Social relation intimacy calculation method and system under social media privacy propagation

Abstract

The invention provides a social relationship intimacy calculation method and system under social media privacy propagation, and relates to the technical field of social network analysis, and the method comprises the steps: building a directed weighted graph based on user privacy propagation behaviors, taking a user as a node, taking a directed propagation relationship as an edge, and fusing an edge weight with a propagation frequency, a content average sensitivity and a topological distance adjustment factor; secondly, a Personalized PageRank algorithm is improved, a path distance Gaussian attenuation transfer matrix is introduced, and a global trust matrix is iteratively calculated; a double-exponential attenuation function is adopted for historical interaction, short-term fast attenuation and long-term slow attenuation are modeled, and a time-varying weight is obtained; quantifying privacy content sensitivity in combination with information entropy and a propagation risk model; identifying a user community and quantifying a community contribution degree; global trust, time-varying weight and sensitivity multiplication are fused, normalization is carried out after community contribution degree correction, and a final intimacy score is generated. According to the method, the relationship intimacy under privacy propagation is accurately described, and dynamic evaluation is supported.

Description

Technical Field

[0001] This invention belongs to the field of social network analysis technology, and relates to a method and system for calculating the intimacy of social relationships under the privacy dissemination of social media, and in particular to a method and system for calculating the intimacy of social relationships under the privacy dissemination of social media scenarios. Background Technology

[0002] As social media permeates all aspects of life, the information shared by users has expanded from public information and shared interests to highly sensitive privacy information, including personal identity, health status, financial status, and geographical location. This type of privacy information is often disseminated through non-public or semi-public channels such as targeted sending, small groups, and multi-level forwarding, forming a complex "privacy dissemination network." In this scenario, accurately assessing the intimacy of social relationships between users is a crucial prerequisite for predicting privacy sharing intentions, implementing refined privacy access control, and providing early warnings of high-risk dissemination links.

[0003] However, traditional and mainstream methods for calculating the intimacy of social relationships are mainly designed around public interaction behaviors, making it difficult to effectively adapt to the special characteristics of privacy dissemination. Existing research mainly focuses on the following technical approaches: (1) Intimacy calculation method based on explicit interaction frequency: This type of method uses the frequency and quantity of explicit interaction behaviors such as likes, comments, public reposts, and the number of mutual friends as the core indicators of intimacy, and calculates it through weighted summation or simple aggregation. This method is simple and intuitive to implement and was widely used in early social platforms. However, its fundamental flaw is that it completely ignores interactions in privacy scenarios and cannot distinguish the difference in intimacy quality represented by "casual chat" and "sharing secrets", resulting in serious distortion of evaluation results in privacy dissemination scenarios.

[0004] (2) Intimacy Model Based on Graph Network Topology: To capture indirect connections, researchers have introduced graph theory methods, using indicators such as the number of common neighbors, the Adamic-Adar index, and Personalized PageRank (PPR) to assess the strength of connections between users through multi-hop paths. This type of method models the transmission of social influence to some extent. However, its edge weights are usually based on the frequency of public interactions or binary relations, failing to incorporate the three core elements of privacy scenarios—"propagation behavior" itself, "sensitivity of propagation content," and "timeliness of propagation"—into the weight system, resulting in a weak correlation between the calculated "connection degree" and "intimacy of privacy sharing."

[0005] (3) Dynamic intimacy calculation method based on time-series behavior modeling: Considering the dynamic evolution of social relationships, some studies have introduced time decay functions (such as exponential decay) to assign time-varying weights to historical interactions, making the impact of recent interactions greater than that of long-term interactions. This method improves the timeliness of intimacy assessment. However, its decay model is usually relatively simple (such as single exponential decay), failing to finely distinguish the differences in decay rate between different types of interactions (such as privacy dissemination vs. public likes), and it is not deeply integrated with other dimensions such as content and network structure.

[0006] (4) Intimacy Calculation Methods Integrating Community Discovery: Recognizing the layered structure of social networks, some methods introduce community detection (such as the Louvain algorithm) when calculating intimacy, giving intimacy bonuses to users belonging to the same community. This simulates the intuition that "insiders are more trustworthy." However, existing methods usually only use the community as a binary label (whether they belong to the same community), failing to quantify the user's active role and intimacy contribution within the community (e.g., who is the core disseminator of private information within the community), and the correction mechanism is crude and lacks personalization.

[0007] Based on the above, existing technologies have the following four specific drawbacks when applied to social media privacy dissemination scenarios: First, existing graph-based methods (such as PPR) fail to highlight the "privacy propagation" behavior in their edge weight definitions, nor do they differentiate the sensitivity of the propagated content, resulting in the graph structure failing to accurately depict the transmission network of social relationship intimacy. Second, existing time-series models, such as the exponential decay formula, use a globally uniform decay parameter, which cannot adapt to the differences in the evolution speed between privacy relationships and ordinary social relationships, and the decay calculation is not related to graph propagation or content sensitivity. Furthermore, there is a lack of measurement in terms of content sensitivity; existing methods either treat all interactions the same or only perform simple category labeling of privacy content, lacking a continuous, computable, and data-driven sensitivity measurement model that integrates content attributes and propagation risks. Finally, existing community correction methods fail to delve into the community to measure the activity and contribution of users as "propagation nodes" of privacy information; the community weighting coefficients are set subjectively, lacking a data-driven personalized adjustment mechanism.

[0008] In summary, when applied to privacy dissemination scenarios, existing technical solutions either neglect core behavioral characteristics, ignore differences in content value, adopt static or single time models, or make too superficial use of network structures, making it difficult to construct a comprehensive, accurate, and dynamic model that can reflect the intimacy calculation process during privacy sharing. Summary of the Invention

[0009] To address the problems in the background technology, this invention provides a method for calculating the intimacy of social relationships under the privacy dissemination of social media. This method proposes a complete, multi-layered coupled calculation framework, from data representation to multi-dimensional feature quantification and collaborative fusion. The core innovation lies in treating the privacy dissemination behavior and its content itself as the most direct "chain of responsibility" for building and measuring intimacy relationships. Through a modular mathematical model, this "chain of responsibility" is deconstructed and comprehensively quantified from multiple perspectives, fundamentally solving the problem of accurately calculating the intimacy of social relationships in the privacy dissemination scenario.

[0010] By constructing a dynamic entity masking network, the weights of the "masked view" and the "retained view" are intelligently adjusted within a unified feature space, and a human-machine collaborative feedback loop based on a difficult example scoring mechanism is introduced, thereby improving the accuracy of intent discovery in an open and dynamic financial environment.

[0011] To achieve the above objectives, this invention provides a method for calculating the intimacy of social relationships under the premise of privacy dissemination on social media, comprising: Based on the privacy information dissemination behavior among users, nodes are defined as users, edges are directed propagation relationships, and edge weights are adjusted by combining propagation frequency, average content sensitivity, and topological distance factors to construct a directed weighted graph for privacy propagation. Based on the directed weighted graph, the Personalized PageRank algorithm is improved by introducing a transition matrix with Gaussian decay of path distance and iteratively calculating the global trust matrix containing direct and indirect paths between users. Based on the behavior of privacy information dissemination among users, a double exponential decay function is used for historical interaction behavior to model short-term rapid decay and long-term slow decay modes respectively, and a time-varying weight that decays over time is calculated for each historical interaction behavior among users. Based on the behavior of privacy information dissemination among users, and combined with the information entropy and dissemination risk model, a sensitivity score is calculated for each piece of privacy content disseminated by a user. Based on the directed weighted graph, the community to which the user belongs is identified, and the user's contribution to community intimacy within each community is quantified; The social relationship intimacy between users is obtained by multiplying and fusing the global trust matrix, time-varying weights, and sensitivity scores, and then corrected based on the community intimacy contribution. The corrected social relationship intimacy is normalized to generate the final intimacy score.

[0012] As a further improvement of the present invention, the formula for calculating the edge weight is as follows: in, Indicates the propagation frequency, Indicates the average sensitivity of the content. Indicates the path length. All are harmonic coefficients, satisfying ,and ; As the core variable, through protrude A key contribution to the intensity of privacy dissemination is the construction of network edge rights that are adapted to different scenarios and centered on content sensitivity.

[0013] As a further improvement to the present invention, the improved Personalized PageRank algorithm introduces a transition matrix with Gaussian decay of path distance, as shown in the formula: The iterative formula is: in, This indicates the introduction of path distance-based... The Gaussian decay term makes the trust transfer probability higher over short distances than over long distances, simulating efficient trust transfer within tight circles; E represents the initial intimacy matrix, and α represents the damping factor; iteration continues until convergence ||T (k) -T (k-1) || F < The multi-hop trust propagation is modeled in a refined manner by using a Gaussian decay transfer matrix to reduce the decay of trust with increasing distance between spheres.

[0014] As a further improvement of the present invention, the double exponential decay function is: Where β represents the weight balancing short-term and long-term effects, λ1 and λ2 represent decay coefficients, and γ represents the power-law exponent. This indicates a rapid short-term decline, simulating the strong but quickly fading impact of recent private interactions on social intimacy. It represents a slow, long-term decay, simulating a persistent but weak relationship left by long-term interactions.

[0015] As a further improvement of the present invention, the formula for calculating the sensitivity of privacy content is as follows: in, Let r(p) represent the probability distribution of content categories, r(p) represent the dissemination risk score, and k represent the adjustment coefficient.

[0016] As a further improvement of the present invention, the formula for calculating the transmission risk score r(p) is as follows: Among them, the spread depth indicates the number of times the private content is forwarded.

[0017] As a further improvement of the present invention, the formula for quantifying a user's contribution to community intimacy in each community is as follows: Where τ(i,p) represents user i's contribution to the spread of privacy content p within community c, and Np represents the breadth of the spread of privacy content p within the community. This represents the logarithmic transformation of the breadth of the spread of private content p within community c; the greater the breadth, the greater the contribution. This indicates the largest contribution from the community to this privacy content.

[0018] As a further improvement to this invention, the global trust matrix, time-varying weights, and sensitivity scores are multiplied and fused to obtain the intimacy of social relationships between users, as shown in the formula: The community correction formula is: in, This represents the global trust value among users derived from the global trust matrix. The average time-varying weight of historical interactions between the two users Indicates from user To users Average sensitivity score of the disseminated content; δ represents the enhancement coefficient, and η represents the nonlinear adjustment factor; Indicates user and users The inner product of the intimacy contribution vectors in each community measures the user's... and users Similarity and overlap of roles in community networks.

[0019] As a further improvement of the present invention, the modified social relationship intimacy is normalized to generate a final intimacy score, the formula of which is: in, This indicates the self-use parameter; σ represents the current batch data. standard deviation It is a constant; Ultimately, the generated It is a value between 0 and 1, with the closer to 1 representing the user's... For users The closer the social relationship in terms of privacy sharing.

[0020] This invention also provides a social relationship intimacy calculation system under social media privacy propagation, including: a data preprocessing and privacy propagation graph construction module, a multi-hop trust transfer calculation module, a double exponential time-varying decay module, a privacy content sensitivity quantification module, a community structure detection and role quantification module, a multi-dimensional intimacy fusion and correction module, and an intimacy normalization and output module; The data preprocessing and privacy propagation graph construction module is used for: This is used to construct a directed weighted graph for privacy propagation based on the behavior of privacy information dissemination among users. Nodes are defined as users, edges are directed propagation relationships, edge weights are calculated, and privacy propagation is constructed. The multi-hop trust transfer calculation module is used for: Connected to the data preprocessing and privacy propagation graph construction module, based on the directed weighted graph, the Personalized PageRank algorithm is improved by introducing a transition matrix with Gaussian decay of path distance to calculate the global trust matrix among users; The dual-exponential time-varying decay module is used for: Connected to the data preprocessing and privacy propagation graph construction module, based on the privacy information propagation behavior between users, time-varying weights are calculated for historical interaction behaviors using a double exponential decay function; The privacy-sensitive content quantification module is used for: Connected to the data preprocessing and privacy propagation graph construction module, and combined with the information entropy and propagation risk model, the sensitivity of privacy content is calculated; The community structure detection and role quantification module is used for: Connected to the data preprocessing and privacy propagation graph construction module, the Louvain algorithm is used to detect community structure and quantify the community intimacy contribution of users in each community; The multi-dimensional intimacy fusion and correction module is used for: It is connected to the multi-hop trust transfer calculation module, the double exponential time-varying decay module, the privacy content sensitivity quantification module, and the community structure detection and role quantification module respectively. It calculates the basic intimacy through multiplication fusion and performs community correction on the basic intimacy based on the community contribution. The intimacy normalization and output module is used for: Connected to the multi-dimensional intimacy fusion and correction module, the corrected intimacy is normalized using an adaptive Sigmoid algorithm to generate the final intimacy score.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention improves the multi-dimensional coupling framework of multi-hop trust transfer, double exponential temporal decay, quantification of privacy content sensitivity, and correction of community role contribution by constructing a directed weighted graph that integrates propagation frequency, content sensitivity, and topological distance. It solves the pain points of existing technologies in privacy propagation scenarios, such as unreasonable edge weights, rigid temporal sequences, lack of sensitivity, and coarse community correction. It achieves accurate, dynamic, and interpretable calculation of social relationship intimacy, and the output of standardized scores can directly drive practical applications such as privacy management and risk warning.

[0022] This invention constructs a scene-adaptive and accurate network representation. By fusing propagation frequency, average content sensitivity (core variable), and topological distance adjustment factor with edge weights, it constructs a directed weighted graph that truly reflects the intensity of privacy propagation and the value of content. This solves the problem that existing graph methods do not highlight the "privacy propagation" behavior and content sensitivity with edge weights, and lays the network foundation for scene adaptation in subsequent calculations.

[0023] This invention achieves refined modeling of multi-hop trust transfer, improves the Personalized PageRank algorithm, introduces a transition matrix with Gaussian decay of path distance to simulate the dynamic transfer process of trust decaying with the distance of social circles, and more realistically depicts the convergence and decay of intimacy in complex networks by iteratively calculating the global trust matrix, thus solving the problem of "coarse trust transfer modeling" in traditional methods.

[0024] This invention enhances the ability to characterize temporal evolution by employing a dual exponential decay function to simultaneously capture two modes of privacy relationships: short-term rapid decay simulates the strong and fleeting impact of recent significant sharing, while long-term slow decay simulates the persistent foundation of long-term mild interaction. This overcomes the rigidity of a single exponential decay function, making intimacy assessment more timely and realistic.

[0025] This invention provides a scientific measure of privacy content sensitivity, integrating information entropy (measuring the inherent uncertainty of content) and dissemination risk (the breadth and depth of response content diffusion), and proposes a continuous sensitivity model. This model solves the problems of missing or coarse sensitivity measurements in existing methods, and can distinguish the differences in the quality of different privacy information, providing a key basis for improving the accuracy of the model.

[0026] This invention optimizes the personalized correction of community effects. It detects communities through the Louvain algorithm, quantifies the community intimacy contribution (continuous value) of users based on their dissemination behavior and influence, and makes a non-linear correction based on role similarity (the inner product of contribution vectors measures the degree of role overlap). It goes beyond the superficial model of "bonus points for being in the same community" and more reasonably reflects the "circle resonance" effect, solving the problem of insufficient personalization in existing community correction.

[0027] This invention achieves the synergistic fusion of multi-dimensional features, coupling trust transmission, time-varying decay, content sensitivity, and community structure. Through multiplicative fusion, it emphasizes the combined effect of relationship strength, timeliness, and content value. Combined with community correction, it reflects the personalized enhancement of role similarity. The whole process is logically clear and highly interpretable, which is superior to the end-to-end black box model and is easy to debug and optimize.

[0028] This invention outputs standardized results that can be directly applied. By using adaptive Sigmoid normalization to map intimacy to the [0,1] interval (automatically adjusting the steepness of the curve according to the data dispersion), the output results have good discriminative power and strong interpretability, and can directly drive practical applications such as dynamic control of privacy permissions, early warning of high-risk transmission links, and assessment of social relationship quality. Attached Figure Description

[0029] Figure 1 This is an overall architecture diagram of a social relationship intimacy calculation method under the social media privacy propagation disclosed in an embodiment of the present invention; Figure 2 This is a flowchart of a method for multi-dimensional intimacy fusion and correction disclosed in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] The present invention will now be described in further detail with reference to the accompanying drawings: like Figure 1As shown, the method for calculating social relationship intimacy under social media privacy propagation provided by this invention follows the logic of "constructing a basic network model → quantifying multi-dimensional intimacy features → performing multi-level information fusion". First, we abandon the crude approach of conflating public interaction with privacy propagation, and instead construct a directed weighted graph specifically for privacy propagation information. The edge weights are not simply the number of interactions, but combine propagation frequency with the average sensitivity of the propagated content, and introduce network topological distance as a regulating factor, thereby generating a high-quality network topology that truly reflects the potential for social relationship intimacy transmission. This step is the cornerstone that distinguishes this method from existing technologies. Based on this topologically constructed network, this invention performs calculations from four dimensions: a multi-hop trust propagation calculation module, a double-exponential time-varying decay module, a privacy content sensitivity quantification module, and a structure-aware community correction module. Each dimension proposes an innovative solution to a weak link in existing research. Finally, the outputs of the above four dimensions are guided to a centralized fusion and correction module. Using the community correction module, based on the similarity of users' contribution vectors in various communities, the basic intimacy is personalized and enhanced. The entire process uses adaptive Sigmoid normalization to output the final result, ensuring the comparability and interpretability of the scores. Specifically, it includes: S1. Based on the privacy information dissemination behavior among users, nodes are defined as users, edges are directed propagation relationships, and edge weights are adjusted by combining propagation frequency, average content sensitivity, and topological distance factors to construct a directed weighted graph for privacy propagation. The formula for calculating the edge weight is as follows: in, Indicates the propagation frequency, Indicates the average sensitivity of the content. Indicates the path length. All are harmonic coefficients, satisfying ,and ; As the core variable, through protrude A key contribution to the intensity of privacy dissemination is the construction of network edge rights that are adapted to different scenarios and centered on content sensitivity.

[0032] Specifically, this step is responsible for transforming raw, mixed social media logs into a structured computational foundation. Input data includes: privacy dissemination logs (sender, recipient, timestamp, content identifier), public interaction logs, user attributes, and group relationship data.

[0033] The processing steps include: (1) Data cleaning: Remove system messages, records that are out of time window, and records that are submitted repeatedly.

[0034] (2) Entity association: aggregate the cross-platform and cross-session behaviors of the same user by ID.

[0035] (3) Construct a directed weighted graph Where V is the set of user nodes and E is the set of directed edges, if user To users If private information has been disseminated, then there is a risk of privacy breaches. W is the edge weight matrix. A directed weighted graph is constructed from this. The edge weights directly reflect the "potential for privacy information transmission" between nodes based on privacy propagation, laying a data foundation for scenario adaptation in subsequent calculations.

[0036] S2. Based on the directed weighted graph, the Personalized PageRank algorithm is improved by introducing a transition matrix with Gaussian decay of path distance and iteratively calculating the global trust matrix containing direct and indirect paths between users. To address the issue of traditional methods' coarse modeling of indirect trust, we improved the classic Personalized PageRank algorithm. The innovation lies in the construction of its transition probability matrix: based on edge weights, we introduced a Gaussian decay term based on the global distance between nodes. This ensures that trust is not only transmitted along edges, but its transmission effect is also constrained by the distance within social circles, thus more realistically simulating the dynamic process of affinity decaying and converging with the number of hops in complex social networks.

[0037] The improved Personalized PageRank algorithm introduces a transition matrix with Gaussian decay of path distance, as shown in the formula: The iterative formula is: In the formula, This indicates the introduction of path distance-based... The Gaussian decay term makes the trust transfer probability higher over short distances than over long distances, simulating efficient trust transfer within tight circles; E represents the initial intimacy matrix, and α represents the damping factor; iteration continues until convergence ||T (k) -T (k-1) || F < The multi-hop trust propagation is modeled in a refined manner by using a Gaussian decay transfer matrix to reduce the decay of trust with increasing distance between spheres.

[0038] Specifically, Figure for step S1 This illustrates the "potential" of direct transmission, but in reality, trust can be transferred through intermediaries (e.g., A trusts B, B trusts C, leading to A developing some indirect trust in C). We need an algorithm to simulate this multi-hop transmission process that decays with the number of hops.

[0039] The graph constructed in step S1 in this step This process runs on a scale designed to quantify direct and indirect privacy trust. Based on the "static" propagation graph constructed in step S1, this step simulates the dynamic diffusion and convergence of privacy trust in social networks, calculating the global trust value between any pair of users that contains direct and indirect paths. It addresses the problem of traditional methods oversimplifying trust transfer modeling, which only considers direct relationships or a limited number of hops.

[0040] PPR naturally simulates the process of trust propagation along edges and returning to the starting point with a certain probability through random walks and restart mechanisms. Its iterative convergence result integrates the influence of paths of all lengths, with longer paths contributing less. We improve upon the classic Personalized PageRank algorithm by replacing its transition matrix with one based on our custom edge weights. This matrix solves the problem of unreasonable edge weight definition in the existing PPR algorithm.

[0041] (1) Construct the propagation transition matrix This formula applies to edge weights Based on this, a path distance-based method was introduced. Gaussian decay term This makes it possible for trust transfer probability to be higher over short distances than over long distances, even with the same edge weights, which is more in line with the actual situation that trust decays with the number of hops.

[0042] The reason for this design is: even from the user's perspective... arrive direct edge rights Very high, but if and They are far apart in the network ( (Large), this attenuation term will make The probability of transfer decreases. This aligns more with social intuition: trust is more easily and efficiently transferred within close circles, and less likely to jump to distant nodes.

[0043] (2) Iteratively calculate the global trust matrix. The innovation of this step lies in incorporating the semantics of privacy scenarios into the transition matrix. It is embedded in the core process of trust iteration, making the final output... The matrix contains a multi-hop global trust relationship based on a privacy propagation network. It will... The static edge weights are transformed into dynamic, global trust metrics.

[0044] S3. Based on the privacy information dissemination behavior between users, a double exponential decay function is used for historical interaction behavior to model short-term rapid decay and long-term slow decay modes respectively, and a time-varying weight that decays over time is calculated for each historical interaction behavior between users. To address the rigidity of the single time decay model, this paper proposes a double exponential smooth decay model. This model includes a rapidly decaying exponential term and a slowly decaying power-law term, which can simultaneously capture two modes that may exist in private relationships: "the strong but fleeting impact of recent significant sharing" and "the durable intimacy base formed by long-term light interaction," thus achieving a more nuanced and realistic portrayal of relationship dynamics.

[0045] The double exponential decay function is: In the formula, β represents the weight that balances the short-term and long-term effects, λ1 and λ2 represent the attenuation coefficients, and γ represents the power law exponent. This indicates a rapid short-term decline, simulating the strong but quickly fading impact of recent private interactions on social intimacy. It represents a slow, long-term decay, simulating a persistent but weak relationship left by long-term interactions.

[0046] Specifically, The trust matrix T calculated in step S2 incorporates historical interactions in its edge weights W, but it does not distinguish the timing of those interactions. A private conversation from yesterday should not contribute the same amount to the current level of intimacy as a conversation from a year ago. Therefore, a separate time-series processing module is needed.

[0047] The single exponential decay model, which assumes that relationship influence decays at a constant rate over time, is inconsistent with complex social relationships, especially those involving privacy. In real-world communication scenarios, social relationship intimacy may follow two patterns: First, recent significant sharing of private information can have a strong but potentially rapidly cooling effect, such as a secret confidant; second, long-term, stable, and casual private interactions can form a slowly decaying base of social intimacy, such as frequently sharing schedules.

[0048] To address the aforementioned issues, this step assigns time-sensitive weights to historical interactions, enabling intimacy calculations to reflect the dynamic evolution of relationships and resolving the "rigid dynamic evaluation" problem of existing methods. In this step, a time-decaying weight is calculated for each historical interaction, ensuring that intimacy reflects the current relationship state. This method abandons the single exponential decay model and proposes a double exponential smooth decay function to more finely characterize the temporal evolution of the impact of privacy interactions.

[0049] ,in, This is the current calculation time. For historical interaction time, To balance the weights of short-term and long-term impacts; This represents a rapid decline in the short term, simulating the strong but quickly fading impact of recent private interactions on social intimacy. It represents a slow, long-term decay, simulating the lasting but weak relationship left by long-term interactions.

[0050] S4. Based on the behavior of privacy information dissemination among users, and combined with the information entropy and dissemination risk model, calculate the sensitivity score for each piece of privacy content disseminated by a user. This study combines information entropy theory with a propagation risk model to calculate the sensitivity of privacy content. It not only considers the uncertainty of content belonging to different privacy categories but also modulates the risks arising from the breadth and depth of its actual propagation using a sigmoid function. This results in a continuous, objective, and computable sensitivity score, providing a key metric for distinguishing the quality differences in privacy information behind everyday scenarios such as "sharing lunch photos" and "confessing medical records."

[0051] The formula for calculating the sensitivity of privacy content is: In the formula, Let r(p) represent the probability distribution of content categories, r(p) represent the dissemination risk score, and k represent the adjustment coefficient.

[0052] Furthermore, the formula for calculating the transmission risk score r(p) is as follows: In the formula, propagation depth represents the number of times the private content is forwarded.

[0053] Specifically, Both the edge weight calculation in step S1 and the trust transfer in step S2 depend on the sensitivity S(p) of the privacy content. Traditional methods using classification labels (such as "high", "medium", "low") are too crude. We need a scientific and computable quantitative method. This step assigns a refined and continuous sensitivity score to each piece of disseminated privacy content, which is the core of distinguishing "privacy information quality".

[0054] The specific process includes: (1) Construct a classification system: Define a set of privacy categories C = {identity information, health information, financial information, location information, family relationships, others}.

[0055] (2) Training the classifier: Train a multi-classification model using labeled text data for any content. Output the probability distribution of each category. .

[0056] (3) Calculate the transmission risk score r(p): Among them, dissemination depth refers to the number of times the content is forwarded.

[0057] (4) Comprehensive Sensitivity Calculation: Combining information entropy theory with propagation risk, a continuous and differentiable sensitivity quantification formula is proposed, overcoming the shortcomings of simple classification. The formula is as follows: Among them, the information entropy part The main purpose is to measure the uncertainty of the probability distribution of privacy content p in a preset category C (such as identity, health, finance, etc.); the risk modulation part of the dissemination. middle This is a propagation risk score that comprehensively considers both the breadth (number of viewers / visibility range) and depth (number of forwards / hops) of propagation. Even if the content itself is not inherently sensitive, if it is accidentally and widely spread, the actual privacy risk it causes will increase dramatically. The sigmoid function compresses the propagation risk score r(p) into the (0,1) interval as a modulation factor. The wider and deeper the content spreads, the greater its potential leakage risk, and the higher the sensitivity score accordingly. The innovation of the comprehensive sensitivity calculation formula lies in multiplying and coupling the "content's own attributes" with the "external propagation effect," which considers both the inherent nature of privacy and its diffusion state in the actual network, achieving a more comprehensive and dynamic sensitivity assessment.

[0058] S5. Based on the directed weighted graph, identify the community to which the user belongs and quantify the user's community intimacy contribution in each community; This step proposes a quantification formula for user community intimacy contribution. This formula calculates a continuous value representing a user's privacy-related influence within the community by analyzing their specific behaviors in disseminating private content (such as initiating or forwarding) and the impact of that content. This allows us to understand a user's online status from the perspective of "what they did" and "how much influence they had" within the community, rather than simply from the perspective of "whether they were present."

[0059] The formula for quantifying a user's contribution to community engagement within each community is as follows: In the formula, τ(i,p) represents the contribution of user i to the spread of privacy content p within community c, and Np represents the breadth of the spread of privacy content p within the community. This represents the logarithmic transformation of the breadth of the spread of private content p within community c; the greater the breadth, the greater the contribution. This indicates the largest contribution from the community to this privacy content.

[0060] Specifically, While step S2 indirectly incorporates network structure information through multi-hop propagation, it doesn't explicitly utilize the strong social prior of "community." Higher default social intimacy typically exists within communities. However, the traditional method of "same community equals bonus" is too crude. We need to go beyond community identification and further differentiate a user's "status" within their community.

[0061] This step delves into the topology of social networks, not only identifying the communities users belong to, but more importantly, quantifying each user's role in privacy dissemination and their level of close contribution within those communities. This step aims to address the shortcomings of existing community correction methods, such as simple weighted formulas, which fail to delve into the quantification within the community itself. We go beyond simple "same community" judgments and propose a "user-community close contribution" quantification method.

[0062] The specific construction process includes: (1) Community detection: The Louvain modularity optimization algorithm is used to detect the network community structure. The community is divided based on the graph G constructed by module 1 to obtain the community structure. The modularity Q is calculated as follows: in, Let the total number of edges be . Here are the elements of the adjacency matrix, and k is the user degree. This is the community indicator function. The purpose of this formula is to measure the quality of community segmentation. A higher Q value indicates a higher edge density and a tighter structure within the community, which is suitable for the "layered" characteristics of privacy dissemination.

[0063] (2) Measuring user community intimacy contribution: We define user In the community Intimacy contribution for: in, On behalf of users Privacy content In the community Internal contributions to the spread (e.g., whether they were the original initiator or a key forwarder), Representative content In the community The logarithmic transformation of the propagation breadth shows that the greater the breadth, the more valuable the contribution. The value representing the largest contribution to the content within the community is primarily used for normalization. This formula transforms a user's behavior within the community (what privacy information was disseminated and how widely it was disseminated) into a continuous value representing their social influence and active role. This provides a detailed basis for subsequent personalized adjustments.

[0064] S6. Multiply and fuse the global trust matrix, time-varying weights, and sensitivity scores to obtain the intimacy of social relationships between users, and correct the intimacy of social relationships based on community intimacy contribution. This paper employs a multiplicative model for basic fusion, emphasizing the combined effect of social relationship strength, timeliness, and content value. Subsequently, a community correction module is used to personalize and enhance basic intimacy based on the similarity of users' contribution vectors across different communities. This correction mechanism allows user pairs with similar roles and matching influence across multiple privacy-related communities to achieve a reasonable increase in intimacy even if their direct interaction data is not prominent, profoundly reflecting the "circle resonance" effect in social networks.

[0065] The intimacy of social relationships between users is obtained by multiplying and fusing the global trust matrix, time-varying weights, and sensitivity scores. The formula is as follows: The community correction formula is: in, This represents the global trust value among users derived from the global trust matrix. The average time-varying weight of historical interactions between the two users Indicates from user To users Average sensitivity score of the disseminated content; δ represents the enhancement coefficient, and η represents the nonlinear adjustment factor; Indicates user and users The inner product of the intimacy contribution vectors in each community measures the user's... and users Similarity and overlap of roles in community networks.

[0066] Specifically, The first five steps generated features characterizing user relationships based on network dissemination, time, content, and structure. Now, these features need to be integrated into a unified score. Simple linear weighting cannot express the complex interactions between features, such as situations with high trust but low content sensitivity, and it is essential to incorporate refined community information. The specific approach for this step is as follows: Figure 2 As shown, specifically: This step is the core of information gathering and decision-making. It organically integrates all the outputs of the preceding modules, first calculating the basic intimacy level, then making personalized adjustments based on community roles, and finally outputting a comprehensive and accurate intermediate intimacy value, including: (1) Basic intimacy calculation: Combine the outputs of steps two, three, and four. For users' intimacy... : ,in, Trust matrix from Module 2 It is the average time-varying weight of the historical interactions between the two parties calculated in Module 3. It is the calculation from Module 4 arrive The average sensitivity of disseminated content. This multiplicative fusion reflects the combined effect of social relationship strength, time freshness, and content value. For example, even with global trust... It's very high, but if the recent interaction weight is high... Very low, or the average sensitivity to the content being disseminated. Very low, so the base intimacy level is... All of these will be significantly suppressed. This aligns with the understanding that "intimacy needs to possess intensity, novelty, and depth simultaneously."

[0067] (2) Structure-aware community modification is an innovation on existing community enhancement methods. The key goal of this modification is to enhance the intimacy between users with similar roles and high contributions within the community network, even without privacy information exchange, because such users are likely to have similar privacy concerns and shared contexts. For example: Users A and C both have high contributions in the "Family Health" community (…). (High value), and both have made moderate contributions to the "Alumni Association" community. Despite the direct interaction between A and C ( While their interactions may not be as frequent as those between A and B, due to the high similarity of their community roles, the final intimacy level between A and C, after adjustment using the community adjustment formula, will be higher. This will be significantly enhanced, more accurately reflecting their potential close ties based on shared community roles.

[0068] Based on the above analysis, we will utilize the refined community roles calculated in step S5. Adjust the base intimacy level: molecular Calculate users and The inner product of the affinity contribution vectors in each community measures the similarity and overlap of the roles of the two individuals in the community network; the denominator is normalized. To enhance the coefficient, This is a non-linear adjustment factor. The correction formula no longer simply adds points based on shared community, but rather weights points based on "what the user did within the community and their influence." Two users active in multiple communities with similar roles will see a significant and reasonable increase in their adjusted intimacy level, even if they don't interact much directly.

[0069] S7. Normalize the corrected social relationship intimacy to generate the final intimacy score.

[0070] The entire process uses adaptive Sigmoid normalization to output the final result, ensuring the comparability and interpretability of the scores.

[0071] The formula is: in, This indicates the self-use parameter; σ represents the current batch data. standard deviation It is a constant; Ultimately, the generated It is a value between 0 and 1, with the closer to 1 representing the user's... For users The closer the social relationship in terms of privacy sharing.

[0072] Specifically, Calculated The distribution of these intimacy values ​​may vary depending on network size and activity. To ensure comparability between intimacy values ​​calculated across different datasets or time points, and to facilitate setting application thresholds (e.g., defining >0.7 as "high intimacy"), normalization is necessary. This involves normalizing the intimacy values ​​output by Module Six, which have varying dimensions and ranges. This is mapped to a unified, well-interpretable standard interval [0,1), generating an intimacy score that can be directly used in downstream applications. .

[0073] The corrected intimacy level is mapped to a standard range for ease of application; the formula is as follows: in, For adaptive parameters, For the current batch The standard deviation is set to automatically adjust the "steepness" of the Sigmoid curve based on the dispersion of the current batch of data. When the data dispersion is large ( When (large), Small size, gentle curve, fine discrimination; concentrated dataset ( Hour, The large size and steep curves provide excellent differentiation. This ensures that the output values ​​effectively differentiate regardless of the data distribution. It is a small constant (e.g., 0.01).

[0074] final, As a value between 0 and 1, the closer it is to 1, the stronger the user's perception. For users The stronger the intimacy of social relationships in terms of privacy sharing, and the better the differentiation, the more directly it can drive applications such as privacy settings and risk alerts.

[0075] In summary, this invention constructs a social relationship intimacy calculation method that combines scenario adaptability, dynamic accuracy, and interpretability through a series of formulas and module designs. It can output quantitative indicators that highly match the intimacy of real-world privacy-sharing social relationships.

[0076] like Figure 1 , 2 As shown, the present invention also provides a social relationship intimacy calculation system under social media privacy propagation, including: a data preprocessing and privacy propagation graph construction module, a multi-hop trust transfer calculation module, a double exponential time-varying decay module, a privacy content sensitivity quantification module, a community structure detection and role quantification module, a multi-dimensional intimacy fusion and correction module, and an intimacy normalization and output module; The data preprocessing and privacy propagation graph construction module is used for: This is used to construct a directed weighted graph for privacy propagation based on the behavior of privacy information dissemination among users. Nodes are defined as users, edges are directed propagation relationships, edge weights are calculated, and privacy propagation is constructed. The multi-hop trust transit calculation module is used for: Connected to the data preprocessing and privacy propagation graph construction module, based on a directed weighted graph, the Personalized PageRank algorithm is improved by introducing a transition matrix with Gaussian decay of path distance to calculate the global trust matrix among users; The dual-exponential time-varying decay module is used for: Connected to the data preprocessing and privacy propagation graph construction module, it calculates time-varying weights for historical interaction behaviors based on the privacy information propagation behavior among users through a double exponential decay function. The privacy content sensitivity quantification module is used for: Connected to the data preprocessing and privacy propagation graph construction module, and combined with the information entropy and propagation risk model, the sensitivity of privacy content is calculated; The community structure detection and role quantification module is used for: Connected to the data preprocessing and privacy propagation graph construction module, the Louvain algorithm is used to detect community structure and quantify users' community intimacy contribution in each community; The multi-dimensional intimacy fusion and correction module is used for: It is connected to the multi-hop trust transfer calculation module, the double exponential time-varying decay module, the privacy content sensitivity quantification module, and the community structure detection and role quantification module, respectively. It calculates the basic intimacy through multiplication fusion and performs community correction on the basic intimacy based on community contribution. The intimacy normalization and output module is used for: It connects with the multi-dimensional intimacy fusion and correction module, and normalizes the corrected intimacy through the adaptive Sigmoid algorithm to generate the final intimacy score.

[0077] Advantages of this invention: This invention provides a complete, multi-layered, coupled framework for calculating intimacy in privacy propagation scenarios, clarifying the technical path from data construction and multi-dimensional feature quantification to fusion correction. Edge weight calculation uses average content sensitivity as a core variable, along with propagation frequency, to define the edge weights of the privacy propagation network, and introduces a topological distance compensation term. In trust iteration, a transition matrix based on custom edge weights and fused path decay is used to achieve refined modeling of trust multi-hop decay propagation. Dual-exponential time-varying decay weights overcome the limitations of single-exponential decay, more accurately simulating the dynamic evolution of privacy relationships. Privacy content comprehensive sensitivity integrates information entropy (intrinsic uncertainty) and propagation risk (spreading effect), achieving continuous and refined quantification. User community intimacy contribution and community correction transcend binary community judgment, achieving personalized intimacy enhancement based on the user's specific behavioral role within the community. The final output intimacy value... It has clear scenario interpretation and can be directly used in practical applications such as high-risk relationship identification and dynamic access control.

[0078] The above are merely preferred embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for calculating the intimacy of social relationships under the privacy dissemination of social media, characterized in that, include: Based on the privacy information dissemination behavior among users, nodes are defined as users, edges are directed propagation relationships, and edge weights are adjusted by combining propagation frequency, average content sensitivity, and topological distance factors to construct a directed weighted graph for privacy propagation. Based on the directed weighted graph, the Personalized PageRank algorithm is improved by introducing a transition matrix with Gaussian decay of path distance and iteratively calculating the global trust matrix containing direct and indirect paths between users. Based on the behavior of privacy information dissemination among users, a double exponential decay function is used for historical interaction behavior to model short-term rapid decay and long-term slow decay modes respectively, and a time-varying weight that decays over time is calculated for each historical interaction behavior among users. Based on the behavior of privacy information dissemination among users, and combined with the information entropy and dissemination risk model, a sensitivity score is calculated for each piece of privacy content disseminated by a user. Based on the directed weighted graph, the community to which the user belongs is identified, and the user's contribution to community intimacy within each community is quantified; The social relationship intimacy between users is obtained by multiplying and fusing the global trust matrix, time-varying weights, and sensitivity scores, and then corrected based on the community intimacy contribution. The corrected social relationship intimacy is normalized to generate the final intimacy score.

2. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that: The formula for calculating the edge weight is: in, Indicates the propagation frequency, Indicates the average sensitivity of the content. Indicates the path length. All are harmonic coefficients, satisfying ,and ; As the core variable, through protrude A key contribution to the intensity of privacy dissemination is the construction of network edge rights that are adapted to different scenarios and centered on content sensitivity.

3. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that: The improved Personalized PageRank algorithm introduces a transition matrix with Gaussian decay of path distance, as shown in the formula: The iterative formula is: in, This indicates the introduction of path distance-based... The Gaussian decay term makes the trust transfer probability higher over short distances than over long distances, simulating efficient trust transfer within tight circles; E represents the initial intimacy matrix, and α represents the damping factor; iteration continues until convergence ||T (k) -T (k-1) || F < We refine the modeling of multi-hop trust decay transmission with increasing circumference by using a Gaussian decay transfer matrix.

4. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that: The double exponential decay function is: Where β represents the weight balancing short-term and long-term effects, λ1 and λ2 represent decay coefficients, and γ represents the power-law exponent. This indicates a rapid short-term decline, simulating the strong but quickly fading impact of recent private interactions on social intimacy. It represents a slow, long-term decay, simulating a persistent but weak relationship left by long-term interactions.

5. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that: The formula for calculating the sensitivity of privacy content is as follows: in, Let r(p) represent the probability distribution of content categories, r(p) represent the dissemination risk score, and k represent the adjustment coefficient.

6. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 5, characterized in that, The formula for calculating the transmission risk score r(p) is as follows: Among them, the spread depth indicates the number of times the private content is forwarded.

7. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that, The formula for quantifying a user's contribution to community intimacy within each community is as follows: Where τ(i,p) represents user i's contribution to the spread of privacy content p within community c, and Np represents the breadth of the spread of privacy content p within the community. This represents the logarithmic transformation of the breadth of the spread of private content p within community c; the greater the breadth, the greater the contribution. This indicates the largest contribution from the community to this privacy content.

8. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that, The social relationship intimacy between users is obtained by multiplying and fusing the global trust matrix, time-varying weights, and sensitivity scores, as shown in the formula: The community correction formula is: in, This represents the global trust value among users derived from the global trust matrix. The average time-varying weight of historical interactions between the two users Indicates from user To users Average sensitivity score of the disseminated content; δ represents the enhancement coefficient, and η represents the nonlinear adjustment factor; Indicates user and users The inner product of the intimacy contribution vectors in each community measures the user's... and users Similarity and overlap of roles in community networks.

9. The method for calculating social relationship intimacy under the privacy dissemination of social media as described in claim 1, characterized in that, The corrected social relationship intimacy is then normalized to generate the final intimacy score, using the following formula: in, This indicates the self-use parameter; σ represents the current batch data. standard deviation It is a constant; Ultimately, the generated It is a value between 0 and 1, with the closer to 1 representing the user's... For users The closer the social relationship in terms of privacy sharing.

10. A system for calculating the intimacy of social relationships under the privacy dissemination of social media, implementing the method for calculating the intimacy of social relationships under the privacy dissemination of social media as described in any one of claims 1 to 9, characterized in that, include: The module includes: data preprocessing and privacy propagation graph construction module, multi-hop trust transfer calculation module, double exponential time-varying decay module, privacy content sensitivity quantification module, community structure detection and role quantification module, multi-dimensional intimacy fusion and correction module, and intimacy normalization and output module. The data preprocessing and privacy propagation graph construction module is used for: This is used to construct a directed weighted graph for privacy propagation based on the behavior of privacy information dissemination among users. Nodes are defined as users, edges are directed propagation relationships, edge weights are calculated, and privacy propagation is constructed. The multi-hop trust transfer calculation module is used for: Connected to the data preprocessing and privacy propagation graph construction module, based on the directed weighted graph, the Personalized PageRank algorithm is improved by introducing a transition matrix with Gaussian decay of path distance to calculate the global trust matrix among users; The dual-exponential time-varying decay module is used for: Connected to the data preprocessing and privacy propagation graph construction module, based on the privacy information propagation behavior between users, time-varying weights are calculated for historical interaction behaviors using a double exponential decay function; The privacy-sensitive content quantification module is used for: Connected to the data preprocessing and privacy propagation graph construction module, and combined with the information entropy and propagation risk model, the sensitivity of privacy content is calculated; The community structure detection and role quantification module is used for: Connected to the data preprocessing and privacy propagation graph construction module, the Louvain algorithm is used to detect community structure and quantify the community intimacy contribution of users in each community; The multi-dimensional intimacy fusion and correction module is used for: It is connected to the multi-hop trust transfer calculation module, the double exponential time-varying decay module, the privacy content sensitivity quantification module, and the community structure detection and role quantification module respectively. It calculates the basic intimacy through multiplication fusion and performs community correction on the basic intimacy based on the community contribution. The intimacy normalization and output module is used for: Connected to the multi-dimensional intimacy fusion and correction module, the corrected intimacy is normalized using an adaptive Sigmoid algorithm to generate the final intimacy score.

Images