compared and analyzed with a preset first comprehensive behavior deviation index reference threshold and a second comprehensive behavior deviation index reference threshold, wherein the first comprehensive behavior deviation index reference threshold is less than the second comprehensive behavior deviation index reference threshold, and the potential robot account is further divided, and the specific division steps are as follows:

[0049] If the comprehensive behavior deviation index is less than the first comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a low-risk account;

[0050] If the comprehensive behavior deviation index is greater than or equal to the first comprehensive behavior deviation index reference threshold and less than the second comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a general account;

[0051] If the comprehensive behavior deviation index is greater than or equal to the second comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a high-risk account.

[0052] A campus social network interactive information exchange platform, including an information collection module, a feature extraction and prediction module, a preliminary classification module, a multi-post comprehensive analysis module, and a risk classification and management module;

[0053] The information collection module collects information from every campus interactive post, collects the interactive content of all accounts in the post, and ensures that all relevant comments are captured;

[0054] The feature extraction and prediction module extracts features from all comments from the same account for the same post. After the feature extraction is completed, the extracted features are deeply analyzed and the processed and analyzed features are input into the pre-trained machine learning model to predict the interactive comments of all accounts.

[0055] The preliminary classification module divides all the accounts in the post into normal accounts and potential robot accounts based on the prediction results of the machine learning model;

[0056] The multi-post comprehensive analysis module obtains interactive comments under multiple posts of accounts marked as potential robot accounts, builds a comprehensive analysis set, and obtains more comprehensive data to more accurately evaluate the behavior patterns of accounts;

[0057] The risk classification and management module conducts an in-depth analysis of the comprehensive analysis set of potential robot accounts. Based on the results of the comprehensive analysis, the potential robot accounts are further divided into high-risk accounts, general accounts and low-risk accounts. Different types of measures are taken for different types of accounts.

[0058] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0059] The present invention ensures the integrity analysis of account behavior by conducting comprehensive information collection and feature extraction on the comment content of each post. After classifying the accounts using a pre-trained machine learning model, the present invention further improves the assessment accuracy of the behavior patterns of potential robot accounts through comprehensive analysis of multiple posts. Finally, by dividing potential robot accounts into high-risk, general and low-risk accounts and implementing targeted management measures, it not only improves the ability to accurately intercept high-risk accounts and avoid the spread of false information, but also effectively reduces misjudgment and interference with normal users. This refined hierarchical management strategy can not only ensure the information authenticity and interactive order of the platform, but also protect the user experience of real users to the greatest extent, providing strong technical support for the healthy development of campus social networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0061] Figure 1 The present invention is a method flow chart of a campus social network interactive information exchange method.

[0062] Figure 2 The present invention is a module schematic diagram of a campus social network interactive information exchange platform. DETAILED DESCRIPTION

[0063] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0064] The present invention provides Figure 1 A campus social network interactive information exchange method shown includes the following steps:

[0065] Collect information on every campus interactive post, collect the interactive content of all accounts in the post, and ensure that all relevant comments are captured;

[0066] This step can be achieved through real-time or batch crawling to collect information. Real-time or batch crawling to collect information means using different data collection methods to obtain all account interaction content in the post from the campus interactive platform. Real-time crawling means starting the data collection program immediately after the post is published, monitoring and crawling new comments and interactive content in real time. This method can ensure the timeliness of information, especially for scenarios that require immediate response, such as public opinion monitoring of emergencies or hot topics. Batch crawling refers to the centralized crawling of multiple posts within a specified time interval, and the unified collection of interactive data within a specified period. This method is more suitable for background batch processing and regular analysis, and can efficiently collect large amounts of data in a short period of time. Combining the two methods, flexible information collection can be achieved under different needs, effectively balancing data integrity and collection efficiency.

[0067] Data privacy and compliance should be taken into account to ensure compliance with relevant laws and regulations during data collection, such as the Personal Information Protection Law of the People's Republic of China. To optimize performance, real-time data synchronization can be achieved to reduce latency and improve the timeliness of data.

[0068] For the same post, all comments from the same account are feature extracted. After feature extraction, the extracted features are deeply analyzed and the processed and analyzed features are input into the pre-trained machine learning model to predict the interactive comments of all accounts.

[0069] For the same post, feature extraction is performed on all comment information from the same account, where the extracted features include the proportion of the same account using highly similar content in multiple comments and the comment frequency of the same account in a short period of time. After analyzing the proportion of the same account using highly similar content in multiple comments and the comment frequency of the same account in a short period of time, a duplicate content rate factor and an active frequency anomaly factor are generated respectively. The duplicate content rate factor is used to quantify the degree to which the same account uses highly similar content in multiple comments, and the active frequency anomaly factor is used to quantify the degree of abnormality in the comment frequency of the same account in a short period of time.

[0070] The duplicate content rate factor and active frequency anomaly factor generated by processing and analyzing the features extracted from the same account are input into the pre-trained machine learning model, and the potential robot coefficient is generated based on the machine learning model. The interactive comment situation of all accounts is predicted by the potential robot coefficient.

[0071] A pre-trained machine learning model refers to a classification or recognition model that is trained and optimized in advance using a large amount of labeled data. The training process involves extracting features (such as comment frequency, content similarity, etc.) from a large amount of known account data, and associating these features with corresponding labels (i.e., whether it is a robot account). The model will gradually learn the impact of different features on account behavior patterns. For example, high rates of duplicate content and abnormally active frequencies may be more common in robot accounts. Through repeated iterative training, the model continuously optimizes parameters so that it has strong recognition capabilities and can accurately distinguish real users from robot accounts from feature data. In addition, the model usually introduces different types of account samples during the training process to ensure that the model has broad adaptability and can effectively predict even in new data environments.

[0072] In practical applications, such a pre-trained model predicts the potential robot coefficient of an account by inputting new feature data (such as duplicate content rate factors and active frequency anomaly factors). The model first extracts characteristic patterns similar to robot behavior from these features and calculates the potential robot coefficient of the account, that is, the model's probability estimate of the account being a robot. The potential robot coefficient can intuitively quantify the risk of each account becoming a robot, helping the system to identify potential robot accounts and take corresponding measures. The reason why pre-trained machine learning models are effective is that they have been trained with a large number of data samples and can generalize well and adapt to a variety of complex scenarios. On campus social platforms, the use of this model can improve the accuracy of identifying robots, thereby effectively maintaining the authenticity of platform interactions and the health of information flow.

[0073] For the same post, if the same account uses a high proportion of highly similar content in multiple comments, it does indicate that the account has a high risk of being a robot account. This behavior usually reflects the lack of diversity and personalization in the generation of comment content, which may be the result of batch generation or copy and paste by automated programs. Real users' comments are usually adjusted according to context and personal thinking, and have a high content difference. Robot accounts, due to performing specific tasks (such as promotion, public opinion guidance, or interference in discussions), often repeatedly post similar or even identical content, thereby achieving the purpose of creating "false heat" or manipulating public opinion. Therefore, the proportion of highly similar content is an important feature for identifying robot accounts. The hidden danger is that it may cause information confusion, mislead the judgment of other users, and even affect the normal order of campus social platforms.

[0074] For the same post, the specific steps to generate the duplicate content rate factor after analyzing the proportion of highly similar content used by the same account in multiple comments are as follows:

[0075] First, for all comments of the same account, the content similarity between each pair of comments is calculated to form a comment content similarity matrix. Assume that the account has a total of several comments under a certain post. Use the text embedding model in natural language processing (NLP) technology (such as Word2Vec, BERT, etc.) to convert each comment into a high-dimensional vector. For any two comments C i and C j , calculate the cosine similarity, the calculation expression is:

[0076] ,

[0077] in, and Comments C i and C j The vector representation of and Represents vectors and vector The Euclidean norm of Sim(C i , C j ) indicates comment C i and C j The similarity of comments is calculated by calculating the pairwise similarity of all comments and constructing a symmetric matrix S. ij , whose element S ij = sin(C i , C j ), indicating comment C i and C j The degree of similarity between

[0078] Based on the similarity matrix S ij , calculate the proportion of duplicate content in the account's comment set. To this end, a similarity threshold θ is introduced. When Sim(C i , C j )≥θ, depending on the comment C i and C j The formula for defining the proportion of duplicate content in highly similar review pairs is:

[0079] ,

[0080] in, is an indicator function, which takes the value 1 when the condition is met, otherwise it takes the value 0. n is the total number of comments made by the account under the post, and R x is the proportion of duplicate content, with a value range of [0, 1], indicating the proportion of highly similar content in the comment set;

[0081] This step calculates the proportion of duplicate content of highly similar pairs in the reviews, which is used to measure the degree of duplication in the review set.

[0082] In obtaining the duplicate content ratio R x After that, a duplicate content rate factor is generated through nonlinear transformation to further amplify the risk of accounts with high duplicate content. The generation expression of the duplicate content rate factor is:

[0083] ,

[0084] In the formula, I RG is the duplicate content rate factor, ∈ is the minimum value to avoid the denominator being zero, Through R x Perform non-linear amplification so that accounts with high repetitive content have higher index values. By combining the number of comments and the proportion of duplicate content, high similarities in large-scale comments are dynamically amplified.

[0085] From the duplicate content rate factor, we can see that for the same post, the larger the performance value of the duplicate content rate factor generated by analyzing the proportion of highly similar content used by the same account in multiple comments, the higher the proportion of highly similar content used by the same account in multiple comments, which indicates that the account is more likely to be a robot account. A high duplicate content rate usually means that the comments lack diversity and personalization, which may be the result of batch generation through automated programs, which is a typical feature of robot accounts. On the contrary, if the value of the duplicate content rate factor is small, it means that the comment content of the account is more diverse and natural, which conforms to the behavior pattern of real users, so the risk of it being a robot account is lower. The duplicate content rate factor provides a reliable quantitative basis for identifying potential robot accounts, which helps to maintain the interactive authenticity and information health of campus social platforms.

[0086] For the same post, if the same account comments frequently in the same post within a short period of time, it does indicate that the account has a high risk of being a robot account. This behavior is usually inconsistent with the interaction habits of normal users. Normal users usually think about the content of the post before replying, while robot accounts may post a large number of comments in a very short period of time through automated programs to achieve the purpose of manipulating public opinion, diverting traffic, or creating false activity. This abnormally high frequency of commenting behavior not only disrupts the normal order of communication, but may also cover up the views of real users, thereby threatening the authenticity of interaction and information quality of campus social platforms.

[0087] For the same post, the specific steps to generate the active frequency abnormality factor after analyzing the comment frequency of the same account in a short period of time are as follows:

[0088] Set a fixed monitoring time window ΔT to count the comment behavior of the same account in this time period. Suppose that within the monitoring time window ΔT, the account has posted a total of y comments, and record the timestamp t1, t2, ..., t y , and calculate the time interval Δt between adjacent comments u =t u+1 -t u (u=1,2,…,y-1), next, define the comment interval fluctuation coefficient φ, which is used to measure the fluctuation degree of the time interval. The calculation expression of the comment interval fluctuation coefficient is:

[0089] ,

[0090] Among them, t u+1 and t u are the timestamps of the u+1th and uth moments, Δt u represents the time interval between the u+1th and uth moments. Similarly, Δt u+1 Represents the time interval between the u+2th and u+1th moments;

[0091] The comment interval fluctuation coefficient can quantify the volatility of the comment interval time interval. The larger the comment interval fluctuation coefficient, the more unstable the comment interval fluctuation is, which may indicate high-frequency behavior of the robot account under program control.

[0092] In the same monitoring time window ΔT, the frequency surge rate λ is defined to measure the sudden increase in the frequency of account comments. At the beginning of the monitoring time window, the number of comments c in each time segment (for example, every ΔT) is counted. k (k = 1, 2, ..., m, where m = ΔT / δT), calculate the frequency surge rate, the calculation expression is:

[0093] ,

[0094] Among them, c k and c k+1 denote the number of comments in the kth and k+1th time segments respectively, and m denotes the total number of time segments in the monitoring time window;

[0095] This step is used to measure the maximum frequency growth rate of the number of comments within the monitoring time window. The higher the frequency surge rate λ value, the more significant the surge in comment frequency at a certain moment, which is usually related to batch operations of robot accounts.

[0096] Combine the comment interval fluctuation coefficient φ and the frequency surge rate λ to generate the active frequency anomaly factor η act , the generated expression is:

[0097] η act =α·φ+β·λ

[0098] , where α and β are the adjustable weight parameters of the comment interval fluctuation coefficient φ and the frequency surge rate λ, respectively, which are used to balance the contribution of the comment interval fluctuation coefficient φ and the frequency surge rate λ, and the active frequency anomaly factor η act Quantifies the abnormality of the account's comment frequency in a short period of time.

[0099] From the active frequency anomaly factor, we can see that for the same post, the larger the performance value of the active frequency anomaly factor generated after analyzing the comment frequency of the same account in a short period of time, the more abnormal the comment frequency of the account in a short period of time, and the greater the difference between its behavioral characteristics and the interaction mode of normal users, so the higher the risk of becoming a robot account. This usually reflects that the account may make high-frequency comments in a short period of time through automated means, trying to create false activity or manipulate public opinion. On the contrary, if the active frequency anomaly factor is small, it means that the comment frequency of the account is closer to the natural interaction rhythm of normal users, and the possibility of a robot account is also lower. Therefore, the active frequency anomaly factor can be used to intuitively evaluate whether the account has abnormal behavior, thereby effectively assisting in identifying potential robot accounts.

[0100] The machine learning model is not specifically limited here, and can achieve the repetition rate factor I RG and the activity frequency abnormality factor η act Perform comprehensive analysis to generate potential robot coefficients Bot In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0101] Potential Robot Coefficient BOt The resulting calculation formula is:

[0102] ,

[0103] Where, u1 and u2 are the repeated content rate factors I RG and the activity frequency abnormality factor η act The preset proportional coefficient, and u1 and u2 are both greater than 0.

[0104] It can be seen from the potential robot coefficient that for the same post, the larger the performance value of the duplicate content rate factor generated after analyzing the proportion of highly similar content used by the same account in multiple comments, the larger the performance value of the active frequency anomaly factor generated after analyzing the comment frequency of the same account in a short period of time. In other words, the larger the performance value of the potential robot coefficient generated when predicting the interactive comments of an account through a machine learning model, the greater the risk that the account is a robot account, and vice versa.

[0105] Based on the prediction results of the machine learning model, all accounts in the post are divided into normal accounts and potential robot accounts;

[0106] The potential robot coefficient generated by predicting the interactive comments of the account through the machine learning model is compared with the pre-set potential robot coefficient reference threshold to divide all the accounts in the post. The specific division steps are as follows:

[0107] If the potential robot coefficient is greater than or equal to the potential robot coefficient reference threshold, the account is classified as a potential robot account;

[0108] Potential robot accounts refer to accounts that are identified as having possible robot behavior characteristics through the prediction of machine learning models. These accounts usually show interactive characteristics that are inconsistent with normal users, such as high frequency of comments in a short period of time, high similarity of comment content, or lack of emotional fluctuations, and may be controlled by automated programs. The behavioral patterns of potential robot accounts are often programmed and repetitive, and may be used to manipulate public opinion, create false activity, or promote specific content, thereby interfering with the normal interactive environment of the platform.

[0109] If the potential robot coefficient is less than the potential robot coefficient reference threshold, the account is classified as a normal account;

[0110] Normal accounts refer to accounts that have been identified as having normal user behavior characteristics after analysis by machine learning models. The behavioral characteristics of such accounts are consistent with the natural interaction patterns on the platform, such as comment frequency and content diversity, which are closer to the daily habits of real users. The interactive content of normal accounts is often diverse and emotionally rich, with human thinking characteristics, and usually does not perform large-scale, high-frequency operations in a short period of time, so it is believed that it will not have a negative impact on the interactive order of the platform.

[0111] For accounts marked as potential robot accounts, obtain the interactive comments of the account under multiple posts, establish a comprehensive analysis set, and obtain more comprehensive data to more accurately evaluate the account's behavior pattern;

[0112] Conduct in-depth analysis of the comprehensive analysis set of potential robot accounts. Based on the results of the comprehensive analysis, the potential robot accounts are further divided into high-risk accounts, general accounts, and low-risk accounts. Different types of measures are taken for different types of accounts.

[0113] For accounts marked as potential robot accounts, collect the interactive comment data of the account in multiple posts, establish a comprehensive analysis set, and set the post set as C, C = {P p}={P1,P2,…,P q}, where P prepresents the pth post. For each post, the potential robot coefficient of the post is calculated by the pre-trained machine learning model, and then the weighted average of the potential robot coefficients of multiple posts of the account is performed. The formula is as follows:

[0114] ,

[0115] Among them, w p is the weight of the pth post, which is usually assigned according to the number of comments on the post or the platform authority. multi It represents the comprehensive potential robot coefficient after weighted average, which reflects the comprehensive potential robot behavior degree of the account in all posts. represents the potential robot coefficient of the p-th post;

[0116] Based on the comprehensive analysis set, the comprehensive behavior deviation index is calculated to evaluate the consistency and abnormality of the account's behavior. The potential robot coefficient of the account in multiple posts is compared with the potential robot coefficient reference threshold. The behavior deviation is defined as:

[0117] ,

[0118] in, represents the behavioral deviation of the pth post, Potential threshold It represents the reference threshold of the potential robot coefficient. The comprehensive behavior deviation index is calculated by the behavior deviation. The formula of the comprehensive behavior deviation index is as follows:

[0119] ,

[0120] In the formula, CBAI μ represents the comprehensive behavioral deviation index;

[0121] This step combines squared deviation and absolute deviation to quantify the volatility of account behavior in multiple posts. Comprehensive Behavior Deviation Index CBAI μ The larger the value, the greater the fluctuation of the potential robot coefficient of the account in different posts and the more abnormal the behavior pattern.

[0122] The comprehensive behavior deviation index is compared and analyzed with the preset first comprehensive behavior deviation index reference threshold and the second comprehensive behavior deviation index reference threshold, wherein the first comprehensive behavior deviation index reference threshold is less than the second comprehensive behavior deviation index reference threshold, and the potential robot account is further divided. The specific division steps are as follows:

[0123] If the comprehensive behavior deviation index is less than the first comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a low-risk account;

[0124] If the comprehensive behavior deviation index is greater than or equal to the first comprehensive behavior deviation index reference threshold and less than the second comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a general account;

[0125] If the comprehensive behavior deviation index is greater than or equal to the second comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a high-risk account.

[0126] High-risk accounts refer to accounts that show obvious robot behavior characteristics after comprehensive analysis, and their behavior patterns are significantly different from those of normal users. Such accounts usually interact frequently in a short period of time, publish highly repetitive or meaningless content, and may even participate in malicious activities, such as manipulating public opinion, spreading false information, or guiding bad trends. Such accounts pose a direct threat to the healthy operation and user experience of the platform, so immediate measures need to be taken, such as freezing account permissions or directly deleting the account to prevent it from continuing to interfere with the order of the platform. At the same time, in order to ensure the accuracy and timeliness of the processing, the relevant data and analysis results of the account should be submitted to the security team for further in-depth investigation and evidence collection, and legal measures can be taken if necessary.

[0127] General accounts refer to accounts that have certain suspicious behaviors but are not sufficient to completely determine that they are robot accounts. Such accounts may show abnormal characteristics in some interactions, such as a comment frequency slightly higher than normal, some repeated content, or a lack of semantic diversity, but they do not reach the significant level of high-risk levels. In order to reduce potential risks, milder restrictions can be taken on these accounts, such as limiting the number or frequency of their daily comments, and requiring them to confirm their identity through secondary verification (such as image verification codes or personalized questions) in specific circumstances. This strategy can effectively manage potential risks while avoiding excessive interference with real users, leaving room for further observation and judgment.

[0128] Low-risk accounts refer to accounts that have only shown slightly abnormal behavior after analysis. Their overall behavior pattern is basically the same as that of normal users, but there are occasional anomalies. For example, the account may have a high frequency of comments in a certain post, but behave completely normally in other posts. In order to protect the user experience and reduce misjudgments, low-risk accounts are usually not subject to immediate restrictions, but are included in the platform's continuous monitoring system. Through regular back-end review and dynamic analysis, the actual situation of these accounts can be further confirmed to avoid affecting the normal user experience of real users due to excessive intervention. Such a monitoring mechanism can balance the needs of platform security and user experience to the greatest extent.

[0129] The present invention ensures the integrity analysis of account behavior by conducting comprehensive information collection and feature extraction on the comment content of each post. After classifying the accounts using a pre-trained machine learning model, the present invention further improves the assessment accuracy of the behavior patterns of potential robot accounts through comprehensive analysis of multiple posts. Finally, by dividing potential robot accounts into high-risk, general and low-risk accounts and implementing targeted management measures, it not only improves the ability to accurately intercept high-risk accounts and avoid the spread of false information, but also effectively reduces misjudgment and interference with normal users. This refined hierarchical management strategy can not only ensure the information authenticity and interactive order of the platform, but also protect the user experience of real users to the greatest extent, providing strong technical support for the healthy development of campus social networks.

[0130] The present invention provides Figure 2 A campus social network interactive information exchange platform is shown, comprising an information collection module, a feature extraction and prediction module, a preliminary classification module, a multi-post comprehensive analysis module, and a risk classification and management module;

[0131] The information collection module collects information from every campus interactive post, collects the interactive content of all accounts in the post, and ensures that all relevant comments are captured;

[0132] The feature extraction and prediction module extracts features from all comments from the same account for the same post. After the feature extraction is completed, the extracted features are deeply analyzed and the processed and analyzed features are input into the pre-trained machine learning model to predict the interactive comments of all accounts.

[0133] The preliminary classification module divides all the accounts in the post into normal accounts and potential robot accounts based on the prediction results of the machine learning model;

[0134] The multi-post comprehensive analysis module obtains interactive comments under multiple posts of accounts marked as potential robot accounts, builds a comprehensive analysis set, and obtains more comprehensive data to more accurately evaluate the behavior patterns of accounts;

[0135] The risk classification and management module conducts in-depth analysis on the comprehensive analysis set of potential robot accounts. Based on the results of the comprehensive analysis, the potential robot accounts are further divided into high-risk accounts, general accounts, and low-risk accounts. Different types of measures are taken for different types of accounts.

[0136] A campus social network interactive information exchange method provided by an embodiment of the present invention is implemented through the above-mentioned campus social network interactive information exchange platform. The specific methods and processes of the campus social network interactive information exchange platform are detailed in the embodiment of the above-mentioned campus social network interactive information exchange method, which will not be repeated here.

[0137] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0138] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0139] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A campus social network interactive information exchange method, characterized in that: The following steps are involved: Collect information on every campus interactive post, collect the interactive content of all accounts in the post, and ensure that all relevant comments are captured; For the same post, all comments from the same account are feature extracted. After feature extraction, the extracted features are deeply analyzed and the processed and analyzed features are input into the pre-trained machine learning model to predict the interactive comments of all accounts. Based on the prediction results of the machine learning model, all accounts in the post are divided into normal accounts and potential robot accounts; For accounts marked as potential robot accounts, obtain the interactive comments of the account under multiple posts, establish a comprehensive analysis set, and obtain more comprehensive data to more accurately evaluate the account's behavior pattern; Conduct an in-depth analysis of the comprehensive analysis set of potential robot accounts. Based on the results of the comprehensive analysis, the potential robot accounts are further divided into high-risk accounts, general accounts, and low-risk accounts. Different types of measures are taken for different types of accounts.

2. A campus social network interactive information exchange method according to claim 1, characterized in that: Information collection is achieved through real-time or batch crawling. Real-time or batch crawling to achieve information collection means using different data collection methods to obtain all account interaction content in the posts from the campus interactive platform; Real-time crawling means starting the data collection program immediately after a post is published, monitoring and crawling new comments and interactive content in real time; Batch crawling refers to the centralized crawling of multiple posts within a specified time interval, and the unified collection of interaction data within the specified period.

3. The campus social network interactive information exchange method according to claim 1, characterized in that: For the same post, feature extraction is performed on all comment information from the same account, where the extracted features include the proportion of the same account using highly similar content in multiple comments and the comment frequency of the same account in a short period of time. After analyzing the proportion of the same account using highly similar content in multiple comments and the comment frequency of the same account in a short period of time, a duplicate content rate factor and an active frequency anomaly factor are generated respectively. The duplicate content rate factor is used to quantify the degree to which the same account uses highly similar content in multiple comments, and the active frequency anomaly factor is used to quantify the degree of abnormality in the comment frequency of the same account in a short period of time.

4. The campus social network interactive information exchange method according to claim 3, characterized in that: The duplicate content rate factor and active frequency anomaly factor generated by processing and analyzing the features extracted from the same account are input into the pre-trained machine learning model, and the potential robot coefficient is generated based on the machine learning model. The interactive comment situation of all accounts is predicted by the potential robot coefficient.

5. The campus social network interactive information exchange method according to claim 4, characterized in that: The potential robot coefficient generated by predicting the interactive comments of the account through the machine learning model is compared with the pre-set potential robot coefficient reference threshold to divide all the accounts in the post. The specific division steps are as follows: If the potential robot coefficient is greater than or equal to the potential robot coefficient reference threshold, the account is classified as a potential robot account; If the potential robot coefficient is less than the potential robot coefficient reference threshold, the account is classified as a normal account.

6. The campus social network interactive information exchange method according to claim 3, characterized in that: For the same post, the specific steps to generate the duplicate content rate factor after analyzing the proportion of highly similar content used by the same account in multiple comments are as follows: First, for all comments of the same account, the content similarity between each pair of comments is calculated to form a comment content similarity matrix. Assume that the account has a total of several comments under a certain post. Use the text embedding model in natural language processing technology to convert each comment into a high-dimensional vector. For any two comments C i and C j , calculate the cosine similarity, the calculation expression is: ， in, and Comments C i and C j The vector representation of and Represents vectors and vector The Euclidean norm of Sim(C i , C j ) indicates comment C i and C j The similarity of comments, i and j represent the index of the comment, and the value range is [-1, 1]. By calculating the pairwise similarity of all comments, a symmetric matrix S is constructed. ij , whose element S ij = sin(C i , C j ), indicating comment C i and C j The degree of similarity between Based on the similarity matrix S ij , calculate the proportion of duplicate content in the account's comment set, introduce a similarity threshold θ, when Sim(C i , C j )≥θ, depending on the comment C i and C j The formula for defining the proportion of duplicate content in highly similar review pairs is: ， in, is an indicator function, which takes the value 1 when the condition is met, otherwise it takes the value 0. n is the total number of comments made by the account under the post, and R x is the proportion of duplicate content, with a value range of [0, 1], indicating the proportion of highly similar content in the comment set; In obtaining the duplicate content ratio R x After that, the duplicate content rate factor is generated through nonlinear transformation to further amplify the risk of high duplicate content accounts. The generation expression of the duplicate content rate factor is: ， In the formula, I RG is the duplicate content rate factor, ∈ is the minimum value to avoid the denominator being zero, Through R x Perform non-linear amplification so that accounts with high repetitive content have higher index values. By combining the number of comments and the proportion of duplicate content, high similarities in large-scale comments are dynamically amplified.

7. A campus social network interactive information exchange method according to claim 6, characterized in that: For the same post, the specific steps to generate the active frequency abnormality factor after analyzing the comment frequency of the same account in a short period of time are as follows: Set a fixed monitoring time window ΔT to count the comment behavior of the same account in this time period. Suppose that within the monitoring time window ΔT, the account has posted a total of y comments, and record the timestamp t1, t2, ..., t y , and calculate the time interval Δt between adjacent comments u =t u+1 -t u (u=1,2,…,y-1), next, define the comment interval fluctuation coefficient φ, which is used to measure the fluctuation degree of the time interval. The calculation expression of the comment interval fluctuation coefficient is: ， Among them, t u+1 and t u are the timestamps of the u+1th and uth moments, Δt u represents the time interval between the u+1th and uth moments. Similarly, Δt u+1 Represents the time interval between the u+2th and u+1th moments; In the same monitoring time window ΔT, the frequency surge rate λ is defined to measure the sudden increase in the frequency of account comments. At the beginning of the monitoring time window, the number of comments in each time segment c is counted. k , k = 1, 2, ..., m, where m = ΔT / δT, calculate the frequency surge rate, the calculation expression is: ， Among them, c k and c k+1 denote the number of comments in the kth and k+1th time segments respectively, and m denotes the total number of time segments in the monitoring time window; Combine the comment interval fluctuation coefficient φ and the frequency surge rate λ to generate the active frequency anomaly factor η act , the generated expression is: or act =a·φ+b·λ, Where α and β are the adjustable weight parameters of the comment interval fluctuation coefficient φ and the frequency surge rate λ, respectively, which are used to balance the contribution of the comment interval fluctuation coefficient φ and the frequency surge rate λ. The active frequency anomaly factor η act Quantifies the abnormality of the account's comment frequency in a short period of time.

8. The campus social network interactive information exchange method according to claim 4, characterized in that: For accounts marked as potential robot accounts, collect the interactive comment data of the account in multiple posts, establish a comprehensive analysis set, and set the post set as C, C = {P p }={P1,P2,…,P q }, where P p represents the pth post. For each post, the potential robot coefficient of the post is calculated by the pre-trained machine learning model, and then the weighted average of the potential robot coefficients of multiple posts of the account is performed. The formula is as follows: ， Among them, w p is the weight of the pth post, PBC multi It represents the comprehensive potential robot coefficient after weighted average, which reflects the comprehensive potential robot behavior degree of the account in all posts. represents the potential robot coefficient of the p-th post; Based on the comprehensive analysis set, the comprehensive behavior deviation index is calculated to evaluate the consistency and abnormality of the account's behavior. The potential robot coefficient of the account in multiple posts is compared with the potential robot coefficient reference threshold. The behavior deviation is defined as: ， in, represents the behavioral deviation of the pth post, Potential threshold It represents the reference threshold of the potential robot coefficient. The comprehensive behavior deviation index is calculated by the behavior deviation. The formula of the comprehensive behavior deviation index is as follows: ， In the formula, CBAI μ Represents the comprehensive behavioral deviation index.

9. The campus social network interactive information exchange method according to claim 8, characterized in that: The comprehensive behavior deviation index is compared and analyzed with the preset first comprehensive behavior deviation index reference threshold and the second comprehensive behavior deviation index reference threshold, wherein the first comprehensive behavior deviation index reference threshold is less than the second comprehensive behavior deviation index reference threshold, and the potential robot account is further divided. The specific division steps are as follows: If the comprehensive behavior deviation index is less than the first comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a low-risk account; If the comprehensive behavior deviation index is greater than or equal to the first comprehensive behavior deviation index reference threshold and less than the second comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a general account; If the comprehensive behavior deviation index is greater than or equal to the second comprehensive behavior deviation index reference threshold, the potential robot account is further classified as a high-risk account.

10. A campus social network interactive information exchange platform, used to implement the campus social network interactive information exchange method described in any one of claims 1 to 9, characterized in that: It includes information collection module, feature extraction and prediction module, preliminary classification module, multi-post comprehensive analysis module and risk classification and management module; The information collection module collects information from every campus interactive post, collects the interactive content of all accounts in the post, and ensures that all relevant comments are captured; The feature extraction and prediction module extracts features from all comments from the same account for the same post. After the feature extraction is completed, the extracted features are deeply analyzed and the processed and analyzed features are input into the pre-trained machine learning model to predict the interactive comments of all accounts. The preliminary classification module divides all the accounts in the post into normal accounts and potential robot accounts based on the prediction results of the machine learning model; The multi-post comprehensive analysis module obtains interactive comments under multiple posts of accounts marked as potential robot accounts, builds a comprehensive analysis set, and obtains more comprehensive data to more accurately evaluate the behavior patterns of accounts; The risk classification and management module conducts an in-depth analysis of the comprehensive analysis set of potential robot accounts. Based on the results of the comprehensive analysis, the potential robot accounts are further divided into high-risk accounts, general accounts and low-risk accounts. Different types of measures are taken for different types of accounts.

Images