A worker recruitment method and system based on social recommendation and trust prediction

By constructing social recommendation and trust prediction models, combined with variable neighborhood taboo search algorithm, the problem of workers recruitment strategies in mobile group intelligence perception systems ignore social attributes, realizing accurate trust prediction and preference calculation, expanding workers' solution space, and improving Task benefits and platform value.

CN118552168BActive Publication Date: 2024-11-22YANTAI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411016895.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-29
Publication Date
2024-11-22
Estimated Expiration
2044-07-29

AI Technical Summary

Technical Problem

The worker recruitment strategy in the existing mobile group intelligence perception system ignores the social attributes between workers, resulting in the task missing out on excellent workers. The existing diffusion model is difficult to capture complex social network relationships and cannot accurately predict trust relationships and worker-task preferences.

Method used

Using a method based on social recommendation and trust prediction, by obtaining workers' social data and task recommendation data, building social recommendation models and trust prediction models, predicting workers-task preference similarity and trust relationships, and socially diffusing based on these models to optimize workers Solve space and task benefits, use variable neighborhood taboo search algorithm to solve the best worker combination.

Benefits of technology

Accurate trust prediction and worker-task preference calculation are achieved, workers' solution space is expanded, task efficiency and platform value are improved, and workers are encouraged to actively participate in task diffusion through social incentive mechanisms.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118552168B_ABST
    Figure CN118552168B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for recruiting workers based on social recommendation and trust prediction, and relates to the cross-technical field of crowd intelligence perception technology and artificial intelligence technology. The method includes social diffusion of tasks based on worker trust relationships to obtain worker solution space for tasks. Under budget constraints, the social influence, preference similarity and task responsiveness of workers in the worker solution space are comprehensively considered, and the worker recruitment problem is modeled as CMCOP. The CMCOP is solved using a variable neighborhood taboo search algorithm to obtain the best combination of workers for the task. Incentive rewards are paid to workers who successfully achieve social recruitment in the existing worker set. The present invention can accurately predict the trust relationship between workers and the worker-task preference score, achieve efficient worker recruitment based on maximizing task benefits, and motivate workers to promote social diffusion of tasks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the cross-technical field of crowd intelligence perception technology and artificial intelligence technology, and in particular to a worker recruitment method and system based on social recommendation and trust prediction. Background Art

[0002] Mobile Crowd Sensing (MCS) is a perception model with workers and task publishers as the core. It aims to use the sensors and computing power of large-scale mobile sensing devices to collect, process and analyze various environmental perception data to complete spatiotemporal perception tasks in the real world. With the characteristics of low cost, high coverage and high flexibility of MCS, the application scope of MCS can cover various perception fields such as environmental monitoring, traffic flow, noise level, meteorological conditions, etc. In MCS, an excellent worker recruitment strategy is an important part to ensure that workers complete tasks efficiently, maximize task benefits and enhance platform value. In the study of mobile crowd sensing worker recruitment, a key issue is to recruit a certain number of workers suitable for completing tasks under a given budget. However, existing recruitment strategies usually only consider the independent space, time and ability attributes of workers, while ignoring the social attributes between workers. This neglect will cause tasks to miss potential excellent workers. In fact, based on social attributes, workers can spread tasks to social friends, thereby expanding the potential worker solution space for task recruitment and providing higher quality worker solutions.

[0003] In order to promote the social diffusion of tasks by users to improve task benefits and increase platform value, MCS faces a series of technical challenges. First, the sparsity of social networks determines that users have relatively few direct trust relationships in social networks. In addition, existing diffusion models are usually based on fixed rules and it is difficult to capture the complex nonlinear relationships in workers' social networks. Therefore, how to establish an accurate trust prediction framework to predict the potential trust relationship between workers and model the social diffusion process of workers to tasks has become the primary challenge faced by the present invention. Secondly, in the past research on MCS task allocation, people often only focused on the interaction between task publishers and workers, and realized worker-task recommendations through a single-modal recommendation network. However, the social network constructed by the interactive behavior between workers can optimize worker-task recommendations and enrich worker-task features. Therefore, how to establish a cross-modal worker-task recommendation framework and reasonably predict the worker-task preference similarity is the second challenge faced by the present invention. Finally, the existing recruitment algorithms are usually based on a fixed set of workers, and the value of the MCS platform is fixed, which is not conducive to the development of the platform. The worker recruitment method based on social diffusion can solve this problem. Therefore, how to model the worker recruitment problem based on task diffusion and design reasonable solutions and dynamic worker incentive mechanisms to motivate workers' social diffusion behavior on tasks and ensure the maximization of task benefits and platform value is the third challenge faced by the present invention. Summary of the invention

[0004] In order to solve the above-mentioned problems, the present invention provides a worker recruitment method and system based on social recommendation and trust prediction, which accurately predicts the trust relationship between workers and worker-task preference calculation, and realizes efficient worker recruitment based on social diffusion, thereby maximizing task benefits and ensuring good platform value.

[0005] In a first aspect, the present invention provides a worker recruitment method based on social recommendation and trust prediction, which adopts the following technical solution:

[0006] A worker recruitment method based on social recommendation and trust prediction, comprising:

[0007] Obtain worker social datasets and task recommendation datasets;

[0008] Build a social recommendation model, use trust data and preference rating data to train the social recommendation model, and predict the worker-task preference similarity.

[0009] Build a trust prediction model, use trust data to train the trust prediction model, and predict the trust relationship between workers;

[0010] Based on the trust relationship between workers, tasks are socially diffused to obtain the worker solution space of tasks; the social influence of workers is updated using the trust prediction model; the worker-task preference similarity is updated using the social recommendation model; the task response ability of workers is calculated based on the distance between workers and tasks and the worker movement speed; the task benefits of worker combinations are calculated under budget constraints, and the worker recruitment problem is modeled as CMCOP;

[0011] The variable neighborhood tabu search algorithm is used to solve CMCOP and obtain the best worker combination for the task.

[0012] Furthermore, the obtaining of the worker social dataset and the task recommendation dataset includes obtaining preference rating data and trust data of the worker social dataset and the task recommendation dataset.

[0013] Furthermore, the social recommendation model is constructed and the trust data and preference rating data are used to train the social recommendation model, including using random embedding and Node2vec pre-trained models to process preference rating data and trust data, obtaining the initial embedding state of tasks and workers, and obtaining worker feature vectors, task feature vectors and multi-layer perceptron weights through task aggregation and worker aggregation based on an attention mechanism, thereby realizing the prediction of preference ratings for task-worker-task.

[0014] Furthermore, the trust prediction model is constructed, and the trust prediction model is trained using the workers' trust data to obtain the workers' trust relationship, including inputting the worker feature vector extracted by social recommendation model training into the trust prediction model as an initial embedding vector, enriching the workers' trust features by fusing task preference features, updating the workers' trust aggregation features and trusted aggregation features through trust aggregation and trusted aggregation, and predicting the trust relationship between workers through a normalized exponential function, and finally obtaining the trust relationship between workers.

[0015] Furthermore, the tasks are socially diffused based on the trust relationship among the workers to obtain the worker solution space of the tasks; the social influence of the workers is updated using the trust prediction model; and the social influence of the workers is updated using the trust prediction model.

[0016] Furthermore, the task benefit of the worker combination is calculated under the budget constraint, and the worker recruitment problem is modeled as CMCOP, including the task benefit calculated by considering the social influence, preference similarity and task responsiveness of the workers in the worker solution space, which is expressed as:

[0017]

[0018]

[0019] in, Indicates being The workers recruited express The number of workers recruited socially in express middle value.

[0020] Furthermore, the variable neighborhood taboo search algorithm is used to solve the CMCOP to obtain the best worker combination for the task, including sorting the workers in the worker solution space in descending order according to the cost performance, searching all sorted workers under the task budget constraint, greedily selecting the workers with the highest cost performance until the budget constraint is not met, and using the greedily selected worker combination as the initial solution, updating it to the current solution, and adding it to the taboo main table, while calculating the task benefit of the current solution; then inputting the initial solution into different heuristic neighborhoods for search to explore new worker solutions, and obtaining the optimal solution after updating and iteration, which is the best worker combination for the task.

[0021] Second, a worker recruitment system based on social recommendation and trust prediction, including:

[0022] A data acquisition module is configured to acquire trust data and preference score data of a worker social dataset and a task recommendation dataset;

[0023] The model building module is configured to build a social recommendation model, train the social recommendation model using trust data and preference score data, and predict the worker-task preference similarity; build a trust prediction model, train the trust prediction model using trust data, and predict the trust relationship between workers;

[0024] The optimization module is configured to perform social diffusion of tasks based on worker trust relationships to obtain worker solution space for tasks; update workers’ social influence using a trust prediction model; update worker-task preference similarity using a social recommendation model; calculate workers’ task response capabilities based on the distance between workers and tasks and workers’ moving speed; calculate the task benefits of worker combinations under budget constraints, and model the worker recruitment problem as a CMCOP;

[0025] The computation module is configured to solve the CMCOP using a variable neighborhood tabu search algorithm to obtain an optimal worker combination for the task.

[0026] In a third aspect, the present invention provides a computer-readable storage medium storing a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device for a method of recruiting workers based on social recommendation and trust prediction.

[0027] In a fourth aspect, the present invention provides a terminal device including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded by the processor and executing the method for recruiting workers based on social recommendation and trust prediction.

[0028] In summary, the present invention has the following beneficial technical effects:

[0029] The present invention discloses a social recommendation framework and a trust prediction framework based on graph neural network. The social recommendation framework processes cross-modal data and realizes worker-task preference prediction. The social prediction model extracts worker trust features and trustee features and independently propagates trust to achieve accurate trust prediction.

[0030] Under budget constraints, this paper aims at the problem of recruiting workers for mobile crowdsensing. It comprehensively considers the social influence, preference similarity and task responsiveness of workers, models the worker recruitment problem as CMCOP, and searches for the approximate optimal worker combination through a variable neighborhood taboo search algorithm.

[0031] The present invention is based on the economic incentive reduction mechanism and adopts a social incentive strategy to attract workers to actively participate in the social diffusion of tasks to obtain potentially higher incentive rewards.

[0032] 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

[0033] Figure 1 is a schematic diagram of a worker recruitment framework based on social recommendation and trust prediction according to Embodiment 1 of the present invention;

[0034] Figure 2 is a schematic diagram of a social recommendation framework according to Embodiment 1 of the present invention;

[0035] Figure 3 is a schematic diagram of a trust prediction framework according to Embodiment 1 of the present invention;

[0036] Figure 4 is an example diagram of social reachability according to Embodiment 1 of the present invention;

[0037] Figure 5 This is an example diagram of a task benefit experiment of different algorithms under different tasks in Example 1 of the present invention;

[0038] Figure 6 This is an example diagram of a task benefit experiment under different budgets using different algorithms in Example 1 of the present invention;

[0039] Figure 7This is an example diagram of a task benefit experiment under different numbers of tasks using different algorithms in Example 1 of the present invention;

[0040] Figure 8 This is an example diagram of an experiment on task benefits and the number of workers in the solution space under different numbers of diffusion layers in Example 1 of the present invention. DETAILED DESCRIPTION

[0041] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0042] Example 1

[0043] In the mobile crowd intelligence perception system, there are three main roles: task publishers, workers, and platforms. Taking the urban noise monitoring project as an example, the task publisher publishes tasks and invites workers to use sensing devices to record the noise levels in different locations. Workers are encouraged to share task information through social networks and invite social friends to join noise monitoring. Through this social diffusion, tasks can be spread rapidly, and the workers' solution space can be expanded to cover more urban areas, thereby improving the quality of workers recruited for tasks. In addition, the accuracy of data collection is determined by the workers' task response time, the social attributes of the tasks, and the similarity of worker-task preferences. Appropriate incentive strategies also affect workers' participation and social diffusion. The lifetime value of workers is higher than the acquisition cost of worker incentives. Therefore, incentivizing more high-quality worker users is crucial to the development of the platform.

[0044] like Figure 1 As shown, the present invention designs a social recommendation framework based on graph attention network and a trust prediction framework based on graph convolutional network. In the social recommendation framework, the concept of hypergraph in graph theory is introduced to model the worker-task recommendation network. The worker feature attributes are updated through task aggregation, and the task features are updated through worker aggregation. Then, the aggregation strategy is optimized through the attention mechanism to obtain the importance features of different input data and improve the accuracy of model prediction. In the trust prediction framework, the initial embedding of the worker is obtained through pre-training of the social recommendation framework. Workers independently propagate trust in different roles as trustors and trustees. Tasks are socially diffused based on the trust relationship of workers to obtain the worker solution space of the task. The social influence of workers in the worker solution space and the worker-task preference similarity are calculated through the social recommendation model and the trust prediction model. Combined with the three-dimensional Euclidean distance between workers and tasks and the worker speed, the task response capability of workers is updated in real time.

[0045] Based on the workers' social influence, preference similarity and task responsiveness, the worker recruitment problem is modeled as CMCOP. This paper proposes a specific variable neighborhood taboo search algorithm to solve CMCOP and obtain the best worker combination for the task. In order to encourage effective exploration and prevent the algorithm from falling into local optimality, heuristic neighborhood search and taboo table are defined. Finally, a social incentive decreasing strategy is adopted to attract workers to actively participate in social diffusion in order to obtain potentially higher incentive rewards.

[0046] Embodiment 1 of the present invention provides a trust prediction and social recommendation framework based on social relations and a variable domain taboo search recruitment method, including a preference calculation stage, a trust prediction stage, a recruitment stage and an incentive stage.

[0047] Preference calculation phase: training the social recommendation model. Using the social recommendation model to predict worker-task preference similarity.

[0048] Trust prediction stage: Train the trust prediction model and use it to predict the trust relationship between workers.

[0049] Recruitment phase: Based on the trust relationship of workers, tasks are socially diffused to obtain the worker solution space of tasks. The preference scores of workers and tasks are predicted by training social recommendation models, and the worker-task preference similarity is updated. The trust relationship of workers is predicted by training trust prediction models, and the social influence of workers is updated. The task response time of workers is calculated based on the three-dimensional Euclidean distance between workers and tasks and the movement speed of workers, and the task response ability of workers is calculated by maximum and minimum normalization. Considering the social influence, preference similarity and task response ability of workers under budget constraints, the worker recruitment problem is modeled as CMCOP. The variable neighborhood taboo search algorithm is used to solve CMCOP and obtain the best worker combination for the task. In order to encourage effective exploration and prevent the algorithm from falling into local optimality, heuristic neighborhood search and taboo table are defined.

[0050] Incentive stage: Count the workers recruited by social networks in the combination of workers recruited by the task, update the worker set, and calculate the current value of the platform. The platform pays incentive rewards to workers who successfully achieve social recruitment in the existing worker set.

[0051] Specifically,

[0052] In step 1, if Figure 2 As shown in the figure, the specific steps for training the recommendation model are:

[0053] Step 1.1: Build an initial model based on the graph attention network, using random embedding and the Node2vec pre-trained model to obtain the initial embedding of the embedding layer tasks and workers.

[0054] Step 1.2: Use task aggregation and worker aggregation to forward diffuse the worker and task feature information.

[0055] Step 1.3: Since the hypergraph contains not only the interactions between tasks and workers , and also includes the task's rating of the worker. Therefore, the present invention considers both the task interaction feature and the rating feature. To formulate the interaction between tasks and workers, The task To the workers Interaction information and preference scores To integrate, The calculation is as follows:

[0056] x j i = g t ( [ e i ⊗ e r i j ] )

[0057] in, represents a vector concatenation operation, express The embedding vector of express The embedding vector of Multilayer Perceptron (MLP) representing task interactions.

[0058] Next, the present invention uses worker aggregation to learn the feature vector of the task Specifically, workers aggregate by considering tasks For workers Interaction information with different weights and preference scores Come and learn Feature representation in a hypergraph. To mathematically represent the worker aggregation, we get the task feature vector , The specific calculation is as follows:

[0059]

[0060] α j i = W 2 ⋅ σ ( W 1 ⋅ [ x j i ⊗ e j ] + b 1 ) + b 2

[0061] in, represents a nonlinear activation function, express and Interacting Attention scores parameterized by a two-layer neural network. express The embedding vector of . , , represents the model trainable parameters, , Represents the model trainable bias.

[0062] Step 1.4: Since the hypergraph contains not only the interactions between workers and tasks , and also includes the preference scores between workers and tasks. Therefore, the present invention considers the worker's preference score characteristics for tasks while considering the task characteristics. To the task Preference rating The interactive features of express, Will arrive Interaction information and preference scores To integrate, The calculation is as follows:

[0063] x i j = g w ( [ e j ⊗ e r j i ] )

[0064] in, express The embedding vector of express The embedding vector of Multilayer Perceptron (MLP) representing worker interactions.

[0065] The feature vector of workers is learned by task aggregation, and the worker feature vector is modeled by using the interaction between workers and tasks in the hypergraph Specifically, task aggregation is done by considering workers Tasks that have been interacted with and workers' preference ratings for these tasks To learn the feature representation of workers in the hypergraph. To mathematically represent the task aggregation, we get the worker feature vector , The calculation of is as follows:

[0066]

[0067] α i j = W 2 ⋅ σ ( W 1 ⋅ [ x i j ⊗ e i ] + b 1 ) + b 2

[0068] in, Indicates workers With the task Interacting Attention score, using a two-layer neural network to parameterize the task attention score .

[0069] Step 1.5: After learning is completed, the feature vectors of workers and tasks are obtained (i.e., and ), concatenate them, and then feed them into the MLP for rating prediction:

[0070] h 0 = [ h i w ⊗ h j t ]

[0071]

[0072]

[0073] in, represents a nonlinear activation function, represents the number of hidden layers, express right Preference score prediction.

[0074] Step 1.6: In order to update the model parameters, the mean square loss function is used as the objective function, which is specifically expressed as:

[0075]

[0076] in, is the number of observed preference ratings, express right The actual preference score.

[0077] In step 2, after the task is released, the feature vectors of the task and the worker are obtained through the trained social recommendation model, and the preference scores of the task and the worker are obtained through MLP and the worker-task preference similarity is calculated, as shown in the following formula:

[0078]

[0079] Since the preference between workers and tasks is undirected during training, the worker-task preference similarity is equal to the task-worker preference similarity, i.e. .

[0080] In step 3, if Figure 3 As shown in the figure, the specific steps for training the trust prediction model are:

[0081] Step 3.1: Input the worker feature vector extracted by social recommendation model training into the trust prediction model as the initial embedding vector, and enrich the worker's trust features by integrating the task preference features. Specifically, the worker's initial trust features h O 0 [ u ] and trusted characteristics h I 0 [ u ] As shown below:

[0082] h O 0 [ u ] = h u w

[0083] h I 0 [ u ] = h u w

[0084] Step 3.2: Due to the asymmetric nature of social trust, workers can assume different roles. The present invention independently considers two types of trust propagation: workers as trustors or trustees. At the same time, the different role characteristics of workers and their label characteristics are considered. right Trust Interaction Feature Vector Interacting with trusted feature vector As shown below:

[0085] A t u v l = h O l − 1 [ v ] ⊗ { W u v l ⋅ ω u v }

[0086] P t v u l = h I l − 1 [ v ] ⊗ { W v u l ⋅ ω v u }

[0087] in, , Represents a trainable weight matrix. Indicates the number of layers of trust propagation. h O l − 1 [ v ] Indicates The worker trust feature vector of the layer. h I l − 1 [ v ] Indicates The trusted feature vector of workers in the layer.

[0088] Based on the different trust roles of workers, the present invention first divides the pairwise trust interactions into two groups: trust aggregation and trusted aggregation. Trust aggregation is to calculate the degree of trust that workers trust other workers. The more workers a worker trusts, the higher the degree of trust the worker obtains. Similarly, trusted aggregation is to calculate the degree to which a worker is trusted by other workers. The more a worker is trusted by other workers, the higher the degree of trust the worker obtains.

[0089] In order to mathematically represent the trust aggregation and the trusted aggregation, we get the worker trust feature vector and the trusted feature vector , and The specific calculation is as follows:

[0090]

[0091]

[0092] in, represents a non-linear activation function. and represents the trainable weight matrix, and represents a trainable bias term, Represents the adjacency matrix.

[0093] Step 3.3: The trust feature vector and of The trusted feature vectors of the layers are connected and input into the fully connected layer, using the Softmax function Trust Relationship Make predictions, The calculation of is as follows:

[0094]

[0095] The model parameters are trained with cross entropy loss as the optimization target to measure the difference between the trust relationship and the true trust relationship. During the entire training process, the Adam optimizer is used to adjust the model parameters to minimize the binary cross entropy loss, thereby achieving the purpose of learning the trust relationship between nodes. The loss function is:

[0096]

[0097]

[0098] represents the number of observed trusts. Represents model parameters. is the weight of the regularization term.

[0099] In step 4, if the worker and workers If the workers are socially reachable and workers The social diffusion probability calculation of is as follows:

[0100]

[0101] in, express and The probability that the predicted trust relationship between them is credible.

[0102] like Figure 4 As shown, in step 5, the present invention uses the characteristics of trust propagation of workers in social networks to model the social diffusion process of tasks under the premise of social accessibility. Specifically, workers can share tasks with their social friends, who then further spread the tasks to their social circles, thus forming a multi-level social network task diffusion model. , through social diffusion of workers’ social networks, a worker solution space consisting of potential trust workers can be constructed , for the task The recruitment offers a wide range of options. Workers' solution space The definition is as follows:

[0103] in, r d ∈ [ 0 , 1 ] A random variable defined as the social diffusion discriminant of participation in the task. express The set of socially reachable workers.

[0104] In step 6, after the task is released, the social influence of the worker is affected by two factors. The first is the number of workers that the worker can socially reach, and the second is the average probability of the worker's social diffusion. It can be understood that the more workers the worker can socially reach and the higher the probability of social diffusion, the greater the social influence of the worker. The social influence calculation of social influence is as follows:

[0105]

[0106] in, Indicates workers The number of workers that are socially reachable.

[0107] In step 7, workers right The preference similarity is calculated by the formula in step 2.

[0108] In step 8, the three-dimensional Euclidean distance is calculated based on the positions of the workers and the tasks. and The Euclidean distance between The calculation is as follows:

[0109]

[0110] implement Recruitment cost The calculation is as follows:

[0111]

[0112] in, represents the distance energy consumption coefficient, Represents the vertical height energy consumption coefficient. Represents the cost factor.

[0113] right Mission responsiveness The calculation is as follows:

[0114]

[0115] in, Indicates workers moving speed. represent The maximum response time of a worker task in It means The shortest response time of a worker's task.

[0116] In step 9, After the release, in the workers' solution space In the calculation, the task benefit is obtained by considering the workers’ social influence, preference similarity and task responsiveness. , The calculation is as follows:

[0117]

[0118]

[0119] in, Indicates being The workers recruited express The number of workers recruited socially in express middle value.

[0120] Under the constraints of the task budget, the goal of this invention is to The optimal combination of workers is recruited for each task, and the task benefit is finally maximized. Formally, the worker recruitment problem based on social diffusion in the present invention can be formulated as CMCOP as follows:

[0121]

[0122] In step 10, the specific steps of solving CMCOP using the variable neighborhood taboo recruitment algorithm to obtain the worker combination that is approximately optimal for the task are:

[0123] Step 10.1 (Greedy Search): Divide the worker's value by the cost to get the worker's cost performance. Then, sort the workers in the worker solution space in descending order according to their cost performance. Under the task budget constraint, search all sorted workers and greedily select the worker with the highest cost performance until the budget constraint is not met. Use the greedily selected worker combination as the initial solution, update it to the current solution, add it to the taboo main table, and calculate the task benefit of the current solution.

[0124] Step 10.2 (Neighborhood Search): Enter the initial solution into different neighborhoods for search to explore new worker solutions. Specifically, there are a total of neighborhoods equal to the number of workers in the initial solution, and the search strategies for different neighborhoods are as follows:

[0125] First neighborhood: Update the initial solution to the current solution of the first neighborhood, and remove the workers included in the current solution from the worker solution space. Next, remove the workers with the highest cost from the current solution to obtain an incomplete solution for the first neighborhood. Then, update the remaining budget, and update the worker solution space while satisfying the remaining budget constraint. Next, adopt a greedy search strategy to select the worker combination with the highest cost-effectiveness to join the incomplete solution of the neighborhood, form the search solution of the first neighborhood and update it to the current solution of the first neighborhood. In order to prevent the algorithm from falling into a local optimum, a taboo subtable for the first neighborhood is designed to record the number of times the incomplete solution of the first neighborhood appears m. When a conflict occurs, the greedy search strategy removes the first m worker elements from the worker solution space to avoid the generation of repeated search solutions.

[0126] Second neighborhood: Update the initial solution to the current solution of the second neighborhood, and remove the workers included in the current solution from the worker solution space. Next, remove the two workers with the highest cost from the current solution to obtain an incomplete solution for the second neighborhood. Then, update the remaining budget and update the worker solution space while satisfying the budget constraint. Next, adopt a greedy search strategy to add the worker combination with the highest cost-effectiveness to the incomplete neighborhood solution, form a search solution for the second neighborhood, and update it to the current solution for the second neighborhood. For conflicts in the second neighborhood, adopt the same strategy as the first neighborhood.

[0127] Nth neighborhood (N is equal to the number of workers in the initial solution): Update the initial solution to the current solution of the Nth neighborhood, and remove the workers included in the current solution from the worker solution space. Next, remove all workers from the current solution to obtain an incomplete solution for the Nth neighborhood. Then, update the remaining budget and update the worker solution space while satisfying the budget constraint. Next, adopt a greedy search strategy to add the most cost-effective worker combination to the incomplete solution of the neighborhood to form a search solution for the Nth neighborhood and update it to the current solution for the Nth neighborhood. For conflicts in the second neighborhood, adopt the same strategy as the first neighborhood. For conflicts in the Nth neighborhood, adopt the same strategy as the first neighborhood.

[0128] For N kinds of neighborhood searches, we set the number of neighborhood iterations k to obtain N*k groups of worker solutions, and calculate the task benefits of worker solutions in different neighborhoods.

[0129] Step 10.3 (Compare and Update):

[0130] The worker solutions with the highest task benefits in different neighborhoods are compared with the current solution and updated. If the task benefit of the worker solution in the neighborhood exceeds the current solution and is not in the taboo main table, , then update the neighborhood worker solution to the current solution and add the updated solution to middle. Used to record the optimal solution explored previously to avoid repeated exploration.

[0131] Step 10.4:

[0132] After the update iteration is completed, the selected current solutions will be organized into a combination of workers to perform the task.

[0133] In step 11, the workers who are recruited through social networking to participate in the task are counted and added to the worker set. The updated worker set is expressed as:

[0134]

[0135] Based on the size of the existing set of workers, the platform pays incentive compensation to workers who successfully achieve social recruitment. The present invention takes into account the dynamic changes in the size of workers when calculating the incentive cost of the platform. For an initially smaller set of workers, the platform provides a higher budget to encourage workers to invite new people to join through social networks. As the number of workers increases, the existing group of workers will naturally encourage new workers to join by leveraging the network effect, so the platform will gradually reduce the social recruitment incentive compensation per unit of workers. In short, if workers can invite more social friends to join the platform earlier, they will receive more incentive income. Specifically, the task After release, workers The dynamic incentive compensation calculation formula is as follows:

[0136]

[0137] in, represents the worker incentive compensation coefficient, Represents the number of workers in the current worker set.

[0138] In order to verify the effectiveness of the method of this embodiment, experiments were conducted using three real datasets: Ciao, Epinions, and Filmtrust, and the experimental results of Ciao were used as a typical example. The above real datasets are derived from existing public data and have been approved by the information owner.

[0139] Variable Neighborhood Taboo Search Algorithm

[0140] Among them, VNTS represents the variable neighborhood taboo search algorithm proposed in the present invention, and the greedy worker value-to-cost ratio (G-PRAM) algorithm, the greedy social diffusion (G-SocialD) algorithm, the population-based evolutionary metaheuristic (MA-RAWR) algorithm, the variable neighborhood taboo search without considering social diffusion (VNTS-SocialF) algorithm and the taboo search recruitment (TSR) algorithm represent the comparison algorithms.

[0141] By comparing the task benefits of different algorithms under different budgets, tasks and number of tasks (e.g. Figure 5 , Figure 6 , Figure 7 As shown in Figure 2, it can be found that the VNTS algorithm outperforms other baseline algorithms in terms of task benefits. This is due to the two main advantages of VNTS:

[0142] 1. Strategic superiority:

[0143] The VNTS algorithm comprehensively considers the balance between worker-task preference similarity, social influence, task responsiveness and the number of recruited workers during the recruitment process, making it outstanding in task efficiency. Specifically, VNTS is superior to the G-PRAM and G-SocialD algorithms in task efficiency. In order to maximize the recruitment of social workers, the G-SocialD algorithm will select workers with the lowest cost to increase the number of workers recruited, but this strategy ignores the impact of worker-task preference similarity and task responsiveness on task efficiency, resulting in the worst performance in task efficiency. Although the G-PRAM algorithm greedily selects workers with the highest value-to-cost ratio, its limitation is that it only selects the current optimal solution at each step, lacks global optimality and backtracking ability, and is prone to fall into the local optimal solution. In addition, the VNTS algorithm obtains more high-quality worker options from the worker solution space of the task through social diffusion before recruitment, making it superior to the TSR and VNTS-SocialF algorithms that do not consider social diffusion in task efficiency.

[0144] 2. Excellent algorithm solving performance:

[0145] VNTS's efficient neighborhood search strategy enables it to find excellent solutions in high-quality search domains. By setting neighborhood taboo tables and conflict strategies, VNTS avoids falling into local optimality, thus outperforming the MA-RAWR algorithm that considers social recruitment in terms of task effectiveness. In the process of exploring the optimal solution, MA-RAWR narrows the search space through a segmented probability extraction strategy, which performs poorly when solving CMCOP because it does not exclude impossible or invalid solutions, resulting in many inferior solutions being explored within a fixed number of iterations. In contrast, VNTS focuses the search on areas where better solutions may be found, improves search efficiency, reduces computational overhead, and prevents falling into local optimality through taboo tables.

[0146] In order to more intuitively compare the number of workers and task benefits of VNTS solution space at different levels of social diffusion, Figure 8 The experiment verified that as the number of social diffusion layers increases, the social diffusion of tasks will bring a larger space of worker solutions to the tasks and provide better potential worker solutions. In addition, the increase in task benefits also verified that the increase in the number of social diffusion layers can improve task benefits, because the number of social workers is an important factor affecting task benefits. Recruiting high-quality social workers can not only improve task benefits, but also increase the value of the platform exponentially as social workers register to perform tasks. At the same time, the strategy based on diminishing incentive rewards promotes the process of social diffusion.

[0147] Extensive experiments are conducted on three real datasets to verify the performance of the proposed variable neighborhood tabu search algorithm. The results show that the variable neighborhood tabu search algorithm outperforms the most advanced baselines in the literature in terms of task efficiency.

[0148] Example 2

[0149] This embodiment provides a worker recruitment system based on social recommendation and trust prediction, including:

[0150] A data acquisition module is configured to acquire trust data and preference score data of a worker social dataset and a task recommendation dataset;

[0151] The model building module is configured to build a social recommendation model, train the social recommendation model using trust data and preference score data, and predict the worker-task preference similarity; build a trust prediction model, train the trust prediction model using trust data, and predict the trust relationship between workers;

[0152] The optimization module is configured to perform social diffusion of tasks based on worker trust relationships to obtain worker solution space for tasks; update workers’ social influence using a trust prediction model; update worker-task preference similarity using a social recommendation model; calculate workers’ task response capabilities based on the distance between workers and tasks and workers’ moving speed; calculate the task benefits of worker combinations under budget constraints, and model the worker recruitment problem as a CMCOP;

[0153] The computation module is configured to solve the CMCOP using a variable neighborhood tabu search algorithm to obtain an optimal worker combination for the task.

[0154] A computer-readable storage medium stores a plurality of instructions, wherein the instructions are suitable for being loaded and executed by a processor of a terminal device, for a worker recruitment method based on social recommendation and trust prediction.

[0155] A terminal device includes a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, wherein the instructions are suitable for being loaded and executed by the processor to implement a worker recruitment method based on social recommendation and trust prediction.

[0156] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A worker recruitment method based on social recommendation and trust prediction, characterized in that: include: Obtain worker social datasets and task recommendation datasets; Build a social recommendation model, use the worker social data set and task recommendation data set to train the social recommendation model, and predict the worker-task preference similarity; Build a trust prediction model, use the workers' social data set to train the trust prediction model, and predict the trust relationship between workers; Based on the trust relationship between workers, tasks are socially diffused to obtain the worker solution space of tasks; the social influence of workers is updated using the trust prediction model; the worker-task preference similarity is updated using the social recommendation model; the task response ability of workers is calculated based on the distance between workers and tasks and the worker movement speed; the task benefits of worker combinations are calculated under budget constraints, and the worker recruitment problem is modeled as CMCOP; The CMCOP is solved using the variable neighborhood tabu search algorithm to obtain the best worker combination for the task; The task benefit of the worker combination is calculated under the budget constraint, and the worker recruitment problem is modeled as CMCOP, including the social influence, preference similarity and task responsiveness of the workers in the worker solution space. The task benefit is calculated and expressed as: in, Indicates that the task The workers recruited express The number of workers recruited socially in express Middle Worker the value of Indicates workers social influence, Indicates workers With the task The preference similarity of Indicates workers To the task Mission responsiveness.

2. A worker recruitment method based on social recommendation and trust prediction according to claim 1, characterized in that: The obtaining of the worker social dataset and the task recommendation dataset includes obtaining trust data and preference score data of the worker social dataset and the task recommendation dataset.

3. A worker recruitment method based on social recommendation and trust prediction according to claim 2, characterized in that: The social recommendation model is constructed, and the social recommendation model is trained using the worker social data set and the task recommendation data set, including constructing an initial model of the social recommendation model based on a graph attention network, using random embedding and a Node2vec pre-trained model to process preference score data and trust data, obtaining the initial embedding state of tasks and workers, obtaining worker feature vectors, task feature vectors and multi-layer perceptron weights through task aggregation and worker aggregation, and then realizing the worker-task preference score prediction to obtain the worker-task preference score.

4. A worker recruitment method based on social recommendation and trust prediction according to claim 3, characterized in that: The trust prediction model is constructed, and the trust prediction model is trained using the worker social data set to predict the trust relationship between the workers, including inputting the worker feature vector extracted by social recommendation model training into the trust prediction model as an initial embedding vector, enriching the worker's trust features by fusing task preference features, updating the worker's trust aggregation features and trusted aggregation features through trust aggregation and trusted aggregation, and predicting the trust relationship between the workers through a normalized exponential function, and finally obtaining the trust relationship between the workers.

5. A worker recruitment method based on social recommendation and trust prediction according to claim 4, characterized in that: The social diffusion of the task based on the trust relationship between the workers to obtain the worker solution space of the task includes using the trust relationship to socially diffuse the socially reachable friends in the worker set to determine the worker solution space of the task; Based on the workers' social networks and trust prediction model, the social influence of workers in the worker solution space of the task is calculated.

6. A worker recruitment method based on social recommendation and trust prediction according to claim 5, characterized in that: The variable neighborhood taboo search algorithm is used to solve the CMCOP to obtain the best worker combination for the task, including sorting the workers in the worker solution space in descending order according to the cost performance, searching all sorted workers under the task budget constraint, greedily selecting the workers with the highest cost performance until the budget constraint is not satisfied, and taking the greedily selected worker combination as the initial solution, updating it to the current solution, and adding it to the taboo main table, while calculating the task benefit of the current solution; then inputting the initial solution into different heuristic neighborhoods for searching to explore new worker solutions, and obtaining the optimal solution after updating and iteration, which is the best worker combination for the task.

7. A worker recruitment system based on social recommendation and trust prediction, executing a worker recruitment method based on social recommendation and trust prediction as claimed in any one of claims 1 to 6, comprising: The data acquisition module is configured to acquire a worker social dataset and a task recommendation dataset; The model building module is configured to build a social recommendation model, train the social recommendation model using the worker social data set and the task recommendation data set, and predict the worker-task preference similarity; Build a trust prediction model, use the workers' social data set to train the trust prediction model, and predict the trust relationship between workers; The optimization module is configured to perform social diffusion of tasks based on the trust relationship between workers to obtain the worker solution space of the task; update the social influence of workers using the trust prediction model; update the worker-task preference similarity using the social recommendation model; calculate the task response ability of workers based on the distance between workers and tasks and the worker's movement speed; Compute the task benefits of worker combinations under budget constraints and model the worker recruitment problem as a CMCOP; The computation module is configured to solve the CMCOP using a variable neighborhood tabu search algorithm to obtain an optimal worker combination for the task.

8. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing a worker recruitment method based on social recommendation and trust prediction as claimed in claim 1.

9. A terminal device, comprising a processor and a computer-readable storage medium, wherein the processor is used to implement each instruction; and the computer-readable storage medium is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the worker recruitment method based on social recommendation and trust prediction as claimed in claim 1.