Cooperative crowd sensing team recruitment method and system based on multi-view trust

Abstract

The invention provides a cooperative crowd sensing team recruitment method and system based on multi-view trust, and belongs to the technical field of crowd sensing. According to the method, a multi-view trusted network is constructed based on an interaction relationship and attribute data of node users; constructing an extensible trust evaluation model based on the graph attention network, and mining a multi-view trust relationship between node users in the multi-view trust network; obtaining a multi-view trust expectation of a specific task, and integrating a multi-view trust relationship between node users into global trust of the specific task; the method comprises the following steps: constructing a team allocation target function by taking a high global trust value in a specific task as a target, solving the target function by using an alternate iteration solving method to obtain an optimal solution, and obtaining an allocation result of a cooperative crowd sensing team through the optimal solution. According to the method, the optimal team can be recruited for the specific task by integrating a plurality of trust relationships on the basis of remarkably reducing the calculation amount, and each user is ensured to be matched with the optimal collaborator.

Description

Technical Field

[0001] The present invention belongs to the technical field of crowd intelligence perception, and in particular to a collaborative crowd intelligence perception team recruitment method and system based on multi-view trust. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Crowd sensing refers to various perception models in which a group of people share data, extract information, measure, and map phenomena of common interest. It is a key component of emerging IoT applications. In recent years, many crowd sensing applications have emerged, including environmental monitoring, traffic management, healthcare, and public safety. For some complex perception tasks, many users need to form a team to obtain higher quality perception data. This team-based perception paradigm is called collaborative crowd sensing. In collaborative perception tasks, users must cooperate and communicate within their teams to better complete the perception tasks. The successful completion of the task depends not only on the quality of individual task completion, but also on the collaborative efficiency of the team. Therefore, a key to improving the service quality of collaborative crowd sensing applications is the recruitment of collaborative teams.

[0004] In recent years, a large number of studies have been devoted to the field of crowdsensing. Alagha et al. proposed a data-stabilized recruitment system that evaluates employees' interests to maximize information quality. Liao et al. used worker feature information extracted from the platform to divide worker groups into communities based on their reputation, preferences, and activity characteristics, thereby improving the efficiency and quality of crowdsourcing task execution. However, they only focused on recruiting enough users, who were treated as independent individuals to complete tasks. They ignored the importance of collaboration between teams. Stable and lasting cooperative relationships between teams are mainly based on mutual trust. When the task requires collaboration between members, team trust is positively correlated with team performance. Song et al. considered mutual trust and willingness between users to form the optimal collaborative team for collaborative crowdsensing tasks, which can effectively improve service quality. However, they ignored the multiple and heterogeneous trust relationships between team members.

[0005] Collaborative crowd sensing requires recruiting teams to perform complex sensing tasks. The higher the mutual trust between team members, the more active the team collaboration will be, thus improving the service quality of collaborative crowd sensing. However, through the analysis of the above-mentioned prior art, the inventors found that the current team recruitment work mainly considers the trust situation in a single aspect, ignoring the diversity of trust relationships between team members. This results in the inability of the current recruitment method to dynamically adjust team members according to task requirements to recruit collaborators who collaborate well in multiple trust views. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a collaborative crowd intelligence perception team recruitment method and system based on multi-view trust, which can recruit the best team for a specific task by integrating multiple trust relationships on the basis of significantly reducing the amount of calculation and ensure that each user is matched with the best collaborator.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] A first aspect of the present invention provides a collaborative crowd-sensing team recruitment method based on multi-view trust.

[0009] The collaborative crowd-sensing team recruitment method based on multi-view trust includes:

[0010] Obtain the interaction relationship and attribute data of collaborative crowd-sensing node users, and build a multi-view trust network based on the interaction relationship and attribute data;

[0011] Building a scalable trust evaluation model based on graph attention network;

[0012] Using the scalable trust evaluation model, the multi-view trust relationship between node users in the multi-view trust network is mined to obtain the importance between any two node users in different trust views;

[0013] Obtain the multi-view trust expectation of a specific task, and integrate the multi-view trust relationship between node users into a task-specific global trust based on the multi-view trust expectation of the specific task; construct the objective function of team allocation with the goal of having a high global trust value in a specific task, use the alternating iterative solution method to solve the objective function, obtain the optimal solution, and obtain the allocation result of the collaborative crowd perception team through the optimal solution.

[0014] Furthermore, trust subgraph extraction, node annotation, feature learning and trust evaluation are performed based on the scalable trust evaluation model to realize the mining of multi-view trust relationships between node users in a multi-view trust network.

[0015] Furthermore, when labeling nodes, the nodes are sorted according to the one-dimensional Weisfeiler-Lehman algorithm combined with the trust value to obtain the label of each node.

[0016] Furthermore, the node labels are represented in the form of one-hot encoding as the feature vector of the node structural attributes.

[0017] Furthermore, the importance between two nodes in different trust views is captured based on a graph attention network. Specifically, the self-attention layer in the graph attention network is used to assign different importances to different nodes.

[0018] Furthermore, in the process of obtaining multi-view trust expectations for a specific task, each trust view has different importance for different perception tasks, and each perception task is represented as a unique vector, namely, the multi-view trust expectation vector.

[0019] Furthermore, the objective function is solved using an alternating iterative solution method to obtain an optimal solution. Specifically, the direct solution of the objective function is divided into two independent sub-problems for solution.

[0020] A second aspect of the present invention provides a collaborative crowd-sensing team recruitment system based on multi-view trust.

[0021] Collaborative crowd intelligence perception team recruitment system based on multi-view trust, including:

[0022] The multi-view trust network construction module is configured to: obtain the interaction relationship and attribute data of the collaborative crowd-sensing node users, and construct the multi-view trust network based on the interaction relationship and attribute data;

[0023] The trust evaluation module is configured to: construct a scalable trust evaluation model based on a graph attention network; use the scalable trust evaluation model to mine the multi-view trust relationship between node users in the multi-view trust network to obtain the importance between any two node users in different trust views;

[0024] The collaborative team recruitment module is configured to: obtain the multi-view trust expectations of a specific task, integrate the multi-view trust relationships between node users into task-specific global trust according to the multi-view trust expectations of the specific task; construct the objective function of team allocation with the goal of having a high global trust value in a specific task, solve the objective function using an alternating iterative solution method, obtain the optimal solution, and obtain the allocation result of the collaborative crowd intelligence perception team through the optimal solution.

[0025] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the collaborative crowd intelligence perception team recruitment method based on multi-view trust as described in the first aspect of the present invention.

[0026] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the collaborative crowd intelligence perception team recruitment method based on multi-view trust as described in the first aspect of the present invention are implemented.

[0027] One or more of the above technical solutions have the following beneficial effects:

[0028] The present invention provides a collaborative crowd-sensing team recruitment method based on multi-view trust. First, a multi-view trust network is constructed based on the acquired user interaction relationship and attribute data. The multiple trust relationships between users can be described based on the multi-view trust network. Secondly, a scalable trust evaluation model based on a graph attention network is introduced to reduce the amount of memory calculation. Finally, according to the multi-view trust matrix and the requirements of specific tasks, the collaborative team is recruited adaptively. Therefore, the present invention can recruit the best team for a specific task by integrating multiple trust relationships on the basis of significantly reducing the amount of calculation, and ensure that each user is matched with the best collaborator.

[0029] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0031] Figure 1 The present invention is a flowchart of a collaborative crowd-sensing team recruitment method based on multi-view trust in the first embodiment of the present invention.

[0032] Figure 2 Schematic diagram of trust value relationship based on two different trust relationships in Embodiment 1 of the present invention.

[0033] Figure 3 Schematic diagram of the mutual trust evaluation process of the multi-view trust network in the first embodiment of the present invention.

[0034] Figure 4 Schematic diagram of the node marking process of the trust subgraph in the first embodiment of the present invention. DETAILED DESCRIPTION

[0035] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0036] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0037] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0038] The overall idea proposed by the present invention is as follows: The present invention provides a collaborative crowd-sensing team recruitment method based on multi-view trust. First, a multi-view trust network model of collaborative crowd-sensing is established to describe the multiple trust relationships between users. Next, a scalable trust evaluation model based on a graph attention network is designed to mine the multi-view trust relationships between users while significantly reducing the computational consumption. Finally, a collaborative team recruitment method based on graph learning is used for team recruitment. The present invention can recruit the best team for a specific task by integrating multiple trust relationships on the basis of significantly reducing the computational workload, and ensure that each user is matched with the best collaborator.

[0039] Embodiment 1

[0040] This embodiment discloses a collaborative crowd-sensing team recruitment method based on multi-view trust.

[0041] like Figure 1 As shown, the collaborative crowdsensing team recruitment method based on multi-view trust includes:

[0042] Step S1, obtaining the interaction relationship and attribute data of the collaborative crowd-sensing node users, and building a multi-view trust network based on the interaction relationship and attribute data;

[0043] Step S2: construct a scalable trust evaluation model based on graph attention network;

[0044] Using the scalable trust evaluation model, the multi-view trust relationship between node users in the multi-view trust network is mined to obtain the importance between any two node users in different trust views;

[0045] Step S3, obtain the multi-view trust expectation of a specific task, and integrate the multi-view trust relationship between node users into a task-specific global trust according to the multi-view trust expectation of the specific task; construct an objective function for team allocation with the goal of having a high global trust value in a specific task, use an alternating iterative solution method to solve the objective function, obtain the optimal solution, and obtain the allocation result of the collaborative crowd perception team through the optimal solution.

[0046] Through the above method steps of the present invention, it is possible to recruit the best team for a specific task by integrating multiple trust relationships, and ensure that each user is matched with the best collaborator, thereby improving the service quality of collaborative crowd intelligence perception. To facilitate understanding of the technical solution of the present invention, this embodiment further explains and illustrates the above method steps.

[0047] In step S1, the interaction relationship and attribute data of the collaborative crowd-sensing node users are obtained, and a multi-view trust network is constructed based on the interaction relationship and attribute data. Specifically:

[0048] In collaborative crowd-sensing scenarios, tasks may need to be completed at a specific location and time, or require users to have certain expertise. Influenced by factors such as IoT devices and personal preferences, users show multiple aspects of trust in the collaborative process. In this study, this embodiment considers the trust of multiple views from different interactions between users. Before building a multi-view trust network, we first analyze the following: Figure 2 The two trust views (trust related to expertise and trust related to spatial location) are shown to illustrate the multi-view trust relationship between users of collaborative crowdsensing nodes and the importance of building multi-view trust relationships for achieving team recruitment.

[0049] like Figure 2 In the two trust views shown, three potential task completers are considered, denoted as A, B, and C. The trust values ​​between these users are: the professional-related trust value between A and B is 0.2, and the spatial-location-related trust value between A and B is 0.8; the professional-related trust value between A and C is 0.6, and the spatial-location-related trust value between A and C is 0.1. Based on these given trust values, user A needs to decide a partner to collaborate on a specific perception task. Due to the expertise and spatial location involved in the perception task, the choice of collaborator will significantly affect the outcome of the task. For tasks that require spatial accuracy, such as drawing noise maps, the high trust between A and B in spatial location makes their collaboration more effective; on the contrary, for tasks that require technical expertise, such as air quality analysis, the higher professional trust between A and C makes their collaboration more efficient. Therefore, it is an urgent problem to design task recruitment schemes in different perception scenarios based on the multi-view trust relationship between users to improve the service quality of the crowd intelligence perception platform.

[0050] In the collaborative crowdsensing platform, users can establish trust relationships based on their original interactions and various aspects of their own attributes. Each trust aspect provides a different view of the trust interaction between users. Multi-view trust relationships between users are crucial for recruiting collaborative teams.

[0051] In order to describe the multi-view trust relationship between users, this embodiment provides the following definition of the multi-view trust network. Specifically, the multi-view trust network of users in the collaborative crowd intelligence perception platform is modeled as v undirected weighted graphs G = {G (1) ,G (2) ,...,G (v)}. Among them, G (v) =(U,E (v) ,T (v) ) represents the trust network of the vth trust view, U = {u1,u2,...,u n} represents n users who can perform the perception task. The vth trust view G (v) Each node u in i Corresponds to a user; Represents the set of mutual trust relationships in the vth trust view; each edge has a weight value The weight value represents the user u i and u j The trust relationship value in the vth view; T (v) is the mutual trust relationship matrix of users symmetrically in the vth view, Represents the (i,j)th element of the matrix.

[0052] The present invention models a multi-view trust network among users of a collaborative crowd intelligence perception platform to describe multiple trust relationships. By modeling a multi-view trust network, collaborative teams can be adaptively recruited for specific tasks, and combining multiple trust relationships during the recruitment phase is conducive to subsequently allocating the best collaborators to users.

[0053] In step S2, a scalable trust evaluation model is constructed based on a graph attention network; the scalable trust evaluation model is used to mine the multi-view trust relationship between node users in a multi-view trust network to obtain the importance between any two node users in different trust views.

[0054] Specifically, each user in a modern online social network can specify a value to represent the trustworthiness of their direct friends; however, in crowd-sensing platforms, direct interactive relationships between users are usually rare. Evaluating the multi-view trust relationship between a pair of users is crucial for recruiting collaborative teams, while traditional trust evaluation models do not take into account multiple trust relationships between users. In addition, large-scale crowd-sensing platforms require a lot of computing resources to evaluate trust relationships in multiple views. As the scale of the network expands, existing models are not suitable for evaluating multi-view trust relationships in crowd-sensing platforms. In order to mine multi-view trust relationships between users, this embodiment proposes a scalable trust evaluation model based on a graph attention network GAT, which can significantly reduce memory overhead. Figure 3 As shown in Figure 2, the scalable trust evaluation model consists of three main steps: (1) trust subgraph extraction; (2) node labeling; (3) feature learning and trust evaluation. Since we use the same model to evaluate the trust between different views, in the subsequent process G will represent any trust view, and nodes and users represent the same meaning.

[0055] Step S2-1, trust subgraph extraction. Trust is transferable, that is, it can be transferred from one person to another; however, trust will gradually decay as the number of trust path hops increases. Therefore, this embodiment believes that the subgraph of the trust network retains rich trust-related information. For the target trust relationship, this embodiment first extracts a set of fixed K nodes; then, extracts the trust subgraphs from the two target users u and u respectively. i and u j Start with a breadth-first search algorithm to obtain a node set L of size K / 2 i and L j ; Finally, merge the two sets to obtain the final node set, while ensuring that the two node sets are different, that is,

[0056] The extraction of trust subgraphs has two important effects on the model: 1) It is scalable because a fixed-size subgraph of K nodes is extracted for each trust link, avoiding the need to input the entire adjacency matrix of the trust network. 2) It is able to generate similar contextual representations for all trust links, even for different trust graphs. By using the same subgraph extraction method, the model can be generalized to different views of trust networks, users, and trust links.

[0057] Step S2-2, node labeling. Before performing a trust assessment, it is defaulted that only the topological structure of the graph is known, so it is necessary to determine two target nodes and determine the structural importance of each node. Specifically, the nodes can be sorted according to the one-dimensional Weisfeiler-Lehman algorithm combined with the trust value to obtain the label of each node. First, in order to maintain the topological directionality of the target trust relationship, label 1 and label 2 are assigned to the target node; the initial label is assigned to it according to the sum of the trust relationships between all nodes in the subgraph and the target node. Then, a unique character signature is assigned to each node according to the Weisfeiler-Lehman algorithm; for this purpose, the initial character signature of the node can be connected to the initial character signatures of its adjacent nodes in lexicographic order (from small to large) to generate a unique character signature; the next node in the ordered list is the node with the lowest character signature in lexicographic order. Finally, this process is iterated repeatedly until each node is assigned an order, as shown in the following example. Figure 4 As shown (the dotted line in the figure represents the trust value t evaluated for the target node, and the trust value is represented in yellow); the node marking process can be implemented by the algorithm process shown in Table 1.

[0058] In practice, the labels of nodes are represented in the form of one-hot encoding as feature vectors of node structural properties. We extract the adjacency matrix of the subgraph, ensuring that its rows and columns are consistent with the ordered set of nodes. The confidence value to be evaluated is set to -1 to make the value invisible to the model.

[0059] Table 1 Node labeling algorithm

[0060]

[0061] Step S2-3, feature learning and trust evaluation. The graph attention network can operate on graph structure data and use the self-attention layer to assign different importance to different nodes. In collaborative crowd sensing, multi-view trust relationships can be represented by graph data. The importance of other nodes in a node’s neighborhood to the node is different in different trust views, and the importance between any two nodes is different in different trust views. Therefore, this embodiment uses a graph attention network to capture the importance between two nodes in different trust views. First, given a subgraph G(L of the target trust relationship), K ), the embedding representations of the two target nodes are connected; then, the connected embedding representation passes through a fully connected layer; and then through a softmax layer. The trust value between the target user pair is expressed as:

[0062]

[0063] in, and Represents user u i and u j The embedding is the feature after h times of weighted aggregation of neighborhood features. is the concatenation operator, W fc is the weight matrix defined in the fully connected layer, and σ is the softmax function. To learn the model parameters, we use the mean square error loss function as the objective function, which is expressed as:

[0064]

[0065] Through the above step S2, this embodiment designs a scalable multi-view trust evaluation model SGN-MT based on graph attention network for the crowd intelligence perception platform. Based on this model, the computational overhead can be significantly reduced; at the same time, the multi-view trust relationship between any user pairs can be effectively evaluated; thus laying the foundation for the subsequent recruitment of the best team for specific tasks.

[0066] Based on the above steps S1 and S2, this embodiment considers the multi-view trust relationship between users in collaborative team recruitment. In collaborative crowd intelligence perception, there are multiple trust views, which can be expressed as TV = {TV 1 ,TV 2 ,...,TV V}; Moreover, in the various perception scenarios initiated by the requester, there are R different types of perception tasks, denoted as ST = {ST1 ,ST 2 ,...,ST R Each type of task ST R There are unique requirements in terms of time, space, and sensing equipment. Therefore, the importance of each trust view to different perception tasks is different. To represent the vth trust view on task ST R While considering the importance of , we choose to represent each task as a unique vector. The vector corresponding to each task constitutes the multi-view trust expectation vector, that is, This vector reflects the expected importance level of various trust views pre-determined by the task publisher based on the specific requirements and characteristics of the task, and can be provided by the task publisher. The multi-view trust expectation quantifies the importance of each trust view and reflects the publisher's expectation of how each view contributes to task execution.

[0067] The trust network between users is usually sparse. Therefore, there are very few user pairs with clear multi-view trust relationships in collaborative crowdsensing platforms, and the multi-view trust relationships between a large number of users are unknown. For effective collaborative team recruitment, it is crucial to evaluate the multi-view trust values ​​between users without direct interaction.

[0068] In step S3, the multi-view trust expectation of a specific task is obtained, and the multi-view trust relationship between node users is integrated into a task-specific global trust according to the multi-view trust expectation of the specific task; the objective function of team allocation is constructed with the goal of having a high global trust value in the specific task, and the objective function is solved using an alternating iterative solution method to obtain the optimal solution. The allocation result of the collaborative crowd intelligence perception team is obtained through the optimal solution, which can be specifically achieved through the following process:

[0069] Step S3-1, collaborative crowd-sensing team recruitment. Recruit the best team for a specific task by dividing all users into c teams. First, this embodiment proposes a recruitment method based on graph learning, which integrates trust relationships on multiple views to form the best team for a specific task and ensures that each user is matched with the best collaborator; then, an optimization algorithm is proposed to solve the resulting team recruitment problem.

[0070] Step S3-1-1, collaborative crowd intelligence sensing recruitment formulation based on multi-view trust relationship. In order to describe the collaborative team for each task, this embodiment defines the concept of task-specific collaborative team, that is, task-based collaborative team. Specifically, Representative for the mission ST R The number of recruited collaborative teams, where c represents the number of recruited teams.

[0071] After trust evaluation is performed on any pair of users in the crowd sensing platform, a trust matrix set T = {T (1) ,T (2) ,...,T (V)}. In order to form a team that can effectively collaborate to complete specific tasks R The team integrated the multi-view trust relationship between users into a task-specific global trust relationship based on the multi-view trust expectation of a specific task. User i A collaborator can be defined as a i The k most trusted users. Each user is assigned to one of the c teams. i and u j The probability of being assigned to the same team is s ij .

[0072] S∈R n×n represents the team allocation matrix (i.e., user recruitment matrix), which is n×n in size and is used to store the probability that any two users are recruited to the same team; each element s in the team allocation matrix S ij Represents user u i and u j The probability of being assigned to the same team. The team assignment matrix S can be viewed as a graph with n users as nodes. Generally, users with higher global trust values ​​in a task are more likely to be assigned to the same team; therefore, user u i and u j Probability of being assigned to the same team The calculation of can be calculated by solving the following problem:

[0073]

[0074] Among them, s j ∈R n×1 is a vector, i.e., the jth column of the team assignment matrix S, i.e., the probability that user j and all users are recruited into a team; its i-th element is s in the team assignment matrix S ij .s ij is an element in the team assignment matrix S, which is used to represent the probability that users i and j are recruited to the same team. represents the importance of the v-th trust view to the R-th task; T (v)Represents the pairwise trust relationship of all users in the vth trust view. A constraint is introduced into formula (3), that is, the sum of each column of the team assignment matrix S is 1. This constraint ensures that each user is accurately assigned to a team; for each user, we can use formula (3) to assign its collaborators. Ideally, when users are divided into c teams, the graph represented by the team assignment matrix S has exactly c connected components. However, the team assignment matrix S may not have exactly c connected components. In order to achieve ideal collaborator assignment, the probability s in formula (3) ij should be constrained to make the collaborator assignment process an adaptive process to ensure that there are exactly c connected components.

[0075] The team allocation matrix S is non-negative, and its Laplace matrix L has the following important property: the number of connected components in the team allocation matrix S is equal to the number of zeros in the eigenvalues ​​of L, c. According to this property, if the rank restriction rank(L) = nc is satisfied, the users have been divided into c teams and the team allocation matrix S is an ideal collaborator allocation matrix. Therefore, the objective function is:

[0076]

[0077] Formula (4) adaptively assigns the best collaborator to each user, which means that the probability of two users being assigned to the same team will change. Therefore, the team assignment matrix S will keep changing until it has c connected components. The learned team assignment matrix S can be used to directly recruit collaborative teams according to Tarjan's strongly connected component algorithm.

[0078] When the smallest c eigenvectors of L are 0, that is, When (λ i represents the i-th smallest eigenvalue), according to Fan's theorem, we have:

[0079]

[0080] Among them, P T =[p1,p2,...,p n ], P represents the transformation matrix, which is used to minimize the transformed trace of matrix L, p i represents the i-th column of the transformation matrix P; Tr(·) represents the trace operation of the matrix, L = D-[(S T +S) / 2] is the Laplacian matrix of the team allocation matrix S; I∈R c×c is an identity matrix, D is a diagonal matrix, Therefore, according to Fan's theorem, equation (4) can be written as:

[0081]

[0082] in, is the trust matrix T (v) The j-th column of is used to represent the trust relationship between user j and other users in the v-th trust view; (v) Each element in express, It is used to represent the trust relationship between users i and j in the vth trust view; γ is a trade-off parameter. By solving formula (6), the team recruitment matrix S can be learned.

[0083] Step S3-2, solving the team recruitment problem. In order to solve the problem in formula (6), this embodiment designs a method using alternating iterations. Therefore, solving the team recruitment problem can be decomposed into two sub-problems, namely, j -subproblem and P-subproblem.

[0084] Step S3-2-1, solve s j - subproblem. First, fix the variable P; then, update s j ; Finally, formula (6) can be rewritten as:

[0085]

[0086] Since each column of the team assignment matrix S is independent, we can solve each column separately. Let b ij express Then solving formula (7) is equivalent to optimizing the following problem:

[0087]

[0088] The Lagrangian function of formula (8) is expressed as:

[0089]

[0090] Here, η is the Lagrange multiplier.

[0091] According to the Karush-Kuhn-Tucker condition, s j The optimal solution is:

[0092]

[0093] Step S3-2-2, solve the P-subproblem: when s is fixed j When , formula (6) can be written as:

[0094]

[0095] Therefore, formula (11) can be solved by calculating the eigenvector of L.

[0096] The above solution process for the team recruitment problem can be completed by Table 2. Specifically, the algorithm in Table 2 is based on a multi-view trust matrix set, a specific task ST R and the number of teams c as input, the output is the task ST R Specific team division results and the team recruitment matrix S. The overall steps of the solution process can be summarized as follows: first, initialize the matrices S and P; then, according to Fan's theorem, s j and P can be optimized simultaneously; alternately optimize formula (10) and formula (11) until S has c connected components, that is, all users are divided into c teams.

[0097] Table 2 Algorithm for collaborative team recruitment problem

[0098]

[0099] Based on the above solution process, the optimal solution for team recruitment can be obtained. Through the above step S3, this embodiment adaptively recruits collaborative teams according to the multi-view trust matrix and the requirements of specific tasks. Specifically, in step S3, this embodiment effectively integrates multiple trust relationships by proposing a collaborative team recruitment method based on graph learning to recruit the best team for a specific task and ensure that each user is matched with the best collaborator. Therefore, this method optimizes the team recruitment results by assigning adaptive and optimal collaborators to each user.

[0100] Embodiment 2

[0101] This embodiment discloses a collaborative crowd-sensing team recruitment system based on multi-view trust.

[0102] Collaborative crowd intelligence perception team recruitment system based on multi-view trust, including:

[0103] The multi-view trust network construction module is configured to: obtain the interaction relationship and attribute data of the collaborative crowd-sensing node users, and construct the multi-view trust network based on the interaction relationship and attribute data;

[0104] The trust evaluation module is configured to: construct a scalable trust evaluation model based on a graph attention network; use the scalable trust evaluation model to mine the multi-view trust relationship between node users in the multi-view trust network to obtain the importance between any two node users in different trust views;

[0105] The collaborative team recruitment module is configured to: obtain the multi-view trust expectations of a specific task, integrate the multi-view trust relationships between node users into task-specific global trust according to the multi-view trust expectations of the specific task; construct the objective function of team allocation with the goal of having a high global trust value in a specific task, solve the objective function using an alternating iterative solution method, obtain the optimal solution, and obtain the allocation result of the collaborative crowd intelligence perception team through the optimal solution.

[0106] Embodiment 3

[0107] The purpose of this embodiment is to provide a computer-readable storage medium.

[0108] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the method for recruiting a collaborative crowd intelligence perception team based on multi-view trust as described in the first embodiment of the present disclosure.

[0109] Embodiment 4

[0110] The purpose of this embodiment is to provide an electronic device.

[0111] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for recruiting a collaborative crowd intelligence perception team based on multi-view trust are implemented as described in the first embodiment of the present disclosure.

[0112] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part 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 to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0113] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0114] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A collaborative crowd-sensing team recruitment method based on multi-view trust, characterized in that: include: Obtain the interaction relationship and attribute data of collaborative crowd-sensing node users, and build a multi-view trust network based on the interaction relationship and attribute data; Building a scalable trust evaluation model based on graph attention network; Using the scalable trust evaluation model, the multi-view trust relationship between node users in the multi-view trust network is mined to obtain the importance between any two node users in different trust views; Obtain the multi-view trust expectation of a specific task, and integrate the multi-view trust relationship between node users into the global trust of the specific task according to the multi-view trust expectation of the specific task; The objective function of team allocation is constructed with the goal of having a high global trust value in a specific task. The objective function is solved using an alternating iterative solution method to obtain the optimal solution. The allocation result of the collaborative crowd perception team is obtained through the optimal solution.

2. The collaborative crowd intelligence perception team recruitment method based on multi-view trust as claimed in claim 1, characterized in that: Based on the scalable trust evaluation model, trust subgraph extraction, node labeling, feature learning and trust evaluation are performed to realize the mining of multi-view trust relationships between node users in a multi-view trust network.

3. The collaborative crowd intelligence perception team recruitment method based on multi-view trust as claimed in claim 2, characterized in that: When labeling nodes, the nodes are sorted according to the one-dimensional Weisfeiler-Lehman algorithm combined with the trust value to obtain the label of each node.

4. The collaborative crowd intelligence perception team recruitment method based on multi-view trust as claimed in claim 3 is characterized in that: The node labels are represented in the form of one-hot encoding as feature vectors of node structural attributes.

5. The collaborative crowd intelligence perception team recruitment method based on multi-view trust as claimed in claim 2, characterized in that: The importance between two nodes in different trust views is captured based on a graph attention network. Specifically, a self-attention layer in the graph attention network is used to assign different importances to different nodes.

6. The collaborative crowd intelligence perception team recruitment method based on multi-view trust as claimed in claim 1, characterized in that: In the process of obtaining multi-view trust expectations for a specific task, each trust view has different importance for different perception tasks, and each perception task is represented as a unique vector, namely the multi-view trust expectation vector.

7. The collaborative crowd intelligence perception team recruitment method based on multi-view trust as claimed in claim 1, characterized in that: The objective function is solved using an alternating iterative solution method to obtain an optimal solution. Specifically, the direct solution of the objective function is divided into two independent sub-problems for solution.

8. A collaborative crowd-sensing team recruitment system based on multi-view trust, characterized by: include: The multi-view trust network construction module is configured to: obtain the interaction relationship and attribute data of the collaborative crowd-sensing node users, and construct the multi-view trust network based on the interaction relationship and attribute data; The trust evaluation module is configured to: construct a scalable trust evaluation model based on a graph attention network; use the scalable trust evaluation model to mine the multi-view trust relationship between node users in the multi-view trust network to obtain the importance between any two node users in different trust views; The collaborative team recruitment module is configured to: obtain the multi-view trust expectation of a specific task, and integrate the multi-view trust relationship between node users into the global trust of the specific task according to the multi-view trust expectation of the specific task; The objective function of team allocation is constructed with the goal of having a high global trust value in a specific task. The objective function is solved using an alternating iterative solution method to obtain the optimal solution. The allocation result of the collaborative crowd perception team is obtained through the optimal solution.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps in the collaborative crowd intelligence perception team recruitment method based on multi-view trust are implemented as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the collaborative crowd sensing team recruitment method based on multi-view trust are implemented as described in any one of claims 1-7.

Images