Mobile crowdsourcing trust evaluation method and system

Abstract

The invention provides a mobile crowdsourcing trust evaluation method and system, and relates to the technical field of mobile crowdsourcing. Comprising the following steps: constructing a hypergraph according to user nodes, working nodes and crowdsourcing tasks; constructing an initial trust evaluation model by using a heterogeneous graph neural network; constructing an optimization loss function by using a cross entropy loss function and a dual-information bottleneck loss function; performing parameter optimization on the initial trust evaluation model by utilizing an optimization loss function and a gradient descent method to obtain a final trust evaluation model; and performing trust prediction on the hypergraph by using the final trust evaluation model to obtain trust prediction values of the user and the working node corresponding to the specific crowdsourcing task. According to the method and the device, the problems that an existing trust relationship prediction scheme based on a graph neural network cannot fully consider environment information and a condition trust relationship under a specific task, so that the scheme cannot adapt to dynamics and diversity in a mobile crowdsourcing task, and the trust prediction accuracy is influenced are solved.

Description

Technical Field

[0001] This application relates to the field of mobile crowdsourcing technology, and in particular to a mobile crowdsourcing trust assessment method and system. Background Technology

[0002] Trust assessment is the process of evaluating trust by quantifying attributes that influence trust. It has numerous applications in sensor networks, social networks, mobile crowdsourcing, and blockchain. Traditional trust assessment schemes are based on heuristic trust theory, typically fusing information from user attributes and user ratings using a predefined trust model to obtain a comprehensive user trust score or trust relationships between users. The effectiveness of such methods heavily relies on the predefined model, lacks universality, and cannot flexibly apply to different scenarios. To address the limitations of traditional trust assessment algorithms, some researchers have applied machine learning algorithms to trust assessment. Machine learning models automate the processing of trust-related data, extracting user trust features and predicting trust relationships between users. Traditional machine learning-based trust assessment algorithms typically rely on matrix factorization theory. They represent the trust relationships between users as a matrix, decompose the matrix to obtain node trust features and trusted features, and then perform matrix multiplication to obtain the complete trust relationship matrix. For example, Fazeli et al. explored the effectiveness of different metric methods in matrix factorization-based trust relationship evaluation and proposed a trust evaluation algorithm based on implicit trust scores. This algorithm incorporates implicit trust scores into the matrix factorization process instead of explicit trust relationships, and experiments demonstrated that the model achieves similar performance even without explicit trust scores. While matrix factorization-based methods have shown good effectiveness, they are based on shallow models and cannot uncover the deep, complex features of nodes, making it difficult to effectively discover the complex nonlinear mapping between user trust features and trust relationships.

[0003] To address the aforementioned problems, some scholars have introduced deep learning models for trust feature extraction and trust relationship prediction. By modeling trust relationships as graphs and processing them using graph convolutional neural networks (GCNNs), these models have achieved remarkable results. For example, Guardian uses GCNNs to process and fuse two types of user trust features separately, constructing a trust relationship prediction scheme based on GCNNs. Building upon this, KGTrust further integrates other social networks besides trust relationship networks, expanding the trust relationship graph and achieving some performance improvement. GATTrust uses an attention mechanism to fuse trust features from different domains to improve trust prediction performance. Furthermore, considering the dynamic nature of trust, a dynamic trust relationship prediction mechanism is constructed based on dynamic graph neural networks. These works effectively solve the trust relationship prediction problem.

[0004] However, the above work has the following problems: Existing trust prediction schemes based on graph neural networks are generally geared towards predicting mutual trust relationships in social networks. These long-term trust relationships cannot adequately reflect the trust relationships between users and work nodes in mobile crowdsourcing, particularly lacking sufficient consideration of environmental information. This is mainly reflected in mobile crowdsourcing, where, compared to the long-term trust relationships between users and work nodes, more attention should be paid to the conditional trust relationships between users and work nodes under specific tasks. For the same work node, differences in its expertise and the type of task lead to different levels of trust from the same user in that work node under different task types. This makes existing trust prediction methods unsuitable for dynamic and environment-dependent mobile crowdsourcing tasks. Existing trust relationship prediction models utilize graph neural networks as feature extractors to extract trust features for trust relationship prediction. However, the extracted trust features typically contain a large amount of redundant information unrelated to trust, which is not conducive to trust relationship prediction and may even introduce noise. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, the purpose of this application is to provide a mobile crowdsourcing trust assessment method and system. This application solves the problem that existing trust relationship prediction schemes based on graph neural networks fail to fully consider environmental information and conditional trust relationships under specific tasks, making them unable to adapt to the dynamism and diversity of mobile crowdsourcing tasks. At the same time, the extracted trust features contain a large amount of redundant information, which affects the accuracy of trust prediction.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a mobile crowdsourcing trust assessment method, including: Construct a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks; An initial trust assessment model was constructed using heterogeneous graph neural networks; An optimized loss function is constructed using the cross-entropy loss function and the double-information bottleneck loss function. The initial trust assessment model is optimized using the optimized loss function and the gradient descent method to obtain the final trust assessment model. By using the final trust assessment model to predict the trust in the hypergraph, the predicted trust values ​​for users and work nodes corresponding to specific crowdsourcing tasks are obtained.

[0007] As a further improvement to this application, the construction of the hypergraph based on user nodes, worker nodes, and crowdsourcing tasks includes: Obtain the text description data of the crowdsourcing task; The task features of the text description data are extracted using a pre-trained text representation model; Construct a hypergraph based on the task characteristics, user nodes, and worker nodes.

[0008] As a further improvement to this application, the expression for the hypergraph is: ; in, It is a set of user nodes. For hypergraphs, It is a set of worker nodes. It is a collection of task nodes; Let be a set of superedges, representing worker nodes. Participating users Released tasks For the set of superedges There exists a set of trust tags. ,and i, j, and k each represent a different number of nodes.

[0009] As a further improvement to this application, the expression for the initial trust assessment model is: ; in, For the initial trust assessment model, User output for the model In the mission Next pair of working nodes The predicted value of trust, For MLP output users eigenvalues, Working nodes for MLP output eigenvalues, for The characteristics are represented.

[0010] As a further improvement to this application, the method of constructing an optimized loss function using the cross-entropy loss function and the dual-information bottleneck loss function includes: Construct a lower bound loss function for the dual-information bottleneck loss function; The target loss function is obtained by reparameterizing the lower bound loss function; The optimized loss function is obtained based on the cross-entropy loss function and the target loss function.

[0011] As a further improvement to this application, the expression for the lower bound loss function is:

[0012] ; in, The number of training samples, Number the training samples; and Variables for user primary characteristics and worker node primary characteristics; It is the first Primary trust characteristics of a trustee; It is the first Primary trust characteristics of each working node; Variables representing higher-order trust characteristics of the trustor. Variables representing higher-order features of the working nodes; Variables representing task characteristics; It is the first Crowdsourcing task Specific characteristics; It is a super-edge variable; For a given number one sample , and Down The conditional probability distribution; This refers to using MLP to analyze the true conditional probability distribution. Approximate; , These are the hyperparameters that balance the loss from the two information bottlenecks; It is given Down The conditional probability distribution; The true distribution of high-level user characteristics Variational approximation; It is given Down The conditional probability distribution; The true distribution of high-level user characteristics Variational approximation; The true distribution of features of high-order working nodes Variational approximation; It is a trust distribution, belonging to the trust label set Y.

[0013] As a further improvement to this application, the expression for the target loss function is:

[0014] ; in, Let be the target loss function. It is an expectation calculation; Refers to the standard normal distribution Upsampling ; This is the formula for calculating the KL divergence; Sampling-based For superedge The result obtained by resampling; Given the sampling results Trust distribution The probability distribution.

[0015] As a further improvement to this application, the expression for the optimized loss function is: ; in, To optimize the loss function, This is the cross-entropy loss function.

[0016] As a further improvement to this application, the expression for the final trust assessment model is as follows: ; in, For the final trust assessment model, For a given task, Initial features extracted using the BERT model For users initial features For working nodes initial features For the model used for primary feature extraction This is an MLP network for high-order trust feature extraction, where y is the trust label. ) represents the true conditional probability distribution The approximation of ) is used to predict the probability distribution of the trust label y based on the subsequent feature extraction results.

[0017] Secondly, this application provides a mobile crowdsourcing trust assessment system, comprising: The hypergraph building module is used to build a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks. The model building module is used to build an initial trust assessment model using heterogeneous graph neural networks; The function building module is used to construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function. The parameter optimization module is used to optimize the parameters of the initial trust evaluation model using the optimization loss function and the gradient descent method to obtain the final trust evaluation model. The trust prediction block is used to predict the trust of the hypergraph using the final trust assessment model, and to obtain the trust prediction values ​​of users and work nodes corresponding to specific crowdsourcing tasks.

[0018] This application discloses the following technical effects: This application provides a trust assessment method for mobile crowdsourcing. By constructing a trust assessment model that combines a hypergraph with a dual variational information bottleneck, this application fully considers environmental information and conditional trust relationships under specific tasks. This solves the problems of insufficient dynamic and diverse adaptability and redundancy of trust features in existing graph neural network-based solutions for mobile crowdsourcing tasks, thereby improving the accuracy of trust prediction. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a mobile crowdsourcing trust assessment method provided in an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of a mobile crowdsourcing trust assessment system provided in an embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 As shown, this application provides a mobile crowdsourcing trust assessment method, including: Step 100: Construct a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks; Step 200: Construct an initial trust assessment model using a heterogeneous graph neural network; Step 300: Construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function; Step 400: Optimize the parameters of the initial trust assessment model using the optimized loss function and gradient descent method to obtain the final trust assessment model; Step 500: Use the final trust assessment model to predict the trust in the hypergraph and obtain the predicted trust values ​​for users and work nodes corresponding to specific crowdsourcing tasks.

[0025] Specifically, this embodiment addresses the problem of user trust assessment of work nodes in mobile crowdsourcing, and is applicable to both centralized and decentralized mobile crowdsourcing systems (such as decentralized SP platforms built using blockchain technology) that include a mobile crowdsourcing platform (Service Provider, SP). The system contains a user set and work nodes. The general operation flow of a mobile crowdsourcing system is as follows: users publish tasks through the SP; the SP evaluates the quality of each work node's completion of the task based on the content, requirements, and budget, predicting the trust in the hypergraph under that task, and using this prediction as the basis for selecting work nodes. The main objective of this embodiment is to complete the trust relationship prediction under a specific task, that is, to complete the trust relationship of the hypergraph under the task.

[0026] This method achieves accurate representation of high-order interactions in mobile crowdsourcing through hypergraph modeling, solving the problem that traditional graph structures cannot cover multi-entity joint decision-making. It significantly improves the model's ability to understand complex task matching processes and enhances the interpretability and accuracy of trust prediction. The process involves: acquiring textual description data of the crowdsourcing task; extracting task features from the textual description data using a pre-trained text representation model; and constructing a hypergraph based on the task features, user nodes, and worker nodes.

[0027] In practical engineering, the text description of mobile crowdsourcing tasks is one of the key factors influencing trust decisions, containing semantic information such as task type, skill requirements, and delivery cycle. Without processing, relying solely on structured fields (such as star ratings and scores) will result in the loss of a significant amount of crucial context. Therefore, this invention extracts the original text description of the task from the platform database, inputs it into a pre-trained text encoder (such as BERT-base-chinese), outputs a 768-dimensional semantic embedding vector, and then compresses it into a 256-dimensional task feature through a fully connected layer. This feature, along with the user and worker nodes, participates in hypergraph construction, forming a ternary hyperedge. In practical applications, Chinese pre-trained models such as ERNIE and RoBERTa can also be used to adapt to localized semantic understanding; this application does not limit this approach.

[0028] Furthermore, the construction of the hypergraph based on user nodes, worker nodes, and crowdsourcing tasks includes: Obtain the text description data of the crowdsourcing task; The task features of the text description data are extracted using a pre-trained text representation model; Construct a hypergraph based on the task characteristics, user nodes, and worker nodes.

[0029] Specifically, pre-trained text representation models (such as BERT) are used for task-related purposes. Detailed text description as input, output Feature representation ; Hypergraph Construction: Constructing heterogeneous graphs with three types of nodes and three types of edges. ;in, It is a set of user nodes. It is a set of worker nodes. It is a collection of task nodes; It is a set of hyperedges, representing worker nodes. Participating users Released tasks For the set of superedges There exists a set of trust tags. ,and ; Attention mechanism for user nodes / worker nodes: Taking user nodes as an example, given a user node... Suppose the set of superedges containing this node is ,make , They are respectively , The features are then used to calculate the update. feature The attention scores are as follows: ; Obtain attention score Then, given the first Layer output is Then the first The layer output is: ; Calculate working node characteristics The method is the same as that of user nodes; Multi-head attention mechanism: Based on the attention mechanism, a multi-head attention mechanism is further designed, that is, it includes... Each independent attention extracts features and performs feature fusion. Assume the first... The features extracted by the attention mechanism are Then the first The layer output is:

[0030] Calculate working node characteristics The method is the same as that of user nodes; When all is completed After feature extraction from the layers, the user node features are output using a Multi-Layer Perception (MLP). Working node characteristics Considering that new task nodes cannot participate in the model training process to obtain task features when evaluating features for new tasks, the original features are used. Used as a feature for trust prediction; The trust output of the multi-MLP is constructed as follows: ; in, For the initial trust assessment model, User output for the model In the mission Next pair of working nodes The predicted value of trust, For MLP output users eigenvalues, Working nodes for MLP output eigenvalues, for The characteristics are represented.

[0031] Furthermore, the construction of the optimized loss function using the cross-entropy loss function and the dual-information bottleneck loss function includes: Construct a lower bound loss function for the dual-information bottleneck loss function; The process of constructing the lower bound loss function of the dual-information bottleneck loss function is as follows: The basic expression for the dual-information loss function is: ,in For mutual information, These are variables representing trust value, user higher-order features, worker node higher-order features, and task features, respectively. and Variables for user primary characteristics and worker node primary characteristics. and To balance the hyperparameters of the two regularization terms.

[0032] Focus on Part 1 ,because , , The corresponding user, worker node, and task can be viewed as a hyperedge. ,by To simplify the description, Expanding, we get:

[0033] ; in Representing variables and The joint probability distribution, It is given back The true conditional probability distribution; yes The true probability distribution.

[0034] make Yes The variational approximation is obtained, and the KL divergence between the two is calculated. Based on the property that the KL divergence is always greater than or equal to 0, we have: ; ; Substitution We can obtain:

[0035] ; yes The information entropy is a fixed value; ignoring this term, we get: ; Focus on Part Two First of all Expanding, we get:

[0036] ; use As The variational approximation of the two can be used to calculate the KL divergence:

[0037] ; Substitution We can obtain:

[0038] ; ; satisfy: ; Further results can be obtained The lower bound,

[0039]

[0040] ; in , which can be regarded as a joint variable of the three, and by using the empirical distribution to replace the true distribution, we can obtain: ; in The number of training samples. For sample index, , For the first Primary feature variables of each sample and their corresponding trust labels.

[0041] The target loss function is obtained by reparameterizing the lower bound loss function; The optimized loss function is obtained based on the cross-entropy loss function and the target loss function.

[0042] Specifically, establish the cross-entropy loss function. ,in The model outputs a trust label. The probability of; Establish a dual-information bottleneck loss function And calculate its lower bound. ; Using loss function Training the network; in, To optimize the loss function, This is the cross-entropy loss function.

[0043] The expression for the final trust assessment model is: ; in, For the final trust assessment model, For a given task, Initial features extracted using the BERT model For users initial features For working nodes initial features For the model used for primary feature extraction This is an MLP network for high-order trust feature extraction, where y is the trust label. ) represents the true conditional probability distribution The approximation of ) is used to predict the probability distribution of the trust label y based on the subsequent feature extraction results.

[0044] More specifically, assuming the user Task posted Extract using BERT Features ; Search and Features and ; Input it into an MLP to obtain predicted trust. ; Furthermore, let's assume four variables. , representing trust relationship, trust characteristics of the trustor, characteristics of the trusted party, and characteristics of the mobile crowdsourcing task, respectively. Then, the trust relationship variable is calculated. and joint variables Mutual Information The lower bound is as follows: ; in , Yes The variational approximation, It is the true conditional probability distribution. These represent the sampling number 1 and 2 respectively. The user node features, worker node features, and crowdsourcing task features contained in each hyperedge sample; calculate The upper realm ,in express The conditional probability distribution, express The probability distribution; calculate The upper realm ,in express The conditional probability distribution, express The probability distribution; get The lower bound:

[0045] ; in, The number of training samples, Number the training samples; and Variables for user primary characteristics and worker node primary characteristics; It is the first Primary trust characteristics of a trustee; It is the first Primary trust characteristics of each working node; Variables representing higher-order trust characteristics of the trustor. Variables representing higher-order features of the working nodes; Variables representing task characteristics; It is the first Crowdsourcing task Specific characteristics; It is a super-edge variable; For a given number one sample , and Down The conditional probability distribution; This refers to using MLP to analyze the true conditional probability distribution. Approximate; , These are the hyperparameters that balance the loss from the two information bottlenecks; It is given Down The conditional probability distribution; The true distribution of high-level user characteristics Variational approximation; It is given Down The conditional probability distribution; The true distribution of high-level user characteristics Variational approximation; The true distribution of features of high-order working nodes Variational approximation; It is a trust distribution, belonging to the trust label set Y.

[0046] Obtaining the final target loss function using reparameterization : ; in, Let be the target loss function. It is an expectation calculation; Refers to the standard normal distribution Upsampling ; This is the formula for calculating the KL divergence; Sampling-based For superedge The result obtained by resampling; Given the sampling results Trust distribution The probability distribution.

[0047] This model supports real-time inference with a single prediction latency of less than 50ms, meeting the needs of online recommendation. It achieves efficient adaptation to new tasks, new users, and new workers, making it suitable for high-frequency crowdsourcing scenarios.

[0048] like Figure 2 As shown, a second objective of this application is to provide a mobile crowdsourcing trust assessment system, based on the aforementioned mobile crowdsourcing trust assessment method, the method comprising: The hypergraph building module is used to build a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks. The model building module is used to build an initial trust assessment model using heterogeneous graph neural networks; The function building module is used to construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function. The parameter optimization module is used to optimize the parameters of the initial trust evaluation model using the optimization loss function and the gradient descent method to obtain the final trust evaluation model. The trust prediction block is used to predict the trust of the hypergraph using the final trust assessment model, and to obtain the trust prediction values ​​of users and work nodes corresponding to specific crowdsourcing tasks.

[0049] A third objective of this application is to provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the aforementioned mobile crowdsourcing trust assessment method. It also includes a communication interface and a bus.

[0050] Among them, the mobile crowdsourcing trust assessment methods include: Step 100: Construct a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks; Step 200: Construct an initial trust assessment model using a heterogeneous graph neural network; Step 300: Construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function; Step 400: Optimize the parameters of the initial trust assessment model using the optimized loss function and gradient descent method to obtain the final trust assessment model; Step 500: Use the final trust assessment model to predict the trust in the hypergraph and obtain the predicted trust values ​​for users and work nodes corresponding to specific crowdsourcing tasks.

[0051] A fourth objective of this application is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned mobile crowdsourcing trust assessment method.

[0052] Among them, the mobile crowdsourcing trust assessment methods include: Step 100: Construct a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks; Step 200: Construct an initial trust assessment model using a heterogeneous graph neural network; Step 300: Construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function; Step 400: Optimize the parameters of the initial trust assessment model using the optimized loss function and gradient descent method to obtain the final trust assessment model; Step 500: Use the final trust assessment model to predict the trust in the hypergraph and obtain the predicted trust values ​​for users and work nodes corresponding to specific crowdsourcing tasks.

[0053] A fifth objective of this application is to provide a computer program product comprising computer instructions that instruct a computer to execute the aforementioned mobile crowdsourcing trust assessment method.

[0054] Among them, the mobile crowdsourcing trust assessment methods include: Step 100: Construct a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks; Step 200: Construct an initial trust assessment model using a heterogeneous graph neural network; Step 300: Construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function; Step 400: Optimize the parameters of the initial trust assessment model using the optimized loss function and gradient descent method to obtain the final trust assessment model; Step 500: Use the final trust assessment model to predict the trust in the hypergraph and obtain the predicted trust values ​​for users and work nodes corresponding to specific crowdsourcing tasks.

[0055] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0056] These computer program instructions can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0057] This application may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, readable storage media, optical storage, etc.) containing computer-usable program code.

[0058] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0059] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort should fall within the scope of protection of this application.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. A mobile crowdsourcing trust assessment method, characterized in that, include: Construct a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks; An initial trust assessment model was constructed using heterogeneous graph neural networks; An optimized loss function is constructed using the cross-entropy loss function and the double-information bottleneck loss function. The initial trust assessment model is optimized using the optimized loss function and the gradient descent method to obtain the final trust assessment model. By using the final trust assessment model to predict the trust in the hypergraph, the predicted trust values ​​for users and work nodes corresponding to specific crowdsourcing tasks are obtained.

2. The mobile crowdsourcing trust assessment method according to claim 1, characterized in that, The construction of the hypergraph based on user nodes, worker nodes, and crowdsourcing tasks includes: Obtain the text description data of the crowdsourcing task; The task features of the text description data are extracted using a pre-trained text representation model; Construct a hypergraph based on the task characteristics, user nodes, and worker nodes.

3. The mobile crowdsourcing trust assessment method according to claim 2, characterized in that, The expression for the hypergraph is: ； in, It is a set of user nodes. For hypergraphs, It is a set of worker nodes. It is a collection of task nodes; Let be a set of superedges, representing worker nodes. Participating users Released tasks For the set of superedges There exists a set of trust tags. ,and i, j, and k all represent different numbers of nodes.

4. The mobile crowdsourcing trust assessment method according to claim 3, characterized in that, The expression for the initial trust assessment model is: ; in, The output of the trust assessment model, User output for trust assessment model In the mission Next pair of working nodes The predicted value of trust, For MLP output users eigenvalues, Working nodes for MLP output eigenvalues, for The feature representation is , and || represents feature concatenation.

5. The mobile crowdsourcing trust assessment method according to claim 1, characterized in that, The construction of the optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function includes: Construct a lower bound loss function for the dual-information bottleneck loss function; The target loss function is obtained by reparameterizing the lower bound loss function; The optimized loss function is obtained based on the cross-entropy loss function and the target loss function.

6. The mobile crowdsourcing trust assessment method according to claim 5, characterized in that, The expression for the lower bound loss function is: ; in, The number of training samples, Number the training samples; and Variables for user primary characteristics and worker node primary characteristics; It is the first Primary trust characteristics of a trustor; It is the first Primary trust characteristics of each working node; Variables representing higher-order trust characteristics of the trustor. Variables representing higher-order features of the working nodes; Variables representing task characteristics; It is the first A crowdsourcing task Specific characteristics; It is a super-edge variable; For a given number one sample , and Down The conditional probability distribution; This refers to using MLP to analyze the true conditional probability distribution. Approximate; , These are the hyperparameters that balance the loss from the two information bottlenecks; It is given Down The conditional probability distribution; The true distribution of high-level user characteristics Variational approximation; It is given Down The conditional probability distribution; The true distribution of high-level user characteristics Variational approximation; The true distribution of features of high-order working nodes Variational approximation; It is a trust distribution, belonging to the trust label set Y.

7. The mobile crowdsourcing trust assessment method according to claim 6, characterized in that, The expression for the target loss function is: ; in, Let be the target loss function. It is an expectation calculation; Refers to the standard normal distribution Upsampling ; This is the formula for calculating the KL divergence; Sampling-based For superedge The result obtained by resampling; Given the sampling results Trust distribution The probability distribution.

8. The mobile crowdsourcing trust assessment method according to claim 7, characterized in that, The expression for the optimization loss function is: ； in, To optimize the loss function, This is the cross-entropy loss function.

9. The mobile crowdsourcing trust assessment method according to claim 1, characterized in that, The expression for the final trust assessment model is: ； in, For the final trust assessment model, For a given task, The initial features extracted using the BERT model; For users initial characteristics, For working nodes initial characteristics, For the model used for primary feature extraction, This is an MLP network for high-order trust feature extraction, where y is the trust label. ) represents the true conditional probability distribution The approximation of ) is used to predict the probability distribution of the trust label y based on the subsequent feature extraction results.

10. A mobile crowdsourcing trust assessment system, characterized in that, include: The hypergraph building module is used to build a hypergraph based on user nodes, worker nodes, and crowdsourcing tasks. The model building module is used to build an initial trust assessment model using heterogeneous graph neural networks; The function building module is used to construct an optimized loss function using the cross-entropy loss function and the double-information bottleneck loss function. The parameter optimization module is used to optimize the parameters of the initial trust evaluation model using the optimization loss function and the gradient descent method to obtain the final trust evaluation model. The trust prediction block is used to predict the trust of the hypergraph using the final trust assessment model, and to obtain the trust prediction values ​​of users and work nodes corresponding to specific crowdsourcing tasks.

Images