Heterogeneous trust-driven social internet of things negative information propagation prediction method and system

Abstract

This invention proposes a heterogeneous trust-driven method and system for predicting the propagation of negative information in the social Internet of Things (IoT), relating to the field of social IoT technology. It includes: constructing user trust networks and device trust networks based on user comment text and device performance indicators in the social IoT, and building a multi-dimensional trust model describing the user-device trust relationship through embedding learning from both the trustor's and trustee's perspectives; establishing a heterogeneous trust-driven information propagation model based on the multi-dimensional trust model and the differences between intra-community and inter-community propagation of negative information; analyzing the negative information propagation process according to the information propagation model to obtain trust prediction results, which are used for trusted device recommendation and high-risk node early warning. This invention improves the accuracy of negative information propagation prediction in the social IoT, providing support for trusted device recommendation and high-risk node early warning.

Description

Technical Field

[0001] This invention relates to the field of social Internet of Things (IoT) technology, and in particular to a method and system for predicting the spread of negative information in social IoT driven by heterogeneous trust. Background Technology

[0002] The Social Internet of Things (SIoT) integrates social relationships with IoT sensing and interaction. Information dissemination is driven by both network topology and trust relationships between nodes. Trust plays a core regulatory role in path selection, dissemination intensity, and range evolution, and is a key factor in ensuring reliable information interaction.

[0003] In scenarios where devices are vulnerable to negative information intrusion such as false alarms and malicious data attacks, accurately depicting the information dissemination patterns driven by trust is of great practical significance for identifying dissemination risks and building security protection mechanisms.

[0004] Existing information dissemination models mostly adopt the homogeneous trust assumption, treating users and devices as a unified trust subject without distinguishing the differences in their trust mechanisms. Furthermore, they only characterize trust from a single structural level, ignoring the differences in trust distribution and interaction characteristics at the community level. This fails to fully reflect the impact of multidimensional trust on the dissemination process, resulting in low accuracy in predicting the scope, speed, and outbreak risk of dissemination, making it difficult to support actual security protection and risk management needs. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes a heterogeneous trust-driven method and system for predicting the spread of negative information in the social Internet of Things (IoT). This method accurately characterizes the patterns of negative information spread in the social IoT and predicts risks, providing a reliable basis for trustworthy device recommendations and security interventions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for predicting the spread of negative information in a social Internet of Things driven by heterogeneous trust, comprising: Based on user comment text and device performance metrics in the social Internet of Things, a user trust network and a device trust network are constructed respectively. Embedding learning is performed from the dual perspectives of the trustor and the trustee to construct a multi-dimensional trust model describing the user-device trust relationship. The social Internet of Things is divided into communities, and the trust differences within and between communities are quantified; Based on a multidimensional trust model and the differences between the propagation of negative information within and between communities, a heterogeneous trust-driven information propagation model is established; the negative information includes false alarms about device infection, misperception data, and malicious data attacks. The negative information propagation process is analyzed based on the information propagation model to obtain trust prediction results, which are used for trusted device recommendations and high-risk node early warning.

[0007] Secondly, the present invention provides a heterogeneous trust-driven social IoT negative information propagation prediction system, comprising: The heterogeneous modeling module is used to construct user trust networks and device trust networks based on user comment text and device performance indicators in the social Internet of Things. It also uses embedding learning from the dual perspectives of trustor and trustee to construct a multi-dimensional trust model that describes the user-device trust relationship. The community segmentation module is used to segment the social Internet of Things into communities and quantify the trust differences within and between communities. The propagation model construction module is used to establish a heterogeneous trust-driven information propagation model based on a multidimensional trust model and the differences between intra-community and inter-community propagation of negative information; the negative information includes false alarm information of device infection, error perception data, and malicious data attacks. The early warning module is used to analyze the negative information propagation process based on the information propagation model to obtain trust prediction results, which are used for trusted device recommendations and high-risk node early warning.

[0008] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the heterogeneous trust-driven social Internet of Things negative information propagation prediction method described in the first aspect.

[0009] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the heterogeneous trust-driven social Internet of Things negative information propagation prediction method described in the first aspect.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, based on the coupling relationship between users and devices in the social Internet of Things (IoT), constructs user trust networks and device trust networks respectively. Embedded learning is performed from the dual perspectives of the trustor and the trustee, fully capturing the multidimensional characteristics and heterogeneous features of trust at the user and device levels, thus improving the accuracy and comprehensiveness of trust representation. By dividing the social IoT trust network into communities, the differences in trust within and between communities can be quantified, more realistically reflecting the distribution patterns and propagation characteristics of trust within the network. The heterogeneous trust-driven information propagation model built upon this foundation organically combines trust heterogeneity with the information propagation process, better reflecting the inherent mechanisms of information diffusion in real-world scenarios. The construction of an evolutionary analysis model enables dynamic analysis of the information propagation process, providing a scientific basis and theoretical support for the design of information control, governance, and trusted propagation mechanisms in the social IoT.

[0011] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0012] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0013] Figure 1 The main flowchart of a heterogeneous trust-driven social IoT negative information propagation prediction method provided in this embodiment of the invention. Detailed Implementation

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] In real-world scenarios of Social Internet of Things (SIoT), users are widely interconnected and interact in real time with various IoT devices, and information such as sensing data, status alarms, and control commands spread rapidly across the network. Whether the information is authentic and reliable, and whether its propagation is controllable, directly affects the security of system operation, while the trust relationship between nodes is a key factor determining the information propagation path, diffusion intensity, and scope of risk.

[0016] In practical applications, trust exhibits a distinct dual characteristic: On the one hand, from the perspective of subject type, the formation mechanisms of user trust and device trust are completely different. User trust relies heavily on historical interaction evaluations, social connections, and subjective preferences, while device trust is determined by objective performance such as operational stability, data accuracy, and communication compliance. Existing models fail to distinguish between the guiding role of user trust based on subjective interaction and the constraining role of device trust based on objective performance. This leads the models to mistakenly regard the two as equivalent propagation nodes, failing to accurately depict the true evolution of negative information in the user-device propagation path, directly causing misidentification of key propagation nodes and misjudgment of risk thresholds.

[0017] On the other hand, from a network structure perspective, nodes within the same community interact frequently and have high trust levels, making it easy for information to spread rapidly in localized areas. Nodes across communities, however, have loose connections and weak trust, directly limiting the spread of information across the entire network. Existing models do not distinguish between these trust differences within and outside communities, often employing a uniform, homogeneous propagation assumption. This leads to models either underestimating the resistance to cross-community propagation or miscalculating the propagation threshold within communities when predicting the dynamics of negative information spread, resulting in systematic biases in the predictions of the information's spread range, speed, and risk level.

[0018] Given that IoT devices are vulnerable to threats from negative information such as false alarms, injection of erroneous perception data, and malicious data attacks, accurately characterizing the information dissemination patterns under these dual differences in trust is crucial for achieving trusted device recommendations, high-risk node identification, and early warning of dissemination risks.

[0019] Most existing information propagation models employ simplified, homogeneous trust assumptions, resulting in a relatively simplistic modeling approach. They typically consider only differences in subject type or community structure, failing to simultaneously account for heterogeneous user-device trust and community-scale trust distribution characteristics. This makes it difficult to fully reproduce the driving and constraining effects of trust on information propagation in real-world networks. Consequently, these models lack accuracy in predicting the speed of negative information propagation, its scope of impact, and the risk of its outbreak. They cannot provide reliable basis for trusted device selection, abnormal node control, and propagation intervention strategies, and thus struggle to meet the actual security management needs of the social Internet of Things.

[0020] To address the aforementioned issues, this invention proposes a heterogeneous trust-driven method, system, medium, and device for predicting the spread of negative information in the social Internet of Things (IoT). Targeting the differentiated mechanisms of user trust and device trust in the social IoT, a heterogeneous trust network is constructed using user comment text and device performance indicators. A dynamic information propagation model is then used to characterize the diffusion pattern of trust within the user-device network, enabling dynamic analysis of negative information propagation and obtaining device trust prediction results. These results are ultimately applied to trusted device recommendations and high-risk node early warning.

[0021] Example 1 like Figure 1As shown, this embodiment discloses a heterogeneous trust-driven method for predicting the spread of negative information in the social Internet of Things, including the following steps: S1: Based on the coupling relationship between users and devices in the social Internet of Things, user trust network and device trust network are constructed respectively, and embedding learning is performed from the dual perspectives of trustor and trustee to obtain a multidimensional heterogeneous trust model that describes the trust heterogeneity between the user layer and the device layer. S2: Divide the social IoT trust network into communities and quantify the trust differences within and between communities; S3: Based on the multidimensional heterogeneous trust model and the trust differences, establish a heterogeneous trust-driven information propagation model; S4: Construct an evolutionary analysis model of the information propagation model to analyze the information propagation process and obtain the dynamic analysis results of information propagation in the social Internet of Things.

[0022] Next, combined Figure 1 This embodiment provides a detailed description of a heterogeneous trust-driven method for predicting the spread of negative information in the social Internet of Things.

[0023] I. Multidimensional Heterogeneous Trust Modeling (a) Network and Problem Definition The social IoT architecture consists of a coupled social layer and an IoT layer. The social layer comprises individual users who establish connections through social interactions, thereby enabling the generation and dissemination of information. The IoT layer consists of smart devices, edge servers, and cloud servers, where there is typically a one-to-one correspondence between individual users and their respective smart devices.

[0024] Since most smart devices have limited computing and storage capabilities, their information processing and interaction often rely on the collaborative support of edge servers and cloud servers. In this process, whether information is received, forwarded, or ignored often depends on the trust relationship between the interacting entities.

[0025] Therefore, in order to describe the information dissemination behavior in the social Internet of Things, it is necessary to formally model the trust relationship at the network structure level.

[0026] To characterize the information dissemination process in the social Internet of Things (IoT), this embodiment constructs a user trust network and a device trust network based on real data, forming a social IoT trust network within a unified framework. The data includes user comment text, device performance metrics, and user / device trust relationship data.

[0027] Among them, user reviews reflect users' subjective evaluation of the device, and a trust score for the device is obtained by quantification.

[0028] Equipment performance indicators include objective attributes such as equipment response time, failure rate, and communication success rate. After normalizing all indicators, a weighted fusion is performed to obtain the objective reliability of the equipment.

[0029] User trust relationship data comes from social interaction behaviors between users, such as information recommendation adoption rate, collaborative operation records, and friend relationships, and is processed to obtain trust scores between users.

[0030] Device trust relationship data comes from historical interaction records between devices, such as communication success rate, data consistency, and task collaboration completion rate, and is processed to obtain a trust score between devices.

[0031] Definition 1: User trust network.

[0032] The user trust network is modeled as a directed graph: .

[0033] in Indicates user, Indicates user trust relationship, Indicates user For users The trust level consists of trust scores between users and trust tendency scores extracted from user comment texts. The comprehensive trust level is obtained by weighted summation, and its value range and weight are determined by the specific application scenario.

[0034] Definition 2: Device Trust Network.

[0035] The device trust network is modeled as a directed graph: .

[0036] in Indicates equipment, Indicates device trust relationship, Indicates device For equipment The trust level is determined by a combination of device performance indicators and trust scores between devices, which are then integrated to obtain the final trust level.

[0037] Definition 3: Social Internet of Things Trust Network.

[0038] Establish a one-to-one correspondence between users and devices (each user is bound to their smart device), and couple the user trust network and the device trust network into an undirected graph: ,in This refers to a social IoT node, where each node is coupled with a user and their smart devices. Indicates the trust relationship between nodes. Represents a node and The trust level between them, this value is a combination of the user-side trust level. Trust level with device side .

[0039] Definition 4: Social IoT Trust Assessment. Given a social IoT trust network... The social IoT trust assessment problem aims to evaluate the trust relationships between nodes that have not directly established a trust relationship, using network structure and known trust levels. Trust levels between ,in .

[0040] (II) Multidimensional Trust Modeling To model the heterogeneity of trust in the entity dimension, a GCN-based SIoT multidimensional trust model is proposed. Its overall framework consists of four modules: an embedding layer, a user / device trust convolutional layer, a prediction layer, and a multidimensional trust fusion layer.

[0041] 1. Embedded layer The embedding layer is used to construct initial feature representations of users and devices, while integrating subjective evaluation information and objective attribute features.

[0042] (1) User embedding: for users The collection of comment texts published on it using Doc2vec Encode the data to obtain the dimension as follows: User embedding : (1) (2) Device embedding: For the device First, use Doc2vec to process the collection of comments it received. Encode the data to obtain the dimension as follows: Comments Embedded : (2) Secondly, the k-dimensional performance index vector of the equipment: Perform Min-Max normalization to obtain the performance embedding. : (3) And embed the performance into a learnable transformation matrix. Projected into the same representation space as the comment embedding: (4) Finally, the comment embedding and the performance embedding are concatenated to obtain the complete device embedding representation. : (5) in, This indicates that the comment is embedded. This indicates performance embedding.

[0043] 2. User / Device Trust Convolutional Layer Trust convolutional layers are designed on both the user trust network and the device trust network to obtain user trust embeddings and device trust embeddings. Considering that each node may act as both a trustor and a trustee, this embodiment jointly models the node embeddings from these two perspectives. The following explanation uses user trust embedding as an example.

[0044] When users When acting as a trustee, assume for right The trust level is first determined by one-hot encoding, and then by transformation matrix. Convert it to the same dimension as the node embedding: (6) Finally, combining neighbor user embedding And trust level embedded modeling right Trust: (7) because There may be multiple trusted neighbors, so mean aggregation is used to aggregate all the information: (8) in and Representing users respectively As a set of neighbors and embeddings of trustors.

[0045] Correspondingly, When acting as a trustee, the process of trust embedding and aggregation is represented as follows: (9) (10) (11) in and Representing users respectively As a set of neighbors and embeddings of the recipient.

[0046] Because the user is portrayed from the perspectives of both the trustor and the trustee. In the trust network, the roles of the two are merged through a fully connected layer, forming the trust embedding of the node at layer l: (12) in These are learnable parameters. This represents a non-linear activation function, where l is the number of trust convolutional layers used to control the range of trust propagation. Specifically, This is the initial embedding obtained from the embedding layer.

[0047] The trust propagation process of a device is the same as that of a user, and the process of obtaining device trust embedding will not be elaborated here.

[0048] 3. Prediction layer For user trust prediction, based on user's... Taking trust prediction as an example, firstly... and The embedded vectors are concatenated, and the learned user embeddings are transformed into a probability distribution of trust relationships through a multilayer perceptron (MLP) and a softmax function, thereby modeling the potential factors of trust levels: (13) in represent right The probability prediction vector for trust level. The specific trust level is... , where index The maximum value of the corresponding vector. Cross-entropy is used as the loss function: (14) in, It is a set of known user pairs and their corresponding trust relationships. This represents all learnable model parameters during the trust prediction process. Control the L2 regularization strength to prevent overfitting. Train the model using the Adam optimizer.

[0049] After training, the model will have the ability to predict the trust level of unknown user pairs: for any user pair By concatenating the embedding vectors of the two and inputting them into the trained MLP and Softmax, the probability distribution of the trust level can be obtained, and the level corresponding to the highest probability is taken as the predicted trust value.

[0050] Accordingly, the device trust prediction process can be described as follows: (15) (16) 4. Multi-dimensional Trust Integration Layer First, based on the above modules, SIoT coupled node pairs with at least one trust rating are... Trust is filled in, and then fused according to trust direction. After training the user trust prediction model and the device trust prediction model, the two are used to fill in the missing trust relationships in the social Internet of Things.

[0051] Specifically, for SIoT coupled node pairs where at least one type of trusted observation exists at the user layer or device layer. ,set up and These represent directed trusts at the user layer and the device layer, respectively. If a trust relationship is missing at the user layer, the user node will be... The embedded vector is input into the trained user trust prediction model to obtain the predicted user trust level. If the device layer trust relationship is missing, then the device node pair will be... The embedded vector is input into the trained device trust prediction model to obtain the predicted device trust level. By using the methods described above, a complete user trust and device trust matrix can be obtained.

[0052] Subsequently, fusion is performed according to the trust direction. Then the coupled nodes... right Directed trust is defined as: (17) in and These represent user trust weight and device trust weight, respectively. Further considering that social IoT nodes frequently need to interact and collaborate, requiring mutual trust, a weighted average method is used to convert directed trust into undirected trust: (18) Considering that traditional methods use fixed values ​​(such as 0 or 0.5) for filling, which ignores node characteristics and network structure information, this embodiment uses the constructed trust model to predict and fill in the missing trust relationships for SIoT coupled node pairs that have at least one type of trust observation in the user layer or device layer, thereby obtaining complete user trust and device trust for further fusion.

[0053] This embodiment constructs separate user trust networks and device trust networks, and extracts embedding features of the two types of trust based on graph convolutional networks. This effectively distinguishes between the guiding role of user trust based on subjective interaction and the constraining role of device trust based on objective performance. Learnable weights are introduced in the fusion stage to achieve adaptive fusion of the two types of trust. This modeling approach solves the problem of existing models treating heterogeneous trusts equally, enabling accurate characterization of the evolution of negative information in the cross-user-device hybrid propagation path. It improves the prediction accuracy of information propagation range, speed, and key propagation nodes, providing a more reliable trust basis for risk warning and security intervention in social IoT platforms.

[0054] (III) Community-based heterogeneous trust mechanism A community-based heterogeneous trust mechanism is introduced. The classic Louvain algorithm is used to identify the community structure in SIoT, and the concepts of "intra-community trust" and "inter-community trust" are proposed.

[0055] The objective function of Louvain's algorithm is to maximize modularity. (19) By maximizing the modularity of the network, automatic identification of community structures in the SIoT trust network is achieved, resulting in higher connection density and trust strength between nodes within a community, while connections between communities are relatively sparse. Based on this division, intra-community trust and inter-community trust can be further defined, thus providing a foundation for characterizing the heterogeneity of trust at different structural scales.

[0056] in, Indicates SIoT node and Undirected trust between them Represents a node Overall trust level This indicates the overall trust level of SIoT. Represents a node The community where it is located express and Are they in the same community?

[0057] To characterize the differences in trust at different community levels and their impact on individual interactions, the following definitions are introduced.

[0058] Definition 1: Trust within a community.

[0059] Intra-community trust refers to the level of trust exhibited by individuals within the same community during interactions, reflecting the strength of trust-driven interactions within the community. Given a community... Trust within a community is defined as the average level of trust between all node pairs within the community. (20) in Indicates community The number of edges in the community. The higher this value, the stronger the trust relationship between nodes within the community, and the more efficient the information dissemination within the community.

[0060] Definition 2: Inter-community trust.

[0061] Inter-community trust refers to the level of trust exhibited by individuals from different communities during interactions, reflecting the strength of cross-community trust-driven interactions. Given two communities... and Since connections between communities are typically sparse, directly using average trust values ​​is insufficient to accurately reflect the level of interaction. Therefore, considering both the degree of connection and trust relationships between communities, inter-community trust is defined as: (twenty one) The higher the value, the closer the trust relationship between different communities, and the greater the risk of negative information spreading across communities.

[0062] By segmenting communities and measuring trust differences, this study accurately characterizes the heterogeneity of trust structures in the social Internet of Things (IoT). Intra-community trust reflects the potential for rapid information dissemination within close-knit groups, while inter-community trust reflects the degree of risk associated with cross-border propagation. Combining these two metrics allows for precise identification of bridging nodes and high-risk propagation paths. This approach addresses the problem of existing models neglecting differences in community structures, providing a structured basis for predicting information dissemination scope, identifying key nodes, and implementing targeted security interventions. This effectively improves the accuracy of risk assessment and the precision of intervention measures.

[0063] II. Heterogeneous Trust-Driven SIoT Information Propagation Model A heterogeneous trust-driven SIoT information propagation model was constructed based on the mean-field method to study the impact of trust mechanisms on the propagation process.

[0064] (a) SIoT Node State Definition Each SIoT node consists of social network users and their owned IoT smart devices. Therefore, social network users and IoT devices are divided into different states. Specifically, social network users are divided into two states: unconscious state (U): the user is unaware of the information transmitted on the device, and conscious state (A): the user is aware of the information transmitted between devices. Devices are divided into three states: susceptible state (S): the device has not received the information but can receive it; infected state (I): the device has received the information and can forward it; and recovered state (R): the device has received the information but no longer forwards it.

[0065] Therefore, four possible states of a social IoT node can be defined: Unconsciously Susceptible (US), Consciously Susceptible (AS), Consciously Infected (AI), and Consciously Recovered (AR). Considering that the user also receives a corresponding message when the device receives information, the user is in a conscious state when the device is in an infected or recovered state, so UI and UR nodes do not exist.

[0066] (II) Modeling of SIoT Information Propagation Driven by Heterogeneous Trust Based on the above node state definition, a heterogeneous trust-driven SIoT information propagation model is proposed. Considering that IoT devices may receive and propagate inaccurate or harmful information, such as malware and false information, this embodiment focuses on the diffusion process of such negative information and constructs an information propagation model based on this.

[0067] Based on the above definition, the information propagation model is described as follows, using... Indicates community The proportion of each state node within the range satisfies Its transmission rate is influenced by trust within the community. Trust between communities The impact of this is shown in Table 1.

[0068] Specifically, given two communities and ,Community Internal information dissemination rate and community and The information transmission rate between people can be calculated as follows: ,in This represents the basic reproduction number. The specific infection process is described below: When a US individual encounters an AI individual, because the user is unconscious, the device is considered to be communicating via the cloud, and two state transitions may occur: one is that the user is aware of the information but the device has not yet received it, resulting in the AS state; the other is that the user is aware of the information and the device has also received it, resulting in the AI ​​state. When a US individual encounters an AI individual within the same community, the probability is... and When becoming AS and AI individuals, and encountering AI individuals from other communities, respectively, the probability is... and Transformed into AS and AI individuals, among which This represents the basic propagation rate.

[0069] Table 1. Transition probabilities of the heterogeneous trust-driven SIoT information propagation model;

[0070] Even when an AS (Autonomous System) individual comes into contact with an AI (AI) individual, they may still become infected and transform into an AI state. However, because the user is in a conscious state, they can utilize local edge devices to accelerate information processing, thereby detecting malicious information earlier and taking appropriate measures to suppress the spread of information on the device. Therefore, a base propagation rate is defined. ,in This indicates the intensity of the measures taken by users; higher intensity results in lower transmission rates. When an AS individual comes into contact with an AI individual within the same community, the probability is used to determine the transmission rate. When becoming an AI individual, and encountering AI individuals from other communities, it is done with probability. To become an AI individual. Furthermore, AI individuals have the potential, with probability... Becoming AR nodes, AS individuals and AR individuals are probabilistically... It becomes a US node.

[0071] Based on the above node state transition relationships, and using mean-field theory to theoretically analyze the information diffusion process, we describe the i-th community. The dynamic differential equation for information propagation is established as follows: (twenty two) The initial conditions of each state node in the network at the start of information propagation are: Belongs to set: in express For all non-negative vectors in the set, for all and All of them have: and .

[0072] This embodiment, by meticulously dividing the consciousness and infection states of nodes and introducing a dual trust mechanism within and between communities, more accurately depicts the diffusion pattern of negative information in a complex social IoT environment; it fully considers the dynamic impact of user consciousness states on information processing and device propagation behavior, and can effectively reveal the regulatory role of trust heterogeneity on propagation paths.

[0073] III. SIoT Information Propagation Evolution Analysis Model In the context of the Social Internet of Things (SIoT), negative information (such as rumors, misleading content, or negative public opinion) can spread rapidly across multiple communities and potentially trigger a chain reaction. Therefore, studying the propagation mechanisms and evolution of negative information is of great significance for predicting propagation trends and developing intervention strategies.

[0074] Based on this, this embodiment systematically analyzes the dynamic characteristics of the constructed model, mainly including key indicators such as the basic regeneration number, equilibrium point and its stability.

[0075] First, the system (22) is simplified to reduce the complexity of the model in order to facilitate the analysis of propagation characteristics. This is done under the assumption that the initial conditions belong to a set. Under the premise that, for all and All of them have: Therefore, system (22) is equivalent to the following dynamical system with dimension 3q: (twenty three) Initial conditions Belongs to set: System (23) can be represented in vector form as follows: (twenty four) Make the following assumptions: Assumption 1: Matrix , and It is not reschedulable, which means that there is a path connecting each pair of communities.

[0076] Assumption 2: This means that the infection can spread within each community.

[0077] This ensures that communities are interconnected and that negative information spreads, which is consistent with reality.

[0078] gather For system (23), it is positively invariant. That is, for all initial values... , All solutions of the time system (23) All kept in the set middle.

[0079] The specific proof is as follows: Consider the boundary... It is sufficient to prove that the system's "direction of motion" at the boundary always points inwards from the set or moves along the boundary to guarantee that it will not leave the set. For any point on the boundary, calculate the derivative of the system at that point and compare it with the outward normal vector of the boundary. If the system "points inwards" or moves along the boundary at the boundary, then the state will not leave the set.

[0080] According to relevant theorems, the boundary consists of the following hyperplanes:

[0081]

[0082]

[0083]

[0084] consider ,in Their outer normal vectors are as follows: make Only the first community is analyzed; the other cases are completely symmetrical.

[0085] Analyze the following scenarios: Scenario 1: ,but:

[0086] Scenario 2: ,but:

[0087] Scenario 3: ,but:

[0088] Scenario 4: ,but:

[0089] Other hyperplanes The proof method is similar and will not be repeated here.

[0090] The above proof ensures that the system's state always remains within a reasonable range and will not result in negative numbers or exceed the total number of people.

[0091] By assuming matrix irreducibility and information propagation, it is proved that the system is positively invariant within a given set, meaning that the state variables always remain within a reasonable range and will not become negative or exceed the total population.

[0092] Clearly, there exists a disease-free equilibrium point in the system. This means that all nodes are in the US state and do not propagate negative information.

[0093] To characterize the propagation capacity of negative information, the classic next-generation matrix method is used to calculate the basic reproduction number. The core idea of ​​this method is to separate the process of "generating new infections" from the process of "state transition or recovery" in the system, thereby measuring the intensity of propagation.

[0094] Furthermore, define the matrix. Then the basic reproduction number of system (23) .

[0095] The specific proof is as follows: In system (23), there are only q state variables related to information infection, that is... However, to facilitate the subsequent proof of the stability of the equilibrium point, the state of system (23) is rearranged as follows: This allows us to write the differential equation as: (25) At the disease-free equilibrium point, linearizing the above system yields the corresponding Jacobian matrix:

[0096]

[0097] Matrix F describes the ability of negative information to spread, representing the spread of newly generated negative information. Matrix V describes the ability of negative information to be eliminated / transferred, representing the gradual disappearance or cessation of the spread of existing negative information.

[0098] Substitute specific And construct the next generation matrix: Define The basic reproduction number is its spectral radius:

[0099] in According to the relevant theorems, we have ,make ,so .

[0100] By defining a disease-free equilibrium point and using the next-generation matrix method to calculate the basic reproduction number, the propagation capacity of negative information is measured.

[0101] To study the future spread trend of negative information, we will analyze the stability of the equilibrium point to determine whether the negative information will continue to spread.

[0102] Furthermore, if Then the informationless equilibrium point of system (23) In the set The situation is locally asymptotically stable, and negative information will not spread. No information balance point It is unstable; even a small amount of negative information can spread and eventually continue to proliferate.

[0103] The specific proof is as follows: It can be proven that matrix V satisfies the definition conditions of a non-singular M-matrix, and .

[0104] According to the relevant theories of nonnegative matrices and M-matrices, when F is a nonnegative matrix and V is a nonsingular M-matrix, matrix F The necessary and sufficient condition for all eigenvalues ​​of V to have negative real parts is: .

[0105] Furthermore, when the basic reproduction number At that time, the system's information-free equilibrium point is locally asymptotically stable; when At that point, the equilibrium point is unstable.

[0106] By analyzing the local stability of the disease-free equilibrium point, it can be known that when When the system is stable, negative information will not spread; when Sometimes it is unstable, and information may continue to spread.

[0107] The matrix and Since the above conditions are met, the following conclusion can be drawn: if Then the disease-free equilibrium point of system (23) In the set The mean is globally asymptotically stable, which further illustrates that: if Not only is the system stable near the equilibrium point, but it also converges to a state of no information regardless of the initial state. That is, no matter the initial state, the system will eventually be free from the propagation of negative information.

[0108] Specifically, the equations in system (23) are rearranged, and let... The system (23) under initial conditions The solution is as follows. According to the relevant theorems, it suffices to prove that for all... All of them have: This means that all individuals are US citizens, and there is no spread of negative information. We will analyze the upper and lower bounds of the US percentage.

[0109] From the first q equations of system (23), we can obtain: To analyze convergence more clearly, a comparison system is constructed to analyze its upper bound. Its dynamic behavior is simpler, but it can "control" the original system. Let the comparison system... It can be obtained that its unique globally asymptotically stable equilibrium point. Therefore, for any ,exist , making all ,have: (26) Therefore, its upper bound can be obtained: (27) set up Then for all From formula (26) and the remaining 2q equations in system (23), we can obtain the following comparative system: (28) set up Then the system (28) can be written as: ,in and These are matrices F and V in state. The value below, the first q components are The remaining 2q components are 0.

[0110] You can choose a sufficiently small one Make Therefore, according to the relevant lemma, the matrix The real parts of all eigenvalues ​​are negative, therefore for any initial conditions ,have: so:

[0111] Therefore, for any ,exist This makes it possible for all ,have: From this and the first q equations of system (23), we can obtain that for all and ,have: Similarly, consider comparing the lower bound of the system analysis: Its globally asymptotically stable equilibrium point is Therefore, for any ,exist This makes it possible for all ,have:

[0112] This means that for any ,have:

[0113] make , obtain the lower realm

[0114] Therefore, the upper bound obtained by combining the previous formula (27) can be obtained as follows: That is, all individuals are ultimately in the US state, with no information transmission.

[0115] Further proof when At this point, the disease-free equilibrium point is globally asymptotically stable, meaning that regardless of the initial state, the system will eventually converge to a state with no information propagation.

[0116] Based on the above propagation evolution analysis, the following quantitative results were obtained: Basic reproduction number : Calculated based on trust networks and community structures ,when Negative information will gradually disappear over time. There is a risk of negative information escalating at this time; Equilibrium point stability: The stability analysis results of the fault-free equilibrium point provide a clear risk threshold for the security protection system. As a condition for triggering an early warning.

[0117] Based on the above analysis results, it can be further applied to the security protection of social IoT platforms to achieve the following functions: Trusted Service Recommendation: Identify highly trustworthy nodes and potentially anomalous nodes based on trust prediction results, and recommend trustworthy devices and services to users; Risk warning: Based on Real-time monitoring triggers an early warning mechanism when the transmission capacity exceeds a threshold; Precise intervention: By combining the characteristics of community structure and trust distribution, key propagation nodes and cross-community bridging nodes are identified, and the platform is guided to adopt targeted intervention strategies, such as information demotion, node isolation, or propagation path control.

[0118] This invention addresses the shortcomings of existing information dissemination models, such as the homogeneous trust assumption, failure to distinguish between user and device trust mechanisms, and neglect of community-level trust characteristics. It constructs a heterogeneous trust-driven method for predicting the spread of negative information in the social Internet of Things (IoT). First, user and device trust networks are constructed based on user comment text and device performance indicators, respectively. A multi-dimensional trust model is established through dual-view embedding learning to accurately characterize the heterogeneous trust relationship between users and devices. Second, the social IoT is divided into communities, quantifying the trust differences within and between communities to compensate for the limitations of a single structural level. Finally, combining the multi-dimensional trust model with community trust differences, the propagation characteristics of negative information within and between communities are distinguished, establishing a heterogeneous trust-driven propagation model that fully reflects the impact of multi-dimensional trust on the propagation process. This invention effectively solves the problem of low prediction accuracy in existing models, accurately predicting the spread, speed, and outbreak risk of negative information, providing reliable technical support for trusted device recommendations, high-risk node early warning, and security protection in real-world scenarios.

[0119] Example 2 This embodiment provides a heterogeneous trust-driven social IoT negative information propagation prediction system, including: The heterogeneous modeling module is used to construct user trust networks and device trust networks based on user comment text and device performance indicators in the social Internet of Things. It also uses embedding learning from the dual perspectives of trustor and trustee to construct a multi-dimensional trust model that describes the user-device trust relationship. The community segmentation module is used to segment the social Internet of Things into communities and quantify the trust differences within and between communities. The propagation model construction module is used to establish a heterogeneous trust-driven information propagation model based on a multidimensional trust model and the differences between intra-community and inter-community propagation of negative information; the negative information includes false alarm information of device infection, error perception data, and malicious data attacks. The early warning module is used to analyze the negative information propagation process based on the information propagation model to obtain trust prediction results, which are used for trusted device recommendations and high-risk node early warning.

[0120] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the heterogeneous trust-driven social IoT negative information propagation prediction method described in Embodiment 1 above.

[0121] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the heterogeneous trust-driven social IoT negative information propagation prediction method described in Embodiment 1 above.

[0122] The steps or modules involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0123] The above description is merely a preferred embodiment of the present invention and is 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 predicting the spread of negative information in a social Internet of Things driven by heterogeneous trust, characterized in that, include: Based on user comment text and device performance metrics in the social Internet of Things, a user trust network and a device trust network are constructed respectively. Embedding learning is performed from the dual perspectives of the trustor and the trustee to construct a multi-dimensional trust model describing the user-device trust relationship. The social Internet of Things is divided into communities, and the trust differences within and between communities are quantified; Based on a multidimensional trust model and the differences between the propagation of negative information within and between communities, a heterogeneous trust-driven information propagation model is established; the negative information includes false alarms about device infection, misperception data, and malicious data attacks. The negative information propagation process is analyzed based on the information propagation model to obtain trust prediction results, which are used for trusted device recommendations and high-risk node early warning.

2. The heterogeneous trust-driven method for predicting the spread of negative information in social IoT as described in claim 1, characterized in that, The construction of a user trust network and a device trust network based on user comment text and device performance metrics in the social Internet of Things (IoT) includes: Users are treated as nodes, trust relationships between users are treated as edges, and user trust levels are treated as edge weights to construct a directed graph as a user trust network; wherein, the user trust level is obtained by fusing trust scores between users with trust tendency scores extracted from user comment text. A directed graph is constructed as a device trust network, with devices as nodes, trust relationships between devices as edges, and device trust levels as edge weights; wherein, the device trust level is obtained by fusing device performance indicators and trust scores between devices.

3. The heterogeneous trust-driven method for predicting the spread of negative information in social IoT as described in claim 1, characterized in that, The multidimensional trust model includes an embedding layer, a user / device trust convolutional layer, a prediction layer, and a multidimensional trust fusion layer. The embedding layer is used to construct the initial feature representations of users and devices, including: encoding the set of comments posted by users to obtain user embeddings; encoding the set of comments received by devices to obtain comment embeddings; and combining the device's performance metric vectors to obtain device embeddings. The user / device trust convolutional layer is used to aggregate and learn the trust relationships of neighboring nodes from the perspectives of both the trustor and the trustee, based on the user trust network and the device trust network respectively, to obtain the user trust embedding and the device trust embedding. The prediction layer is used to concatenate the embedding vectors of target node pairs, and transform them into a probability distribution of trust relationships through a multilayer perceptron and a softmax function to obtain trust prediction results between users and between devices. The multidimensional trust fusion layer is used to perform weighted fusion of user trust prediction results and device trust prediction results for the same coupled node pair to obtain the bidirectional trust of the node pair, and then fuse them into undirected trust between nodes.

4. The heterogeneous trust-driven method for predicting the spread of negative information in social IoT as described in claim 1, characterized in that, The process of dividing the social Internet of Things into communities and quantifying trust differences within and between communities specifically includes: A community detection algorithm is used to divide the social IoT trust network into communities and identify the trust community structure in the network. Based on the partitioning results, the average level of trust value among all nodes within the same community is calculated as the trust value within the community. Based on the partitioning results, and taking into account the connection density and trust relationship between different communities, the average level of trust value between cross-community nodes is calculated as the trust between communities, so as to quantify the trust heterogeneity at the structural scale of the social Internet of Things.

5. The heterogeneous trust-driven method for predicting the spread of negative information in social IoT as described in claim 1, characterized in that, Based on a multidimensional trust model and the differences between intra-community and inter-community dissemination of negative information, a heterogeneous trust-driven information dissemination model is established, specifically including: Construct a composite state space for social IoT nodes, combining user awareness state and device infection state to define multiple node types; Establish a propagation mechanism based on trust differences to distinguish between intra-community propagation and inter-community propagation, so that nodes follow differentiated state transition rules in different propagation scenarios; Introducing the regulatory role of user consciousness on propagation behavior, the enhanced information processing ability of conscious state nodes is quantified as a mechanism to inhibit propagation probability; Based on state partitioning and transition rules, a differential equation model describing the dynamic changes in the proportion of various types of nodes within each community is established using the mean field method. This forms an information propagation model driven by trust within and between communities, which is used to describe the impact of trust heterogeneity on the diffusion path of negative information.

6. The heterogeneous trust-driven method for predicting the spread of negative information in social IoT as described in claim 1, characterized in that, The step of analyzing the negative information propagation process based on the information propagation model to obtain trust prediction results, which are used for trusted device recommendation and high-risk node early warning, specifically includes: Based on the aforementioned information propagation model, the basic reproduction number of negative information propagation is calculated using the next-generation matrix method, and the propagation trend of negative information is determined based on the basic reproduction number. When the basic reproduction number is less than the preset threshold, it is determined that the negative information will gradually disappear. When the basic reproduction number exceeds a preset threshold, it is determined that negative information poses a risk of spreading, triggering an early warning mechanism; Based on the user trust embedding and device trust embedding in the multidimensional trust model, high-trust nodes with trust levels above a preset threshold and abnormal nodes with trust levels below a preset threshold are identified. Devices corresponding to the high-trust nodes are recommended as trusted devices, and abnormal nodes are warned as high-risk nodes.

7. A heterogeneous trust-driven social IoT negative information propagation prediction system, characterized in that, include: The heterogeneous modeling module is used to construct user trust networks and device trust networks based on user comment text and device performance indicators in the social Internet of Things. It also uses embedding learning from the dual perspectives of trustor and trustee to construct a multi-dimensional trust model that describes the user-device trust relationship. The community segmentation module is used to segment the social Internet of Things into communities and quantify the trust differences within and between communities. The propagation model construction module is used to establish a heterogeneous trust-driven information propagation model based on a multidimensional trust model and the differences between intra-community and inter-community propagation of negative information; the negative information includes false alarm information of device infection, error perception data, and malicious data attacks. The early warning module is used to analyze the negative information propagation process based on the information propagation model to obtain trust prediction results, which are used for trusted device recommendations and high-risk node early warning.

8. The heterogeneous trust-driven social IoT negative information propagation prediction system as described in claim 7, characterized in that, The construction of a user trust network and a device trust network based on user comment text and device performance metrics in the social Internet of Things (IoT) includes: Users are treated as nodes, trust relationships between users are treated as edges, and user trust levels are treated as edge weights to construct a directed graph as a user trust network; wherein, the user trust level is obtained by fusing trust scores between users with trust tendency scores extracted from user comment text. A directed graph is constructed as a device trust network, with devices as nodes, trust relationships between devices as edges, and device trust levels as edge weights; wherein, the device trust level is obtained by fusing device performance indicators and trust scores between devices.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the heterogeneous trust-driven social IoT negative information propagation prediction method as described in any one of claims 1-6.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the heterogeneous trust-driven social IoT negative information propagation prediction method as described in any one of claims 1-6.

Images