True value mining method for multi-source unstructured text service in meta-universe crowdsourcing environment

Abstract

The invention discloses a truth value mining method for a multi-source unstructured text service in a meta-universe crowdsourcing environment, and provides a solution of three stages, namely a semantic preprocessing stage, a high-dimensional content vector representation of a crowdsourcing answer is constructed, and clustering is carried out; in the truth value optimization and feature generation stage, the optimized truth value mining model is used for evaluating the work quality of crowdsourcing workers, and features are dynamically generated. And a task clustering and class cluster mapping stage: clustering the tasks with generated features, accurately estimating the real class of each task, and finally determining a true value. According to the method, the accuracy and efficiency of true value mining are improved, and the applicability of the method in a complex unstructured data environment is expanded. By integrating the multi-source sensing data in the meta-universe crowdsourcing environment, the intelligent service meeting the user requirements can be accurately pushed, and the service quality and the user experience of the meta-universe are improved.

Description

Technical Field

[0001] The invention relates to a truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment, and belongs to the field of metaverse and crowdsourcing technology. Background Art

[0002] In recent years, the emergence of crowdsourcing platforms has created a distributed way of solving problems. Requesters publish their problems in the form of tasks to crowdsourcing platforms, and crowdsourcing platforms are responsible for distributing these tasks to workers. Since the data provided by different workers may conflict, crowdsourcing platforms often use truth mining methods to resolve data conflicts and mine truth values ​​in order to provide accurate answers to requesters.

[0003] With the rise of the concept of the metaverse, crowdsourcing platforms have been further expanded and applied in virtual environments. The metaverse contains a large amount of multi-source perceptual data generated by users, which includes both numerical data and categorical data, and increasingly involves unstructured text data. For example, users may submit text information related to virtual goods, virtual identities, or digital content in the metaverse. In this case, multi-source unstructured text data presents complex and diverse characteristics. Specifically, multi-source unstructured text data contains colloquial expressions, emotional descriptions, and non-standardized formats. Traditional truth mining methods usually rely on a single data type or structured data, lacking effective unstructured text data processing capabilities, resulting in limited applicability in the metaverse crowdsourcing environment.

[0004] At the same time, most of the existing truth mining methods are based on probabilistic graph models, which model different scenarios by setting prior parameters to infer probabilities. However, these methods require a large number of prior parameters to be preset, and the model training and inference process consumes huge computing resources. In addition, as the amount of data increases, the computing time increases exponentially, which cannot meet the high concurrency and real-time requirements in the Metaverse crowdsourcing environment.

[0005] In addition, users in the Metaverse have increasing expectations for intelligent services and pursue highly personalized and high-quality experiences. This places higher demands on truth mining methods, requiring them to be able to accurately mine the truth of multi-source unstructured text data and generate results that meet user needs. However, methods based on probabilistic graphical models are difficult to flexibly adapt to the diversity and dynamics of user-generated content due to the fixed prior parameters of the model. At the same time, these methods have large differences in performance in different scenarios and data sets, further limiting their ability to meet the high standards of user personalized services in the Metaverse environment.

[0006] In view of this, the present invention proposes a truth value mining method for multi-source unstructured text services in a Metaverse crowdsourcing environment. The method can solve the problems of high computational complexity, high modeling difficulty, and poor adaptability in the prior art when processing unstructured text data. In addition, the method can improve the service quality and user experience in the Metaverse crowdsourcing environment by collecting and integrating text information of multi-source perception data in the Metaverse crowdsourcing environment and pushing intelligent services that meet user needs. Summary of the invention

[0007] The present invention aims at the technical problems existing in the prior art and provides a method for mining the truth value of multi-source unstructured text services in the metaverse crowdsourcing environment. Through efficient processing in stages and a dynamic adaptive feature generation mechanism, the present invention overcomes the shortcomings of the prior art and greatly improves the accuracy and efficiency of truth value discovery. At the same time, the method can flexibly adapt to complex and changing data environments and provide strong support for intelligent services in the metaverse.

[0008] To achieve the above object, the technical solution adopted by the present invention is: A truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment includes the following stages: A. Semantic preprocessing stage: Based on the data obtained from the Metaverse crowdsourcing platform, we build a worker set, a task set, and a crowdsourcing answer set. We use context embedding based on BERT (Bidirectional Encoder Representations from Transformers) to build a high-dimensional content vector representation of the crowdsourcing answer. We use the KANN-DBSCAN (K-Average Nearest NeighborDensity-Based Spatial Clustering of Applications withNoise) algorithm to adaptively cluster the high-dimensional content vector representation of the answer, and assign a category label to each crowdsourcing answer. B. Truth value optimization and feature generation phase: A truth value mining model is built based on the worker set, task set, and category label set attached to the crowdsourced answers. The goal is to minimize the difference between the truth value and the category label of the crowdsourced answer, and then evaluate the quality of the worker's work. Based on the evaluation results, features are generated from two dimensions, one dimension is the confidence of the category label, and the other dimension is the average difference of the category label confidence; C. Task clustering and cluster mapping stage: For tasks with generated features, the K-means clustering algorithm is used to cluster and obtain clusters equal to the number of task categories. A set of category confidences in each cluster is constructed, and the category with the largest category confidence is selected as the category of each task in the cluster. Each task is mapped one by one to its corresponding category to determine the true category of each task, and finally the true answer to the task on the Metaverse crowdsourcing platform is obtained.

[0009] In the above scheme, the semantic preprocessing stage can process unstructured text data, overcoming the shortcomings of the existing technology in text data processing; the truth value optimization and feature generation stage can flexibly adapt to the diversity and dynamics of user-generated content, overcoming the shortcomings of the existing technology in meeting the high standards of user personalized service requirements; the task clustering and cluster mapping stage can process large amounts of complex data in a lightweight manner, overcoming the shortcomings of the existing technology in meeting the high concurrency and real-time requirements in the metaverse crowdsourcing environment.

[0010] The specific steps of the semantic preprocessing stage are as follows: A1. Define and construct the worker set, task set, and crowdsourcing answer set based on the data obtained from the Metaverse crowdsourcing platform, where the worker set is represented as , represents the total number of workers, and the task set is represented by , represents the total number of tasks, and the crowdsourced answer set of each worker is represented as , Indicates workers To the task Based on the crowdsourced answers, each worker For each task Crowdsourced answers submitted Represented as a triple , triple set ; A2. Define a high-dimensional content vector representation set ,Building a Crowdsourced Answer Collection Using BERT-Based Contextual Embeddings The high-dimensional content vector representation set ; A3. Use the KANN-DBSCAN algorithm to represent the high-dimensional content vector set of the answer Perform adaptive clustering to identify each crowdsourced answer Assign a class label .

[0011] Step A3 specifically includes the following steps: A3.1 Definition of any task have crowdsourced answers, and the corresponding high-dimensional content vector representation set is , For any two vectors and , Calculate the cosine distance between the two , construct the distance matrix , A3.2 Distance Matrix Sort each row in descending order; A3.3 Defining the distance matrix mean , calculate the distance matrix The distance matrix mean of each column in ,generate indivual ; A3.4 Define the minimum number , Represented as a distance matrix The distance between any two vectors in each column is less than the matrix mean The mean of the number of vectors, calculated ,generate indivual ,in Represented as a distance matrix The distance between any two vectors in each column is less than the matrix mean Vector collection of , A3.5 For each pair of parameters , To traverse, use The algorithm clusters the data set.

[0012] In A3.5, The algorithm specifically includes the following steps: A3.5.1 Randomly select an unvisited vector , calculate , A3.5.2 If ,vector and Together they form a new cluster, and recursively process all unvisited vectors in the current cluster in the same way to expand the cluster; A3.5.3 If ,vector is the noise vector; A3.5.4 Repeat steps A3.5.1 to A3.5.3 for other vectors that have not been visited until all vectors belong to a cluster or are noise vectors; A3.5.5 All noise vectors also form a cluster; A3.5.6 Return all cluster sets.

[0013] In step B, the truth value optimization and feature generation phase includes the following steps: B1. After the semantic preprocessing stage, each worker For each task Crowdsourced answers submitted Each comes with a category label , represented as a four-tuple , a set of four , defines the set of worker's work quality , define and build workers The set of category labels for all submitted tasks , defines the set of truth values ​​for all tasks , Represented as a task The truth value of B2. Build a truth mining model and define the objective function , where the constraints are ,when When ,when When , B3. Initialize the work quality of any worker ， , B4. Minimize the objective function, iterate steps B5 and B6 until the objective function converges, and obtain the work quality of each worker and the truth value of all tasks ; B5. Update the true value , B6. Update Worker Quality , B7. Define and build category sets , Indicates the number of categories. For each task For example, any category label There are corresponding categories Correspondingly, , all category labels for this task are The set of workers is represented as , B8. Define each task The category confidence set , , Indicates the task The categories of all crowdsourced answers are The sum of the quality of workers , B9. Calculate category confidence and construct each task The category confidence set , and each task The feature set , where the category confidence satisfies , B10. Give each task The feature set Add an additional feature , , construct the final feature set .

[0014] In step C, the task clustering and cluster mapping phase includes the following steps: C1. Use the K-means clustering algorithm to cluster all tasks with generated features to obtain clusters equal to the number of categories K. C2. Define the set of class confidences in the cluster , each cluster The cluster class confidence set , , Indicates that the category label in the cluster is mission; C3. Calculate the confidence set of the classes in the cluster , C4. For each cluster Select , , each task in the cluster Map to the corresponding category In this way, the true category of each task is determined, and finally the true answer to the task on the Metaverse crowdsourcing platform is obtained.

[0015] C1. Use the K-means clustering algorithm to cluster all tasks with generated features to obtain clusters equal to the number of categories K. Specifically, the following steps are included: C1.1 Defining clusters in the K-means clustering algorithm , is the number of clusters, set the number of clusters to the number of classes in the task ,Right now , C1.2 Define cluster center. ,choose The biggest task As the initial center of each cluster, if the task has been selected as the cluster center, select the next one in descending order Mission as the initial center of the cluster, and so on; C1.3 Calculate the Euclidean distance between each task and the initial center of each cluster, and assign each task to the cluster closest to it; C1.4 For each cluster, recalculate the center of the cluster. The center of the cluster is the mean of all tasks in the cluster. C1.5 Repeat steps C1.3 and C1.4 until the cluster center no longer changes; C1.6 Return Clusters.

[0016] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the method for truth mining of multi-source unstructured text services in a metaverse crowdsourcing environment is implemented.

[0017] A computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment.

[0018] Beneficial effects: Compared with the prior art, the truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment provided by the present invention has the following beneficial effects: (1) The present invention achieves efficient truth mining of complex unstructured text data through three stages: semantic preprocessing stage, truth optimization and feature generation stage, and task clustering and cluster mapping stage. This phased processing method not only significantly improves the processing efficiency, but also ensures the accuracy and reliability of truth mining. In the highly interactive and complex environment of the metaverse, this phased approach can flexibly adapt to multi-source perception data and improve the overall service quality; (2) In the semantic preprocessing stage, the present invention introduces BERT-based context embedding and KANN-DBSCAN algorithm. BERT-based context embedding can deeply mine the semantic information of text data and provide support for the construction of high-dimensional content vectors. The KANN-DBSCAN algorithm accurately identifies and isolates noise data by clustering high-dimensional content vectors. This combined strategy not only efficiently processes unstructured text data, but also lays a solid foundation for subsequent truth value mining; (3) In the truth value optimization and feature generation stage, the present invention designs an iterative optimization mechanism and an adaptive feature generation mechanism. The iterative optimization mechanism can dynamically update the worker quality and the truth value, and the adaptive feature generation mechanism can dynamically generate the confidence feature of the category label according to the worker's work quality, and combine the average difference feature of the confidence to enhance the tolerance to the fluctuation of the worker's behavior data, thereby improving the robustness of the system; (4) In the task clustering and cluster mapping stage, the present invention uses the K-means clustering algorithm to generate clusters equal to the number of task categories, and based on the highest category confidence value, accurately completes the mapping of tasks and categories, and finally determines the true answer to the task on the Metaverse crowdsourcing platform. This stage uses lightweight processing methods to efficiently deal with massive and complex data, effectively reducing the computing and resource burden, and overcoming the problems of high concurrency and insufficient real-time requirements faced by existing technologies in the Metaverse crowdsourcing environment; (5) The method of the present invention not only significantly improves the efficiency and accuracy of truth value mining, but also provides strong data support for intelligent services in the Metaverse. By integrating multi-source crowdsourcing data and optimizing data quality, the present invention provides more accurate and reliable basic data for various intelligent applications in the Metaverse (such as personalized recommendations, virtual assistants, etc.), which not only promotes the innovative development of intelligent services, but also significantly improves user experience and meets users' needs for highly personalized, high-quality intelligent services. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the framework of the method scenario and basic principle of the present invention. Figure 2 It is a schematic diagram of the overall implementation process of the method of the present invention. DETAILED DESCRIPTION

[0020] The present invention is further explained below in conjunction with the accompanying drawings and specific implementation examples. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, those skilled in the art will understand the various aspects of the present invention. All modifications of equivalent forms fall within the scope defined by the claims attached to this application.

[0021] Example: Figure 1As shown in FIG. 1 , the present invention establishes a multi-source unstructured text service truth value mining framework in a metaverse crowdsourcing environment, such as Figure 2 As shown, the overall implementation process of the method established by the present invention is as follows. The specific implementation case is as follows: the truth value mining method of multi-source unstructured text services in the metaverse crowdsourcing environment includes three stages: A. Semantic preprocessing stage: A.1 Based on the data obtained from the Metaverse crowdsourcing platform, as shown in Table 1, define and construct a set of workers , task set , a collection of crowdsourced answers ,On this basis, the crowdsourced answer submitted by each worker for each task can be expressed as a triple, such as: , Table 1: A2. Define and build a collection of high-dimensional content vector representations of crowdsourced answers using BERT-based contextual embeddings , A3. Use the KANN-DBSCAN algorithm to represent the high-dimensional content vector set of the answer Perform adaptive clustering to identify each crowdsourced answer Assign a class label , as shown in Table 2; Table 2: B. Truth value optimization and feature generation stage: B1. After the semantic preprocessing stage, each crowdsourcing answer submitted by each worker for each task is accompanied by a category label, which can be represented as a four-tuple, such as , define and build a set of worker work qualities , define and build each worker The set of category labels for all submitted tasks , such as workers The set of category labels for all submitted tasks , defines the set of truth values ​​for all tasks , Represented as a task The truth value of B2. Build a truth mining model and define the objective function , where the constraints are ,when When ,when When , B3. Initialize the work quality of any worker , , B4. Minimize the objective function, iterate steps B5 and B6 until the objective function converges, and finally obtain Worker's work quality and truth value , B5. Update the true value , B6. Update Worker Quality , B7. Define and build category sets , for each task , Category labels have corresponding categories, such as ,also Indicates that for this task, all category labels are A collection of workers, for example, for tasks , , B8. Define each task The category confidence set , , Indicates the task The categories of all crowdsourced answers are The sum of the quality of workers , B9. Calculate category confidence, e.g. , build each task The category confidence set , and each task The feature set ,satisfy , feature set , , , , , B10. Give each task The feature set Add an additional feature , construct the final feature set , The final feature set , , , , C. Task clustering and cluster mapping stage, C1. Use the K-means clustering algorithm to cluster all tasks with generated features and get 3 clusters , , , , C2. Define the set of class confidences in the cluster , each cluster The cluster class confidence set , , Indicates that the category label in the cluster is mission; C3. Calculate the confidence set of the classes in the cluster , , , , C4. For each cluster Select , ,for Select ,for , select ,for , select . Each task in the cluster Map to the corresponding category In order to determine the true category of each task, that is, the final true value. , Mapped to "A", Mapped to "B", , Mapped to "A", i.e. Finally, we obtain the real answers to the tasks on the Metaverse crowdsourcing platform, as shown in Table 3; Table 3: It should be noted that the above embodiments are not intended to limit the protection scope of the present invention, and equivalent changes or substitutions made on the basis of the above technical solutions all fall within the protection scope of the claims of the present invention.

Claims

1. A truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment, characterized by: The following steps are involved: A. Semantic preprocessing stage: Based on the data obtained from the Metaverse crowdsourcing platform, we build a set of workers, a set of tasks, and a set of crowdsourcing answers. We use BERT-based context embedding to build a high-dimensional content vector representation of the crowdsourcing answers. We use the KANN-DBSCAN algorithm to adaptively cluster the high-dimensional content vector representation of the crowdsourcing answers and assign a category label to each crowdsourcing answer. B. Truth value optimization and feature generation phase: Build a truth value mining model based on the worker set, task set, and category label set attached to the crowdsourcing answers. C. Task clustering and cluster mapping stage: For tasks with generated features, the K-means clustering algorithm is used for clustering to obtain clusters equal to the number of task categories. A set of category confidences in each cluster is constructed, and the category with the largest category confidence is selected as the category of each task in the cluster. Each task is mapped one by one to its corresponding category, so as to determine the true category of each task and finally obtain the true answer to the task on the Metaverse crowdsourcing platform.

2. The truth value mining method for multi-source unstructured text services in the metaverse crowdsourcing environment according to claim 1 is characterized in that: In the semantic preprocessing stage, the specific steps are as follows: A1. Define and construct the worker set, task set, and crowdsourcing answer set based on the data obtained from the Metaverse crowdsourcing platform. The worker set is represented as , represents the total number of workers, and the task set is represented by , represents the total number of tasks, and the crowdsourced answer set of each worker is represented as , Indicates workers To the task Based on the crowdsourced answers, each worker For each task Crowdsourced answers submitted Represented as a triple , triple set ; A2. Define a high-dimensional content vector representation set, , building a crowdsourced answer collection using BERT-based contextual embeddings The high-dimensional content vector representation set ; A3. Use the KANN-DBSCAN algorithm to represent the high-dimensional content vector set of the answer Perform adaptive clustering to identify each crowdsourced answer Assign a class label .

3. The truth value mining method for multi-source unstructured text services in the metaverse crowdsourcing environment according to claim 2 is characterized in that: Step A3 specifically includes the following steps: A3.1 Definition of any task have crowdsourced answers, and the corresponding high-dimensional content vector representation set is , for any two vectors and , Calculate the cosine distance between the two , construct the distance matrix , A3.2 Distance Matrix Sort each row in descending order; A3.3 Defining the distance matrix mean , calculate the distance matrix The distance matrix mean of each column in ,generate indivual ; A3.4 Define the minimum number , Represented as a distance matrix The distance between any two vectors in each column is less than the matrix mean The mean of the number of vectors, calculated ,generate indivual ,in Represented as a distance matrix The distance between any two vectors in each column is less than the matrix mean Vector collection of ， A3.5 For each pair of parameters , To traverse, use The algorithm clusters the data set.

4. The truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment according to claim 3 is characterized in that: In A3.5, The algorithm specifically includes the following steps: A3.5.1 Randomly select an unvisited vector , calculate , A3.5.2 If ,vector and Together they form a new cluster, and recursively process all unvisited vectors in the current cluster in the same way to expand the cluster; A3.5.3 If ,vector is the noise vector; A3.5.4 Repeat steps A3.5.1 to A3.5.3 for other vectors that have not been visited until all vectors belong to a cluster or are noise vectors; A3.5.5 All noise vectors also form a cluster; A3.5.6 Return all cluster sets.

5. The truth value mining method for multi-source unstructured text services in the metaverse crowdsourcing environment according to claim 1 is characterized in that: In step B, the truth value optimization and feature generation phase includes the following steps: B1. After the semantic preprocessing stage, each worker For each task Crowdsourced answers submitted Each comes with a category label , represented as a four-tuple , a set of four , defines the set of worker's work quality , define and build workers The set of category labels for all submitted tasks , defines the set of truth values ​​for all tasks , Represented as a task The truth value of B2. Build a truth mining model and define the objective function , where the constraints are ,when When ,when When , B3. Initialize the work quality of any worker ， , B4. Minimize the objective function, iterate steps B5 and B6 until the objective function converges, and obtain the work quality of each worker and the truth value of all tasks ; B5. Update the true value , B6. Update Worker Quality , B7. Define and build category sets , Indicates the number of categories. For each task For example, any category label There are corresponding categories Correspondingly, , all category labels for this task are The set of workers is represented as , B8. Define each task The category confidence set , , Indicates the task The categories of all crowdsourced answers are The sum of the quality of workers , B9. Calculate category confidence and construct each task The category confidence set , and each task The feature set , where the category confidence satisfies , B10. Give each task The feature set Add an additional feature , , construct the final feature set .

6. The truth value mining method for multi-source unstructured text services in the metaverse crowdsourcing environment according to claim 1 is characterized in that: In step C, the task clustering and cluster mapping phase includes the following steps: C1. Use the K-means clustering algorithm to cluster all tasks with generated features to obtain clusters equal to the number of categories K. C2. Define the set of class confidences in the cluster , each cluster The cluster class confidence set , , Indicates that the category label in the cluster is mission; C3. Calculate the confidence set of the classes in the cluster , C4. For each cluster Select , , each task in the cluster Map to the corresponding category In this way, the true category of each task is determined, and finally the true answer to the task on the Metaverse crowdsourcing platform is obtained.

7. The truth value mining method for multi-source unstructured text services in a metaverse crowdsourcing environment according to claim 6 is characterized in that: C1. Use the K-means clustering algorithm to cluster all tasks with generated features to obtain clusters equal to the number of categories K. Specifically, the following steps are included: C1.1 Defining clusters in the K-means clustering algorithm , is the number of clusters, set the number of clusters to the number of classes in the task ,Right now , C1.2 Define cluster center. ,choose The biggest task As the initial center of each cluster, if the task has been selected as the cluster center, select the next one in descending order Mission As a cluster The initial center of , and so on; C1.3 Calculate the Euclidean distance between each task and the initial center of each cluster, and assign each task to the cluster closest to it; C1.4 For each cluster, recalculate the center of the cluster. The center of the cluster is the mean of all tasks in the cluster. C1.5 Repeat steps C1.3 and C1.4 until the cluster center no longer changes; C1.6 Return Clusters.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the truth value mining method for multi-source unstructured text services in the metaverse crowdsourcing environment as described in any one of claims 1 to 7 above.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by the processor, the truth value mining method for multi-source unstructured text services in the metaverse crowdsourcing environment as described in any one of claims 1-7 is implemented.

Images