Task classification label determination method and device and electronic equipment

Abstract

The invention discloses a method and device for determining classification labels of tasks and electronic equipment. The method relates to the field of data classification, and comprises the following steps: performing multi-modal semantic preprocessing on target knowledge graph data and target task associated data of a plurality of tasks to obtain multi-modal semantic feature data; extracting deep semantic features, and performing clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters; generating target feedback path data, performing dependency relationship modeling, generating dependency data, decomposing the dependency data, and generating a target dependency matrix; combining the target dependency matrix, the target feedback path data, the task feature cluster and the dynamic clustering data to generate target classification input data; and inputting the target classification input data into the target classification model to obtain a classification label of the task. Through the method and the device, the problem of low efficiency and accuracy when labels of multiple tasks are classified in the prior art is solved.

Description

Technical Field

[0001] The present application relates to the field of data classification, and in particular, to a method, device, and electronic device for determining a classification label of a task. Background Art

[0002] In recent years, with the rapid development of Robotic Process Automation (RPA) technology, RPA robots can automatically execute a large number of repetitive and standardized business processes. Its application areas include manufacturing, the internet, finance, retail, healthcare, and other fields. They can effectively improve work efficiency, reduce human errors, and standardize business processes to achieve automation, intelligence, and digital transformation. However, when handling complex and changing business processes, RPA robots are difficult to complete solely through program code or predefined rules. They often need to make decisions based on learned experience, knowledge, and insights. Therefore, researching deep learning-based RPA robot path planning methods to improve the intelligence level of RPA robots is a key issue that needs to be addressed in the RPA application field.

[0003] Currently, most RPA path planning research focuses on single-task, single-label problems. This means that an RPA robot only needs to perform a specific task with a single label or output, such as document classification, data capture, or image recognition. However, with the expansion of RPA application scenarios, an increasing number of RPA processes involve multiple tasks, and executing these tasks often requires processing multiple outputs. This means that a single input data set contains multiple tasks, each with its own output or label. For example, in a document containing multiple tables, an RPA robot needs to perform different operations on each table, such as data capture and table conversion. Each table contains different data and requires a different output.

[0004] In order to solve the problem of low efficiency and accuracy in classifying labels of multiple tasks in related technologies, existing multi-task multi-label classification methods mainly use multi-task learning algorithms or meta-learning algorithms. However, these methods are difficult to achieve ideal results when dealing with complex and changeable business processes.

[0005] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0006] The embodiments of the present application provide a method for determining classification labels of tasks, so as to at least solve the problem of low efficiency and accuracy when classifying labels of multiple tasks in the related art.

[0007] According to one aspect of an embodiment of the present application, a method for determining a classification label of a task is provided, comprising: performing multimodal semantic preprocessing on target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task; extracting deep semantic features from the multimodal semantic feature data, and clustering the deep semantic features to obtain dynamic clustering data and task feature clusters; generating target feedback path data for the task feature cluster, performing dependency modeling based on the target feedback path data to generate dependency data, performing dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; combining the target dependency matrix, target feedback path data, task feature clusters, and dynamic clustering data to generate target classification input data; and inputting the target classification input data into a target classification model to obtain a classification label for each task.

[0008] Optionally, before performing multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks, the method also includes: extracting initial knowledge graph data and initial task association data from the historical process knowledge base; performing denoising processing on the initial knowledge graph data and initial task association data to generate denoised knowledge graph data and denoised task association data; performing target processing on the denoised knowledge graph data and denoised task association data to obtain target knowledge graph data and target task association data, wherein the target processing includes at least one of the following: performing consistency verification on date and task identifier data.

[0009] Optionally, multimodal semantic preprocessing is performed on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task, including: performing knowledge graph embedding processing on the target knowledge graph data to generate task semantic embedding data, wherein the knowledge graph embedding processing is to extract features of entities and relationships in the target knowledge graph data through a multi-layer convolutional neural network, and convert the semantic relationships between tasks into numerical representations; input the target task association data into the relational convolutional neural network to generate task relationship feature data; and perform weighted fusion on the task semantic embedding data and the task relationship feature data to obtain multimodal semantic feature data.

[0010] Optionally, deep semantic features are extracted from multimodal semantic feature data, and the deep semantic features are clustered to obtain dynamic clustering data and task feature clusters, including: extracting deep semantic features of tasks from multimodal semantic feature data through a multi-layer semantic aggregation network to obtain preliminary label feature data; calculating the feature similarity between each two tasks based on the preliminary label feature data, and generating a task feature similarity matrix; clustering all tasks based on the task feature similarity matrix to obtain clustering results; dynamically adjusting the clustering results based on feature changes of tasks at different times to generate dynamic clustering data, wherein the dynamic clustering data is the preliminary result of clustering deep semantic features, including feature changes of tasks at different times; and converting the dynamic clustering data into task feature clusters.

[0011] Optionally, converting dynamic clustering data into task feature clusters includes: merging the features of each task within the same cluster group according to the cluster grouping information of the dynamic clustering data to form a corresponding feature cluster, wherein the cluster grouping information includes at least one of the following: the cluster or group identifier to which each task belongs, the cluster center or representative vector, the similarity mapping between tasks, and the cluster evolution information at different time points; and packaging and outputting each feature cluster to obtain a task feature cluster.

[0012] Optionally, the target feedback path data of the task feature cluster is determined by: combining the task feature similarity matrix and task dependencies to construct preliminary feedback path data, wherein the preliminary feedback path data reflects the current execution order of the tasks; inputting the preliminary feedback path data into the recursive feedback graph convolutional network, and processing to obtain the target feedback path data.

[0013] Optionally, the preliminary feedback path data is input into the recursive feedback graph convolutional network, and processing to obtain the target feedback path data includes: the recursive feedback graph convolutional network performs convolution calculation on the preliminary feedback path data through a convolutional layer in each round of iteration to generate an updated path dependency relationship, generates recursive feedback optimization data according to the updated path dependency relationship, and optimizes the preliminary feedback path data according to the recursive feedback optimization data; after the iteration end condition is met, the optimization result of the preliminary feedback path data outputted from the last iteration is determined as the target feedback path data.

[0014] Optionally, performing dependency matrix decomposition on the dependency data to generate a target dependency matrix includes: splitting the dependency relationship between tasks into a low-rank matrix through a matrix decomposition formula; using the low-rank matrix to capture the explicit dependency data and initial implicit dependency data between tasks, wherein the explicit dependency data is used to reflect the significant correlation between tasks, and the initial implicit dependency data is used to reflect the hidden dependency structure between tasks; iteratively updating the initial implicit dependency matrix through a gradient descent algorithm to obtain the implicit dependency data of the target; and weightedly fusing the explicit dependency data and the implicit dependency data to generate a target dependency matrix.

[0015] Optionally, after inputting the target classification input data into the target classification model and obtaining the classification label of each task, the method also includes: generating the classification confidence of the task, and determining the classification result of each task based on the classification label and classification confidence of each task; adjusting the execution order of each corresponding task based on the classification label and classification confidence, so that the execution order of each task conforms to the target dependency matrix and target feedback path data.

[0016] According to another aspect of an embodiment of the present application, a device for determining a classification label of a task is provided. The device includes: a preprocessing unit for performing multimodal semantic preprocessing on target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data of each task; a clustering unit for extracting deep semantic features in the multimodal semantic feature data and performing clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters; a dependency processing unit for generating target feedback path data of the task feature cluster, performing dependency modeling based on the target feedback path data to generate dependency data, performing dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; an input data generation unit for combining the target dependency matrix, the target feedback path data, the task feature cluster and the dynamic clustering data to generate target classification input data; and a label generation unit for inputting the target classification input data into the target classification model to obtain a classification label for each task.

[0017] According to another aspect of an embodiment of the present application, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein when the program runs, a method for determining a classification label of a task in any one of the embodiments of the present application is executed.

[0018] In an embodiment of the present application, multimodal semantic preprocessing is performed on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data of each task; deep semantic features in the multimodal semantic feature data are extracted, and the deep semantic features are clustered to obtain dynamic clustering data and task feature clusters; target feedback path data of the task feature cluster is generated, dependency modeling is performed based on the target feedback path data to generate dependency data, dependency matrix decomposition is performed on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; the target dependency matrix, target feedback path data and target feedback data are combined into a target dependency matrix. Path data, task feature clusters and dynamic clustering data are combined to generate target classification input data; the target classification input data is input into the target classification model to obtain the classification label of each task, which solves the problem of low efficiency and accuracy in classifying labels of multiple tasks in related technologies. The target classification input data is obtained by performing multimodal semantic preprocessing, clustering processing, high-order dependency modeling and matrix decomposition on the target knowledge graph data and target task association data of the task, and the target classification input data is input into the target classification model to obtain the classification label of each task, achieving efficient and accurate classification effect for the labels of complex tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0020] Figure 1 The present invention is a hardware structure block diagram of a computer terminal for implementing a method for determining a classification label of a task;

[0021] Figure 2 is a flowchart of a method for determining a classification label of a task provided in an embodiment of the present application;

[0022] Figure 3 is a flow chart of a dynamic collaborative clustering method provided according to an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a recursive feedback graph convolutional network provided according to an embodiment of the present application;

[0024] Figure 5 is a flowchart of a high-order dependency matrix decomposition method provided according to an embodiment of the present application;

[0025] Figure 6 This is a flow chart for classifying target classification input data according to the target classification model provided in the embodiment of the present application;

[0026] Figure 7is a flowchart of a method for determining a classification label of an optional task provided in an embodiment of the present application;

[0027] Figure 8 is a schematic diagram of a device for determining a classification label of a task provided in an embodiment of the present application;

[0028] Figure 9 This is a structural block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following interpretations:

[0032] It should be noted that the collected information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this application are information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation portals for users to choose to authorize or refuse. For example, an interface is set up between this system and relevant users or institutions to provide users with corresponding operation portals for users to choose to agree or refuse the automated decision-making results; if the user chooses to refuse, the expert decision-making process will be entered.

[0033] Example 1

[0034] According to an embodiment of the present application, an embodiment of a method for determining a classification label of a task is also provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0035] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 This is a hardware structure diagram of a computer terminal for implementing a method for determining a classification label of a task. Figure 1 As shown, the computer terminal 10 may include one or more ( Figure 1 The computer system includes a processor 102 (shown as 102a, 102b, ..., 102n) (the processor 102 may include but is not limited to a processing device such as a microprocessor (MCU) or a programmable logic device (FPGA), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, the computer system may also include: a display, an input / output interface (I / O interface, Input / Output interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0036] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry". The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuitry may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0037] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method for determining the classification label of the task in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, realizes the above-mentioned method for determining the classification label of the task. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely located relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0038] The transmission device 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission device 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0039] The display may be, for example, a touch screen liquid crystal display (LCD), which enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0040] Under the above operating environment, this application provides Figure 2 The method for determining the classification label of the task shown. Figure 2 is a flowchart of a method for determining a classification label of a task according to an embodiment of the present application, such as Figure 2 As shown:

[0041] Step S201: perform multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task.

[0042] Specifically, multiple tasks refer to different workflows that need to be categorized and managed. For example, these can be Robotic Process Automation tasks, where robots simulate human operations on computers. Multimodal semantic preprocessing embeds the target knowledge graph data and target task association data contained in multiple tasks in the same preprocessing process and fuses them into unified features, resulting in multimodal semantic feature data. Multimodal semantic feature data can represent the semantic relationships and associations between tasks.

[0043] The target knowledge graph data includes entities, relationships, and triples in the process, representing the connections and interactions between tasks. An entity refers to a "node" or "object" in the knowledge graph and can be any core element or concept in the process. For example, "task," "department," and "system interface" in an RPA environment can all be defined as an entity. A relationship refers to the semantic connection between entities, also known as an "edge" or "relationship type." For example, "Task A depends on Task B," "System X belongs to Department Y," and "Task Z is the responsibility of Department X" are all relationships between entities. Triples are the basic unit for representing relationships in the knowledge graph and are represented as (head entity, relationship, tail entity). For example, "(Task A, dependency, Task B)" constitutes a triple, describing the semantic relationship between entities A and B. The target task association data is historical association information between tasks, reflecting the dependencies and collaboration between tasks during execution.

[0044] Step S202 : extracting deep semantic features from the multimodal semantic feature data, and performing clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters.

[0045] For example, a multi-layer semantic aggregation network can be used to mine deep semantic features in multimodal semantic feature data. Deep semantic features can better reveal the key attributes, potential dependencies and similarities of the task than unextracted features. Deep semantic features can be label features. These features are then clustered using a dynamic collaborative clustering algorithm to form dynamic clustering data. The dynamic collaborative clustering algorithm includes similarity calculation and time dependency adjustment.

[0046] After obtaining the dynamic clustering data, it is transformed to construct a task feature cluster. The task feature cluster not only contains the semantic features of the task, but also includes the similarity of the tasks and the time-adjusted dependencies. Based on the task feature cluster, the target feedback path data can be generated.

[0047] Step S203, generating target feedback path data of the task feature cluster, performing dependency relationship modeling based on the target feedback path data to generate dependency data, performing dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks.

[0048] Specifically, based on the task feature cluster, target feedback path data is generated. The target feedback path data represents the optimal execution order of tasks during execution. That is, the target feedback path data reflects the optimal execution path of the task after multiple rounds of recursive feedback, ensuring that the task execution path can maintain the optimal state in a complex multi-task environment.

[0049] Subsequently, a recurrent feedback graph convolutional network (RFGCN) can be used to conduct in-depth analysis of the target feedback path data, build a high-order dependency model, and generate dependency data. Dependency data refers to the information that can characterize the interdependencies between tasks after dependency modeling of task feature clusters. These dependencies in the dependency data are then converted into a more intuitive target dependency matrix through higher-order dependency matrix decomposition (HDMD). The conversion process captures the complex dependencies between tasks. The resulting target dependency matrix provides precise guidance for task classification and path optimization.

[0050] Step S204 , combining the target dependency matrix, target feedback path data, task feature clusters, and dynamic clustering data to generate target classification input data.

[0051] Specifically, the target dependency matrix reflecting the dependency relationship between tasks, the target feedback path data representing the optimal execution order of tasks, the task feature cluster containing semantic and relational features, and the dynamic clustering data reflecting the real-time multi-task structure are integrated into target classification input data through a weighted fusion mechanism.

[0052] Since the generated target classification input data contains optimized task dependencies, execution order, and semantic features, it comprehensively and accurately reflects the attributes of tasks and their relationships with each other, providing comprehensive and structured input for subsequent classification steps, ensuring that the target classification model can capture all key information and obtain accurate classification results.

[0053] It should be noted that task feature clusters are high-level data structures that can be directly used by algorithms such as recursive feedback graph convolutional networks and high-order dependency matrix factorization. Dynamic clustering data, based on task feature clusters, also contains the intermediate information and procedural parameters required to generate these clusters. This data is also valuable for the target classification model. For example, to obtain historical cluster migration information, certain intermediate clustering metrics, and time decay strategies for tasks, dynamic clustering data is required. When determining the classification label for a task, the system simultaneously references both sets of data: classification-oriented aggregate features from the task feature clusters and lower-level or procedural information from the dynamic clustering data.

[0054] Step S205: input the target classification input data into the target classification model to obtain the classification label of each task.

[0055] For example, the target classification model can utilize a multi-layer neural network structure and, through multi-task learning, simultaneously classify tasks, thereby accurately assigning classification labels to each task. This process fully utilizes the target classification input data and ensures the accuracy of the classification results. The classification label refers to the identification obtained after analyzing and categorizing each task using the target classification model. It can represent the task's type, priority, domain, execution conditions, or any other attributes that help manage and identify tasks.

[0056] In an embodiment of the present application, multimodal semantic preprocessing is performed on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data of each task; deep semantic features in the multimodal semantic feature data are extracted, and the deep semantic features are clustered to obtain dynamic clustering data and task feature clusters; target feedback path data of the task feature cluster is generated, dependency modeling is performed based on the target feedback path data to generate dependency data, dependency matrix decomposition is performed on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; the target dependency matrix, target feedback path data and target feedback data are combined into a target dependency matrix. Path data, task feature clusters and dynamic clustering data are combined to generate target classification input data; the target classification input data is input into the target classification model to obtain the classification label of each task, which solves the problem of low efficiency and accuracy in classifying labels of multiple tasks in related technologies. The target classification input data is obtained by performing multimodal semantic preprocessing, clustering processing, high-order dependency modeling and matrix decomposition on the target knowledge graph data and target task association data of the task, and the target classification input data is input into the target classification model to obtain the classification label of each task, achieving efficient and accurate classification effect for the labels of complex tasks.

[0057] In order to ensure that the method for determining the classification label of the task can be processed based on high-quality and highly consistent data, optionally, in the method for determining the classification label of the task provided in the embodiment of the present application, before performing multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks, the method also includes: extracting initial knowledge graph data and initial task association data from the historical process knowledge base; performing denoising processing on the initial knowledge graph data and initial task association data to generate denoised knowledge graph data and denoised task association data; performing target processing on the denoised knowledge graph data and denoised task association data to obtain target knowledge graph data and target task association data, wherein the target processing includes at least one of the following: performing consistency verification on date and task identifier data.

[0058] Specifically, before executing multimodal semantic preprocessing, the system performs a series of data preparation and optimization steps. First, the system extracts initial knowledge graph data and initial task association data from the historical process knowledge base. This data contains a large amount of task attributes, relationships between tasks, and historical execution information, which can be processed and used as input for the target classification model.

[0059] Next, to improve data accuracy and remove noise that could affect classification, the system performs denoising on the initial knowledge graph data and initial task-related data, generating denoised knowledge graph data and denoised task-related data. The denoising process uses a data cleaning algorithm to remove redundant and noisy information, ensuring the accuracy and quality of the dataset.

[0060] The system then performs targeted processing on the denoised data, including consistency checks on key fields such as dates and task identifiers, to generate the target knowledge graph data and target task-related data. This consistency check ensures that all task data is highly consistent in format and representation, avoiding classification errors caused by inconsistent data formats. For example, all date fields are converted to a unified format, and all task identifiers are checked for uniqueness and accuracy.

[0061] The embodiment of the present application not only denoises the initial knowledge graph data and the initial task-related data, but also enhances the reference value of the data through consistent target processing, thereby providing more accurate and consistent data support for subsequent multi-task multi-label classification, ensuring the validity of the classification results.

[0062] In order to convert the target knowledge graph data and target task association data into a format that the system can deeply understand and operate, optionally, in the method for determining the classification label of the task provided in the embodiment of the present application, the target knowledge graph data and target task association data of multiple tasks are subjected to multimodal semantic preprocessing to obtain multimodal semantic feature data for each task, including: performing knowledge graph embedding processing on the target knowledge graph data to generate task semantic embedding data, wherein the knowledge graph embedding processing is to extract features of entities and relationships in the target knowledge graph data through a multi-layer convolutional neural network, and convert the semantic relationships between tasks into numerical representations; input the target task association data into the relational convolutional neural network to generate task relationship feature data; and perform weighted fusion on the task semantic embedding data and the task relationship feature data to obtain multimodal semantic feature data.

[0063] Specifically, the system first performs knowledge graph embedding on the target knowledge graph data. Using a multi-layer convolutional neural network, it deeply explores the intrinsic characteristics of the data and transforms abstract semantic relationships into numerical task semantic embeddings. This conversion process is able to capture the complex semantic connections between tasks.

[0064] When embedding knowledge graphs, the calculation process of each layer of convolutional network can be expressed as:

[0065] H (l+1) =σ(W (l) *H (l) +b (l) );

[0066] In this formula, H (l+1) represents the output feature representation of the (l+1)th layer, W (l) Represents the convolution weight matrix of the lth layer, H (l) represents the input feature representation of the lth layer, b (l) Denotes the bias term of layer l, σ is the activation function, which can be ReLU, and * represents the convolution operation. Through layer-by-layer processing of the multi-layer convolutional network, the system extracts task semantic features at different levels and generates task semantic embedding data.

[0067] At the same time, the system inputs the processed target task association data into a relational convolutional neural network (RelationalGCN) to further refine the dependencies and associations between tasks and generate task relationship feature data. Task relationship feature data not only reflects the direct connections between tasks but also encompasses indirect collaboration and dependencies between tasks. Task semantic embedding data and task relationship feature data, as two different dimensions of feature data, represent the semantic relationships and dependency relationships between tasks, respectively.

[0068] After extracting the task semantic embedding data and task relationship feature data, the system uses a weighted fusion mechanism to integrate the features extracted from both semantic and relationship perspectives, combining the feature data from the two different dimensions into unified multimodal semantic feature data. This weighted fusion process ensures that weights are dynamically assigned based on the feature relevance of the data during the fusion process, resulting in higher expressiveness and accuracy in the resulting feature data. Specifically, the higher the correlation between the "task semantic embedding data" and the "task relationship feature data," or the higher their contribution to label classification, the higher the weight assigned to them during the fusion process. If a feature has a weaker relationship with label classification, the system will reduce its influence on the final fused feature.

[0069] The embodiment of the present application obtains task semantic embedding data and task relationship feature data through knowledge graph embedding processing and relational convolutional neural network processing, and then obtains multimodal semantic feature data through weighted fusion, thereby enhancing the expressiveness of features and the accuracy of classification, providing high-quality input for the step of extracting deep semantic features, and ensuring the efficiency and accuracy of the entire task label classification process.

[0070] In order to perform multi-level analysis and aggregation on feature data and more comprehensively reflect task features, optionally, in the method for determining the classification label of the task provided in the embodiment of the present application, deep semantic features in the multimodal semantic feature data are extracted, and the deep semantic features are clustered to obtain dynamic clustering data and task feature clusters, including: extracting the deep semantic features of the task from the multimodal semantic feature data through a multi-layer semantic aggregation network to obtain preliminary label feature data; calculating the feature similarity between each two tasks based on the preliminary label feature data, and generating a task feature similarity matrix; clustering all tasks based on the task feature similarity matrix to obtain clustering results; dynamically adjusting the clustering results according to the feature changes of the task at different times to generate dynamic clustering data, wherein the dynamic clustering data is the preliminary result of clustering the deep semantic features, including the feature changes of the task at different times; and converting the dynamic clustering data into task feature clusters.

[0071] Figure 3 is a flow chart of a dynamic collaborative clustering method provided in an embodiment of the present application, such as Figure 3 As shown in the figure, dynamic collaborative clustering starts with inputting task feature data, uses Euclidean distance to calculate the similarity between tasks, and then generates dynamic clustering data through a time-dependent adjustment mechanism.

[0072] The step of extracting the deep semantic features of a task can be as follows: A multi-layer semantic aggregation network performs a deep feature extraction on the multimodal semantic feature data, gradually aggregating elements related to the task label (such as historical label information and the task's latent category distribution). After aggregation, a representation vector containing "preliminary label information / features" is generated, namely preliminary label feature data. This preliminary label feature data contains both the multimodal semantic embedding of the task itself and reflects preliminary label tendency or label similarity.

[0073] After obtaining preliminary label feature data, the feature similarity between each two tasks is calculated. Based on this, a task feature similarity matrix is ​​constructed. This matrix intuitively reflects the semantic similarity between all tasks and is key to cluster analysis. Based on this matrix, the system uses a dynamic collaborative clustering algorithm to perform initial clustering on all tasks, generating clustering results.

[0074] The similarity is calculated using the Euclidean distance formula, which is calculated as follows:

[0075]

[0076] Among them, x i represents the i-th eigenvalue of task x, y i represents the i-th eigenvalue of task y, and n represents the characteristic dimension of the task. Using this formula, the system calculates the similarity between each task and generates a task similarity matrix to represent the characteristic similarity between tasks.

[0077] It should be noted that dynamic clustering data is generated to enable the system to dynamically adjust clustering results based on changes in task features at different times. This data not only reflects the clustering of tasks but also includes information on adjustments to task features over time. This ensures the real-time and adaptability of clustering results, allowing task groupings to maintain a reasonable structure over time. Similarity calculations not only consider static feature similarity but also incorporate time-dependent adjustment mechanisms and task performance data.

[0078] Time-dependent adjustment is achieved using the weighted moving average formula:

[0079] S t =α·T t +(1-α)·S t-1 ;

[0080] Among them, S t Indicates the similarity of tasks at the current time t, T t represents the feature value of the task at the current time t, α is the time-dependent weighting coefficient, which is used to control the weight distribution of current features and historical features, S t-1represents the similarity of tasks at the previous time t-1. Through this formula, the system can balance the current and historical characteristics of tasks, ensuring that the similarity of tasks is reasonably adjusted over time.

[0081] After similarity calculation and time dependency adjustment, the system generates dynamic clustering data, which represents the optimized hierarchical structure of tasks and is dynamically adjusted according to the current performance of the tasks to ensure the accuracy and flexibility of task classification in different scenarios.

[0082] The embodiment of the present application forms preliminary label feature data through a multi-layer semantic aggregation network, uses a similarity matrix for dynamic clustering, and converts the dynamic clustering data into task feature clusters, thereby achieving a deep understanding of task features and precise hierarchical processing, and ensuring that complex dependencies between tasks are effectively represented, laying the foundation for the subsequent determination of the target feedback path.

[0083] In order to extract the common characteristics of tasks based on the clustering results, optionally, in the method for determining the classification labels of tasks provided in an embodiment of the present application, converting dynamic clustering data into task feature clusters includes: merging the features of each task within the same cluster group according to the clustering grouping information of the dynamic clustering data to form a corresponding feature cluster, wherein the clustering grouping information includes at least one of the following: the cluster or group identifier to which each task belongs, the cluster center or representative vector, the similarity mapping between tasks, and the cluster evolution information at different time points; and packaging and outputting each feature cluster to obtain a task feature cluster.

[0084] Specifically, the system first analyzes the dynamic clustering data and extracts cluster grouping information. This information includes the specific cluster identifier for each task, the task vector representing the cluster center, a detailed mapping of inter-task similarities, and the evolution of task clusters at different points in time. Based on this information, the system can locate the set of tasks within each cluster and identify their common characteristics.

[0085] Next, the task features within the same cluster are merged. This process combines the feature vectors of each task to extract a comprehensive vector that summarizes the characteristics of the entire cluster, forming a feature cluster representing the characteristics of each task group. The feature clusters of each task group are packaged and output to form a task feature cluster. A task feature cluster is a high-density data unit containing a series of task information after clustering and feature integration. It not only integrates the deep semantic features of the tasks, but also reflects the similarity and time-adjusted dependencies of the tasks.

[0086] The embodiment of the present application parses dynamic clustering data, extracts clustering grouping information therein, and then merges task features based on the clustering grouping information, thereby simplifying the complex dynamic clustering data into easy-to-process task feature clusters, providing input for the subsequent steps of determining target feedback path data, and promoting the efficient advancement of the entire task label classification process.

[0087] In order to optimize the execution order of tasks to achieve more efficient task classification, optionally, in the method for determining the classification label of the task provided in the embodiment of the present application, the target feedback path data of the task feature cluster is determined in the following manner: combining the task feature similarity matrix and the task dependency, constructing preliminary feedback path data, wherein the preliminary feedback path data reflects the current execution order of the tasks; inputting the preliminary feedback path data into the recursive feedback graph convolutional network, and processing to obtain the target feedback path data.

[0088] Specifically, the feature similarity matrix reveals the similarities between tasks, while task dependencies clarify the execution logic between tasks. Therefore, the construction of preliminary feedback path data is essentially the system's attempt to predict a path sequence suitable for multi-task execution based on the current task features and dependencies. This reflects the basic execution process that tasks should follow before classification. "Feedback" refers to the performance indicators, error information, or other indicators obtained by the system after path execution, which are fed back into the path optimization process to update path weights or execution order.

[0089] The key formula for generating preliminary feedback path data is as follows:

[0090]

[0091] Among them, Path t Indicates the path weight at the current time t, Path t-1 represents the path weight at the previous moment, η is the learning rate, is the gradient of the loss function, providing feedback for path adjustment. This formula indicates that the task path weights will be adaptively adjusted based on the gradient of the execution loss function. This dynamic feedback allows the system to optimize task paths in a timely manner, ensuring they remain optimized in the ever-changing task execution environment.

[0092] It should be noted that a variable feedback path mechanism is used in the construction of the initial feedback path. The variable feedback path mechanism ensures that each task path is optimized in a timely manner according to the current execution status by dynamically adjusting task performance. Among them, task performance refers to execution-level indicators such as task efficiency, success rate, error rate, and duration; system performance-related indicators such as resource consumption and access volume; business indicators such as actual revenue, user satisfaction, or other customized metrics. The objects of dynamic adjustment are path weights, or directly adjusting the specific task order (letting some be executed first or later), as well as the dependency direction / priority between tasks.

[0093] After obtaining the preliminary feedback path data, the preliminary feedback path data is input into the recursive feedback graph convolutional network for processing to obtain more optimized target feedback path data. Figure 4 is a schematic diagram of a recursive feedback graph convolutional network provided according to an embodiment of the present application, such as Figure 4 As shown in the figure, RFGCN performs feature conversion and optimization through a series of convolutional layers, including the first convolution layer, the second convolution layer, and the third convolution layer, and finally outputs the result of the optimal path. It will also provide feedback to the first convolution layer of the network after the third convolution layer, and continuously adjust its internal parameters to optimize the execution path between tasks.

[0094] Among them, the recursive feedback graph convolutional network is able to calculate and adjust the bidirectional dependencies between tasks through multiple rounds of recursive calculations, ensuring that the task path is not only based on the current task performance, but also can capture the deeper dependency structure between tasks through the recursive feedback mechanism. This network uses the characteristics and dependencies of the tasks to gradually refine and adjust the task execution path through multi-layer convolution operations to ensure that the path conforms to the logic of task dependencies and maximizes the efficiency and accuracy of classification. The target feedback path data finally output is the result of deep network optimization. It represents the optimal execution order of tasks after considering task similarity and dependencies, and provides path suggestions for subsequent multi-task multi-label classification.

[0095] When computing recursively, the core operation of the recursive feedback graph convolutional network is expressed by the following formula:

[0096]

[0097] In this formula, H (l+1) represents the task feature matrix of the l+1 layer, D is the degree matrix, which represents the degree of the node, A is the task dependency adjacency matrix, which represents the connection relationship between tasks, and W (l) is a trainable weight matrix, H (l)represents the input feature matrix of layer l, and σ is the activation function. Through this formula, the system performs convolution calculations on the task features of each layer, thereby gradually optimizing the dependencies of the task paths.

[0098] The embodiment of the present application constructs preliminary feedback path data by combining the task feature similarity matrix and task dependency, and then inputs this data into the recursive feedback graph convolutional network for processing to determine the target feedback path data, thereby ensuring the efficiency and accuracy of task classification.

[0099] In order to generate optimized feedback path data that is more in line with the task execution logic, optionally, in the method for determining the classification label of the task provided in an embodiment of the present application, the preliminary feedback path data is input into the recursive feedback graph convolutional network, and the processing to obtain the target feedback path data includes: the recursive feedback graph convolutional network performs convolution calculation on the preliminary feedback path data through the convolution layer in each round of iteration to generate an updated path dependency relationship, generates recursive feedback optimization data based on the updated path dependency relationship, and optimizes the preliminary feedback path data based on the recursive feedback optimization data; after the iteration end condition is reached, the optimization result of the preliminary feedback path data output by the last iteration is determined as the target feedback path data.

[0100] Specifically, at the beginning of each iteration, the recursive feedback graph convolutional network performs a convolution operation based on the current preliminary feedback path data. By adjusting the weight parameters of the convolutional layer, it captures and calculates the direct and indirect dependencies between tasks, generating updated path dependencies. Then, based on these updated path dependencies, the system generates recursive feedback optimization data. This recursive feedback optimization data contains guidance for adjusting the task execution order based on the newly discovered dependencies.

[0101] Next, the recursive feedback graph convolutional network further optimizes the preliminary feedback path data based on the recursive feedback optimization data, adjusting the execution order and dependency weights between tasks. This optimization process is repeated in each iteration until the preset iteration end conditions are met.

[0102] When the iteration termination criteria are met—that is, when the network optimization process no longer significantly improves or the predetermined number of iterations has been reached—the system determines the optimization results of the preliminary feedback path data output from the last iteration as the target feedback path data. The target feedback path data represents the task execution sequence after multiple rounds of deep network optimization. It takes into account the complex dependencies between tasks and reflects real-time task performance and feedback data. Using the target feedback path data ensures efficient and accurate task classification results.

[0103] It's important to note that the core optimization principle of the recursive feedback graph convolutional network lies in its iterative feedback mechanism, which gradually approaches the optimal path through multiple cycles. After each iteration, the system feeds the newly refined path optimization results back to the convolutional network, updating the task path dependency matrix and generating recursive feedback optimization data. This process ensures that the system captures bidirectional dependencies between tasks, ensuring that the task path is not only applicable to the current task but also optimizes the global task execution path.

[0104] The embodiment of the present application uses this cyclic optimization mechanism of the recursive feedback graph convolutional network to not only process the immediate dependencies between tasks, but also gradually deepen the understanding of the high-order dependency structure between tasks, and ultimately output optimized feedback path data.

[0105] In order to ensure that each task can fully utilize its dependency information with other tasks during classification and execution, optionally, in the method for determining the classification label of the task provided in the embodiment of the present application, the dependency data is subjected to dependency matrix decomposition to generate a target dependency matrix, including: splitting the dependency relationship between tasks into a low-rank matrix through a matrix decomposition formula; using the low-rank matrix to capture the explicit dependency data and initial implicit dependency data between tasks, wherein the explicit dependency data is used to reflect the significant correlation between tasks, and the initial implicit dependency data is used to reflect the hidden dependency structure between tasks; iteratively updating the initial implicit dependency matrix through a gradient descent algorithm to obtain the implicit dependency data of the target; and weightedly fusing the explicit dependency data and the implicit dependency data to generate a target dependency matrix.

[0106] Specifically, before matrix decomposition, the dependencies between tasks are recursively modeled. The recursive modeling process is essentially to continuously update, sum, nonlinearly transform, and normalize the task interaction matrix, so that it gradually evolves into a matrix that is closer to the actual dependency structure. The goal of this step is to capture the deep dependency structure between tasks through multiple rounds of calculations. Among them, recursive modeling is not just about establishing a matrix, but performing multiple rounds of iterations based on the initial matrix to eventually obtain an accurate high-order dependency matrix. First, a task interaction matrix is ​​constructed based on the dependencies between tasks. Each element of the task interaction matrix represents the degree of dependency between two tasks. The calculation formula for generating the high-order dependency matrix is ​​as follows:

[0107] A (l+1) =f(A (l) ,W (l) );

[0108] Among them, A (l+1) represents the task dependency matrix after the l+1th round of recursion, A (l) represents the task dependency matrix of round l, W (l)is the recursive dependency weight matrix, and f is a nonlinear activation function used to normalize and adjust task dependencies. Through multiple rounds of recursion, the system gradually optimizes inter-task dependencies, ensuring that each task's dependency path captures global dependency information. After the recursion completes, high-order dependency data is generated. This high-order dependency data can be the "final high-order dependency matrix" and its necessary supplementary information.

[0109] It should be noted that the task dependency matrix and the task interaction matrix can exist in different states of the same data: one is the initial or intermediate state (the task interaction matrix), and the other is the result of recursive and feedback optimization (the task dependency matrix after multiple rounds of recursion). After completing the modeling of high-order dependency relationships, the system uses high-order dependency matrix decomposition technology to decompose the high-order dependency matrix containing all task dependencies into a low-rank matrix using a matrix decomposition formula. This process aims to reduce the complexity of task dependencies, making them easier to understand and process. The generation of a low-rank matrix is ​​essentially a mathematical abstraction of task dependencies, simplifying complex many-to-many dependencies into a few pairs of basic dependency units.

[0110] The next step in recursive modeling is matrix factorization. Specifically, Figure 5 is a flowchart of a high-order dependency matrix decomposition method provided in an embodiment of the present application, such as Figure 5 As shown in the figure, the high-order dependency matrix is ​​input into the decomposition process, and the complex task dependency relationship is decomposed into two parts, the explicit dependency matrix and the implicit dependency matrix, through the matrix decomposition technology, and then input into the task classification model together.

[0111] The core formula of matrix decomposition is as follows:

[0112] M=U∑V T ;

[0113] In this formula, M represents the high-order dependency matrix, U and V are the eigenvector matrices of the tasks, ∑ is a diagonal matrix that represents the dependency strength between tasks, and T is the transpose symbol that represents the transpose matrix of matrix V. Through this matrix decomposition formula, the system can decompose complex dependency relationships into low-rank matrices to capture the explicit and implicit dependencies between tasks.

[0114] Then, based on the low-rank matrix, explicit dependency data and initial implicit dependency data are extracted. Explicit dependency data directly reflects significant connections between tasks through intuitive matrix elements, such as direct pre- and post-dependencies or parallel execution conditions. This data provides clear guidance for task classification. Initial implicit dependency data, on the other hand, focuses on revealing dependencies that are not apparent in the higher-order dependency matrix but have a potential impact on task execution, such as indirect multi-step dependencies or cross-influences. Extracting this data helps the system more comprehensively understand the deep connections between tasks.

[0115] Next, the system uses a gradient descent algorithm to iteratively update the initial implicit dependency matrix to obtain optimized implicit dependency data. This process involves fine-tuning the matrix elements to minimize the gap between the dependency relationships and the actual task execution. This allows the implicit dependency data to more accurately reflect the complex dependencies between tasks, improving the accuracy of task path planning and optimization.

[0116] Among them, the gradient descent optimization formula is:

[0117]

[0118] Among them, M new represents the implicit dependency matrix after optimization, M is the implicit dependency matrix before optimization, η is the learning rate, Represents the gradient of the loss function with respect to the dependency matrix.

[0119] Finally, through weighted fusion, the explicit dependency data is combined with the optimized implicit dependency data to generate a target dependency matrix. Weighted fusion adaptively adjusts the weight ratio of explicit and implicit dependencies, ensuring that both dependencies can dynamically adapt to task requirements in different scenarios. It considers both the direct and obvious dependencies between tasks while also not neglecting the complex connections hidden behind them. The construction of the target dependency matrix provides a comprehensive and accurate description of task dependencies for subsequent classification tasks, enhancing the intelligence and accuracy of the classification process.

[0120] The embodiment of the present application generates a high-order dependency matrix through recursive modeling, then performs dependency matrix decomposition, refines the dependency relationship between tasks into explicit and implicit dependency data, and then uses a gradient descent algorithm to iteratively optimize the implicit dependency matrix. Finally, a target dependency matrix is ​​generated by weighted fusion of explicit and implicit dependency data. The target dependency matrix provides strong support for the global optimization of tasks, ensuring that the classification and execution of each task can fully consider its complex dependency information with other tasks, thereby achieving the effect of improving task processing efficiency and classification accuracy, and optimizing the global task execution path.

[0121] In order to further support task management, optionally, in the method for determining the classification label of the task provided in the embodiment of the present application, the target classification input data is input into the target classification model to obtain the classification label of each task. The method also includes: generating the classification confidence of the task, and determining the classification result of each task based on the classification label and classification confidence of each task; adjusting the execution order of each corresponding task based on the classification label and classification confidence, so that the execution order of each task conforms to the target dependency matrix and target feedback path data.

[0122] Figure 6This is a flow chart of classifying target classification input data according to the target classification model provided in the embodiment of the present application, such as Figure 6 As shown in the figure, starting with the input feature data, the data is processed by a multi-layer neural network and optimized for different classification tasks 1, 2, and 3. Each classification task has its own specific output label, for example, task 1 corresponds to label 1, and so on. Figure 6 It demonstrates how the system uses multi-layer neural networks to process multiple classification tasks simultaneously, and achieves efficient classification of multiple types of labels by adjusting the network structure and parameters.

[0123] After data aggregation is complete, the system feeds the generated target classification input data into the target classification model. The target classification model, based on a multi-layer neural network structure, uses multi-task learning to simultaneously classify multiple labels within the target classification input data. The advantage of multi-task learning lies in its ability to simultaneously optimize the classification accuracy of labels across multiple tasks and improve classification accuracy and efficiency by leveraging information between tasks.

[0124] Specifically, the method for determining a task's classification label not only provides the classification label but also generates a classification confidence score, a metric used to reflect the reliability of the classification label. The classification confidence score can be generated based on the probability distribution of the model output, reflecting the classifier's confidence level that a task belongs to a specific label. The system comprehensively determines the classification result for each task based on the classification label and its corresponding classification confidence score. This result not only includes the classification decision but also carries an assessment of the model's reliability.

[0125] It should be noted that when it is necessary to adjust the order of tasks, verify the authenticity of labels, or balance resources, the classification confidence will be referred to to enhance the reliability of the decision.

[0126] The following formula is used to perform weighted integration of task dependencies to generate the final classification input data:

[0127] C final =α1·D features +α2·M dependence +α3·P path ;

[0128] Among them, C final Represents the final classification input data, D features Represents the task feature cluster data, M dependence represents the optimal dependency matrix of the task, P pathrepresents the optimized feedback path data. α1, α2, and α3 are weighting coefficients used in the integration process to balance the contributions of feature data, dependency data, and path data. Through this formula, the system integrates various core data to generate optimal classification input data, supporting the final task classification.

[0129] After determining the classification results, the system post-processes the classification results and adjusts the execution order of each task based on the classification label and classification confidence to ensure that the classification results of each task are consistent with the global task execution path. This adjustment process takes into account the complexity of dependencies between tasks and the impact of classification results on the execution order. By comparing the classification results with the target dependency matrix and target feedback path data, the system can determine the execution order of tasks and the potential impact of classification confidence on execution efficiency. Specifically, the system tends to execute tasks corresponding to classification results with higher confidence first.

[0130] Among them, the system classifies multi-task labels through the following classification model formula:

[0131] y=f(W x ·X+b);

[0132] Among them, y represents the classification output, W x represents the weight matrix of the classification model, X represents the classification input data, b represents the bias term, and f is the activation function, which can be either Softmax or Sigmoid. This classification model outputs the classification label for each task through deep learning of the task input data, and ensures that each task classification result has optimal classification accuracy under the support of the global dependency structure.

[0133] Ultimately, the system outputs the classification results to the task management module for practical application during task execution. This classification result data includes not only the task's final classification label but also the task execution order and optimal path generated by the system based on the global dependency structure. Through this process, the system ensures the accuracy of task classification and the rationality of execution order, providing a complete solution for determining task classification labels.

[0134] The embodiment of the present application determines the detailed classification results by utilizing the classification confidence of the task and combining the classification label of each task. Then, the execution order of the tasks is adjusted based on this classification information to ensure that it is consistent with the optimized target dependency matrix and feedback path data, thereby achieving the effect of enhancing the reliability of task classification, optimizing the task execution process, and improving overall management efficiency.

[0135] The present application also provides an optional method for determining a task classification label. Figure 7This is a flow chart of a method for determining a classification label of an optional task provided in an embodiment of the present application, such as Figure 7 As shown in the figure, its contents include: starting from data collection and preprocessing, through feature extraction, multi-layer semantic aggregation, dynamic collaborative clustering, recursive feedback path optimization, high-order dependency modeling and matrix decomposition, to the final classification and result output.

[0136] Specifically, the steps of data collection and preprocessing include data collection, denoising and standardization. After processing, the target knowledge graph data and target task association data can be obtained, which constitute the input of feature extraction.

[0137] The feature extraction step can be multimodal semantic preprocessing. The process of multimodal semantic preprocessing is knowledge graph embedding processing, task relationship feature extraction and weighted fusion processing, and then multimodal semantic feature data is generated. The multimodal semantic feature data not only represents the semantic relationship and dependency between tasks, but also ensures the quality and consistency of the data.

[0138] The multi-layer semantic aggregation step is to perform multi-layer semantic aggregation on the multimodal semantic feature data, extract the deep semantic features of the task, and obtain preliminary label feature data.

[0139] The dynamic collaborative clustering step dynamically clusters the preliminary label feature data using a dynamic collaborative clustering algorithm to generate dynamic cluster data and task feature clusters. During this process, task features are dynamically adjusted in real time to ensure accurate grouping of tasks at different time points. The hierarchical structure between tasks is continuously optimized based on task performance and similarity, thereby improving the system's responsiveness and real-time classification capabilities.

[0140] The recursive feedback path optimization step is based on task feature clusters. It dynamically optimizes the execution path of the task through a variable feedback path mechanism and a recursive feedback graph convolutional network to generate target feedback path data. This step processes the bidirectional dependencies between tasks through recursive calculations, realizes the dynamic optimization of the task classification path, and captures the deep dependencies between tasks through multiple rounds of convolution operations, ensuring that each task always maintains the optimal path adjustment and dependency management during execution.

[0141] The system can respond to task changes in real time, adaptively adjusting task paths and target classification models, and automatically optimizing model parameters based on task changes. This mechanism significantly improves the system's ability to handle dynamic task environments, enabling the target classification model to maintain efficient and stable classification accuracy as tasks increase, decrease, or change, reducing the need for model retraining. Compared to existing technologies, this reduces the time and computing resources required for retraining, enabling more flexible and efficient task classification.

[0142] Next, the high-order dependency modeling and matrix decomposition step is based on the target feedback path data. Through the high-order interactive dependency model (H technology) and high-order dependency matrix decomposition (HDMD), the deep dependencies between tasks are captured to obtain the target dependency matrix. This step aims to further optimize the dependency structure between tasks, ensuring that each task not only depends on the current task performance, but also has an in-depth understanding of its high-order dependencies in the entire task network. Through recursive modeling and matrix decomposition, the system can capture complex task dependencies, ensuring that the classification model can identify potential task associations when processing large-scale tasks, and improving the overall accuracy of the classification results.

[0143] The system uses a recursive feedback graph convolutional network and high-order dependency matrix decomposition technology to address the inadequate handling of complex task dependencies in existing technologies. It can recursively capture the deep interactions and dependencies between tasks and generate optimal task paths. This recursive modeling technology significantly improves the accuracy of task classification and the efficiency of task path optimization. Especially when dealing with high-dimensional, multi-dependent tasks, the system can identify deeper correlations, addressing the limitations of existing technologies in handling complex dependencies.

[0144] Finally, the classification and result output step refers to combining the target dependency matrix, target feedback path data, task feature clusters and dynamic clustering data to generate target classification input data, inputting the target classification input data into the target classification model, and having the target classification model output the classification label for each task. The classification result of each task is then determined in combination with the classification confidence of the task, and the task is executed according to the classification result, which can ensure the accuracy and efficiency of task execution.

[0145] Furthermore, the target classification model features automated training and self-calibration capabilities. By continuously adjusting and optimizing the model, the need for external intervention is reduced. The system can self-calibrate based on historical data and real-time task performance, ensuring accurate and consistent classification results.

[0146] The embodiment of the present application obtains multimodal semantic feature data by performing data collection, preprocessing and feature extraction, and performs multi-layer semantic aggregation on the multimodal semantic feature data, as well as automatically detects and optimizes the classification results through the self-correction mechanism of dynamic collaborative clustering and the recursive feedback path optimization of the recursive feedback network, so that the system can optimize the classification effects of multiple tasks at the same time through the improved multi-task learning framework. During the processing, the combination of the loss function of multi-label classification and the multi-task loss function enables the system to effectively handle the parallel classification of complex tasks, improves the overall classification efficiency and accuracy, reduces the inefficiency and error problems of related technologies, and enhances the processing power and accuracy of the system when processing big data. Compared with related technologies, the present invention can perform well in larger-scale task classification scenarios and maintain an efficient automated workflow.

[0147] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0148] Example 2

[0149] The present application also provides a device for determining a task classification label. It should be noted that the device for determining a task classification label in the present application can be used to execute the method for determining a task classification label provided in the present application. The following describes the device for determining a task classification label provided in the present application.

[0150] According to an embodiment of the present application, a device for implementing the method for determining the classification label of the above task is also provided. Figure 8 is a schematic diagram of a device for determining a classification label of a task according to an embodiment of the present application, such as Figure 8 As shown, the device includes:

[0151] A preprocessing unit 801 is used to perform multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task;

[0152] A clustering unit 802 is used to extract deep semantic features from the multimodal semantic feature data and perform clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters;

[0153] A dependency processing unit 803 is configured to generate target feedback path data for a task feature cluster, perform dependency relationship modeling based on the target feedback path data to generate dependency data, and perform dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks;

[0154] An input data generating unit 804 is used to combine the target dependency matrix, the target feedback path data, the task feature cluster and the dynamic clustering data to generate target classification input data;

[0155] The label generation unit 805 is used to input the target classification input data into the target classification model to obtain the classification label of each task.

[0156] In the embodiment of the present application, the preprocessing unit 801 performs multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data of each task; the clustering unit 802 extracts deep semantic features in the multimodal semantic feature data and performs clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters; the dependency processing unit 803 generates target feedback path data of the task feature cluster, performs dependency relationship modeling based on the target feedback path data, generates dependency data, performs dependency matrix decomposition on the dependency data, and generates a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; the input data generation unit 80 4. Combine the target dependency matrix, target feedback path data, task feature clusters and dynamic clustering data to generate target classification input data; the label generation unit 805 inputs the target classification input data into the target classification model to obtain the classification label of each task, which solves the problem of low efficiency and accuracy in classifying labels of multiple tasks in related technologies. By performing multimodal semantic preprocessing, clustering processing, high-order dependency relationship modeling and matrix decomposition on the target knowledge graph data and target task association data of the task, the target classification input data is obtained, and the target classification input data is input into the target classification model to obtain the classification label of each task, thereby achieving efficient and accurate classification effect of labels for complex tasks.

[0157] Optionally, in the device for determining the classification label of a task provided in an embodiment of the present application, the device also includes: an initial data extraction module, used to extract initial knowledge graph data and initial task association data from the historical process knowledge base; a denoising module, used to perform denoising processing on the initial knowledge graph data and the initial task association data to generate denoised knowledge graph data and denoised task association data; a standardization module, used to perform target processing on the denoised knowledge graph data and the denoised task association data to obtain target knowledge graph data and target task association data, wherein the target processing includes at least one of the following: a consistency verification module, used to perform consistency verification on date and task identifier data.

[0158] Optionally, in the device for determining the classification label of the task provided in the embodiment of the present application, the preprocessing unit 801 includes: an embedding module, which is used to perform knowledge graph embedding processing on the target knowledge graph data to generate task semantic embedding data, wherein the knowledge graph embedding processing is to extract features of entities and relationships in the target knowledge graph data through a multi-layer convolutional neural network, and convert the semantic relationship between tasks into a numerical representation; a relational convolution processing module, which is used to input the target task association data into the relational convolutional neural network to generate task relationship feature data; a weighted fusion module, which is used to perform weighted fusion on the task semantic embedding data and the task relationship feature data to obtain multimodal semantic feature data.

[0159] Optionally, in the device for determining the classification label of the task provided in the embodiment of the present application, the clustering unit 802 includes: a semantic extraction module, which is used to extract the deep semantic features of the task from the multimodal semantic feature data through a multi-layer semantic aggregation network to obtain preliminary label feature data; a similarity calculation module, which is used to calculate the feature similarity between each two tasks based on the preliminary label feature data, and generate a task feature similarity matrix; a clustering module, which is used to cluster all tasks according to the task feature similarity matrix to obtain clustering results; a dynamic adjustment module, which is used to dynamically adjust the clustering results according to the feature changes of the task at different times to generate dynamic clustering data, wherein the dynamic clustering data is the preliminary result of clustering the deep semantic features, including the feature changes of the task at different times; a conversion module, which is used to convert the dynamic clustering data into a task feature cluster.

[0160] Optionally, in the device for determining the classification label of the task provided in the embodiment of the present application, the conversion module includes: a feature merging module, which is used to merge the features of each task in the same cluster group according to the cluster grouping information of the dynamic clustering data to form a corresponding feature cluster, wherein the cluster grouping information includes at least one of the following: the cluster or group identifier to which each task belongs, the cluster center or representative vector, the similarity mapping between tasks, and the cluster evolution information at different time points; a packaging module, which is used to package and output each feature cluster to obtain a task feature cluster.

[0161] Optionally, in the device for determining the classification label of a task provided in an embodiment of the present application, the dependency processing unit 803 includes: a preliminary feedback module, used to combine the task feature similarity matrix and the task dependency relationship to construct preliminary feedback path data, wherein the preliminary feedback path data reflects the current execution order of the task; and a recursive processing module, used to input the preliminary feedback path data into a recursive feedback graph convolutional network to process and obtain target feedback path data.

[0162] Optionally, in the device for determining the classification label of the task provided in an embodiment of the present application, the recursive processing module includes: a convolution calculation module, which is used for the recursive feedback graph convolution network to perform convolution calculation on the preliminary feedback path data through the convolution layer in each round of iteration, generate an updated path dependency relationship, generate recursive feedback optimization data according to the updated path dependency relationship, and optimize the preliminary feedback path data according to the recursive feedback optimization data; a target feedback path data determination module, which is used to determine the optimization result of the preliminary feedback path data output by the last iteration as the target feedback path data after the iteration end condition is met.

[0163] Optionally, in the device for determining the classification label of the task provided in the embodiment of the present application, the dependency processing unit 803 also includes: a matrix decomposition module, which is used to split the dependency relationship between tasks into a low-rank matrix through a matrix decomposition formula; a data capture module, which is used to use the low-rank matrix to capture the explicit dependency data and initial implicit dependency data between tasks, wherein the explicit dependency data is used to reflect the significant correlation between tasks, and the initial implicit dependency data is used to reflect the hidden dependency structure between tasks; an iterative update module, which is used to iteratively update the initial implicit dependency matrix through a gradient descent algorithm to obtain the implicit dependency data of the target; and a target dependency matrix generation module, which is used to weightedly fuse the explicit dependency data and the implicit dependency data to generate a target dependency matrix.

[0164] Optionally, in the device for determining the classification label of a task provided in an embodiment of the present application, the label generation unit 805 includes: a classification result determination module, used to generate the classification confidence of the task, and determine the classification result of each task based on the classification label and classification confidence of each task; and a sequence adjustment module, used to adjust the execution order of each corresponding task based on the classification label and classification confidence, so that the execution order of each task conforms to the target dependency matrix and target feedback path data.

[0165] It should be noted that the above modules correspond to steps S201 to S205 in Example 1, and the examples and application scenarios implemented by each module and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules or units can be hardware components or software components stored in a memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be run as part of the device in the computer terminal 10 provided in Example 1.

[0166] Example 3

[0167] An embodiment of the present application may provide an electronic device, Figure 9 This is a structural block diagram of an electronic device according to an embodiment of the present application. Figure 9As shown, the electronic device may include: one or more ( Figure 9 Only one is shown) processor 902, memory 904, storage controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0168] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the methods and devices in the embodiments of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, implementing the above-mentioned method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the terminal via a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0169] The processor can call the information and application programs stored in the memory through the transmission device to perform the following steps: perform multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task; extract deep semantic features in the multimodal semantic feature data, and cluster the deep semantic features to obtain dynamic clustering data and task feature clusters; generate target feedback path data for the task feature cluster, perform dependency modeling based on the target feedback path data to generate dependency data, perform dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; combine the target dependency matrix, target feedback path data, task feature clusters and dynamic clustering data to generate target classification input data; input the target classification input data into the target classification model to obtain a classification label for each task.

[0170] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: before performing multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks, the method also includes: extracting initial knowledge graph data and initial task association data from the historical process knowledge base; performing denoising processing on the initial knowledge graph data and initial task association data to generate denoised knowledge graph data and denoised task association data; performing target processing on the denoised knowledge graph data and denoised task association data to obtain target knowledge graph data and target task association data, wherein the target processing includes at least one of the following: performing consistency verification on date and task identifier data.

[0171] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: performing multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task, including: performing knowledge graph embedding processing on the target knowledge graph data to generate task semantic embedding data, wherein the knowledge graph embedding processing is to extract features of entities and relationships in the target knowledge graph data through a multi-layer convolutional neural network, and convert the semantic relationships between tasks into numerical representations; inputting the target task association data into the relational convolutional neural network to generate task relationship feature data; and performing weighted fusion on the task semantic embedding data and the task relationship feature data to obtain multimodal semantic feature data.

[0172] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: extracting deep semantic features from the multimodal semantic feature data, and clustering the deep semantic features to obtain dynamic clustering data and task feature clusters, including: extracting deep semantic features of the task from the multimodal semantic feature data through a multi-layer semantic aggregation network to obtain preliminary label feature data; calculating the feature similarity between each two tasks based on the preliminary label feature data, and generating a task feature similarity matrix; clustering all tasks based on the task feature similarity matrix to obtain clustering results; dynamically adjusting the clustering results based on the feature changes of the tasks at different times to generate dynamic clustering data, wherein the dynamic clustering data is the preliminary result of clustering the deep semantic features, including the feature changes of the tasks at different times; converting the dynamic clustering data into task feature clusters.

[0173] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: converting the dynamic clustering data into a task feature cluster includes: merging the features of each task in the same cluster group according to the cluster grouping information of the dynamic clustering data to form a corresponding feature cluster, wherein the cluster grouping information includes at least one of the following: the cluster or group identifier to which each task belongs, the cluster center or representative vector, the similarity mapping between tasks, and the cluster evolution information at different time points; packaging and outputting each feature cluster to obtain a task feature cluster.

[0174] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: the target feedback path data of the task feature cluster is determined by: combining the task feature similarity matrix and the task dependency to construct preliminary feedback path data, wherein the preliminary feedback path data reflects the current execution order of the task; the preliminary feedback path data is input into the recursive feedback graph convolutional network, and processed to obtain the target feedback path data.

[0175] The processor can also call the information and application stored in the memory through the transmission device to perform the following steps: inputting the preliminary feedback path data into the recursive feedback graph convolutional network, and processing to obtain the target feedback path data includes: the recursive feedback graph convolutional network performs convolution calculation on the preliminary feedback path data through the convolution layer in each round of iteration to generate an updated path dependency relationship, generates recursive feedback optimization data based on the updated path dependency relationship, and optimizes the preliminary feedback path data based on the recursive feedback optimization data; after the iteration end condition is met, the optimization result of the preliminary feedback path data outputted by the last iteration is determined as the target feedback path data.

[0176] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: performing dependency matrix decomposition on the dependency data to generate a target dependency matrix, including: splitting the dependency relationship between tasks into a low-rank matrix through a matrix decomposition formula; using the low-rank matrix to capture the explicit dependency data and initial implicit dependency data between tasks, wherein the explicit dependency data is used to reflect the significant association between tasks, and the initial implicit dependency data is used to reflect the hidden dependency structure between tasks; iteratively updating the initial implicit dependency matrix through a gradient descent algorithm to obtain the implicit dependency data of the target; weighted fusion of the explicit dependency data and the implicit dependency data to generate the target dependency matrix.

[0177] The processor can also call the information and application programs stored in the memory through the transmission device to perform the following steps: input the target classification input data into the target classification model to obtain the classification label of each task, and the method also includes: generating the classification confidence of the task, and determining the classification result of each task according to the classification label and classification confidence of each task; adjusting the execution order of each corresponding task according to the classification label and classification confidence, so that the execution order of each task conforms to the target dependency matrix and target feedback path data.

[0178] According to the embodiment of the present application, multimodal semantic preprocessing is performed on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data of each task; deep semantic features in the multimodal semantic feature data are extracted, and the deep semantic features are clustered to obtain dynamic clustering data and task feature clusters; target feedback path data of the task feature cluster is generated, dependency modeling is performed based on the target feedback path data to generate dependency data, dependency matrix decomposition is performed on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; the target dependency matrix, target feedback path data and target feedback data are combined into a target dependency matrix. Path data, task feature clusters and dynamic clustering data are combined to generate target classification input data; the target classification input data is input into the target classification model to obtain the classification label of each task, which solves the problem of low efficiency and accuracy in classifying labels of multiple tasks in related technologies. The target classification input data is obtained by performing multimodal semantic preprocessing, clustering processing, high-order dependency modeling and matrix decomposition on the target knowledge graph data and target task association data of the task, and the target classification input data is input into the target classification model to obtain the classification label of each task, achieving efficient and accurate classification effect for the labels of complex tasks.

[0179] It can be understood by those skilled in the art that Figure 9 The structure shown is for illustration only, and the electronic device may also be a terminal device such as a smart phone, a tablet computer, a PDA, a mobile Internet device (MID), or a PAD. Figure 9 It does not limit the structure of the above electronic device. For example, the electronic device may also include Figure 9 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 9 Different configurations shown.

[0180] A person skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0181] Example 4

[0182] The embodiment of the present application further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the method for determining the classification label of the task provided in the first embodiment.

[0183] Optionally, in this embodiment, the above-mentioned storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0184] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing the steps of the method for determining a classification label for a task.

[0185] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0186] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0187] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0188] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0189] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0190] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0191] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for determining a classification label of a task, characterized in that: include: Perform multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task; Extracting deep semantic features from the multimodal semantic feature data, and performing clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters; generating target feedback path data of the task feature cluster, performing dependency relationship modeling based on the target feedback path data to generate dependency data, and performing dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; combining the target dependency matrix, the target feedback path data, the task feature cluster, and the dynamic clustering data to generate target classification input data; The target classification input data is input into the target classification model to obtain the classification label of each task.

2. The method for determining the classification label of a task according to claim 1, characterized in that: Before performing multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks, the method further includes: Extract initial knowledge graph data and initial task association data from the historical process knowledge base; Performing denoising on the initial knowledge graph data and the initial task association data to generate denoised knowledge graph data and denoised task association data; Target processing is performed on the denoised knowledge graph data and the denoised task association data to obtain target knowledge graph data and target task association data, wherein the target processing includes at least one of the following: consistency verification of date and task identifier data.

3. The method for determining a task classification label according to claim 1, wherein: The multimodal semantic preprocessing of the target knowledge graph data and target task association data of multiple tasks to obtain the multimodal semantic feature data of each task includes: Performing knowledge graph embedding processing on the target knowledge graph data to generate task semantic embedding data, wherein the knowledge graph embedding processing is to extract features of entities and relationships in the target knowledge graph data through a multi-layer convolutional neural network, and convert the semantic relationships between tasks into numerical representations; Inputting the target task association data into a relational convolutional neural network to generate task relationship feature data; The task semantic embedding data and the task relationship feature data are weightedly fused to obtain the multimodal semantic feature data.

4. The method for determining a task classification label according to claim 1, wherein: Extracting deep semantic features from the multimodal semantic feature data and clustering the deep semantic features to obtain dynamic clustering data and task feature clusters includes: Extracting deep semantic features of the task from the multimodal semantic feature data through a multi-layer semantic aggregation network to obtain preliminary label feature data; Calculating the feature similarity between each two tasks based on the preliminary label feature data, and generating a task feature similarity matrix; Clustering all tasks according to the task feature similarity matrix to obtain clustering results; Dynamically adjusting the clustering results according to feature changes of the task at different times to generate the dynamic clustering data, wherein the dynamic clustering data is a preliminary result of clustering the deep semantic features, including feature changes of the task at different times; The dynamic clustering data is converted into task feature clusters.

5. The method for determining a task classification label according to claim 4, characterized in that: Converting the dynamic clustering data into task feature clusters includes: Based on the cluster grouping information of the dynamic clustering data, the features of each task in the same cluster group are merged to form a corresponding feature cluster, wherein the cluster grouping information includes at least one of the following: the cluster or group identifier to which each task belongs, the cluster center or representative vector, the similarity mapping between tasks, and the cluster evolution information at different time points; Each of the feature clusters is packaged and output to obtain the task feature cluster.

6. The method for determining a task classification label according to claim 1, wherein: The target feedback path data of the task feature cluster is determined by: Combining the task feature similarity matrix and the task dependency relationship, constructing preliminary feedback path data, wherein the preliminary feedback path data reflects the current execution order of the tasks; The preliminary feedback path data is input into a recursive feedback graph convolutional network and processed to obtain the target feedback path data.

7. The method for determining a task classification label according to claim 6, characterized in that: Inputting the preliminary feedback path data into the recursive feedback graph convolutional network to obtain target feedback path data includes: The recursive feedback graph convolutional network performs convolution calculation on the preliminary feedback path data through a convolution layer in each iteration to generate an updated path dependency relationship, generates recursive feedback optimization data based on the updated path dependency relationship, and optimizes the preliminary feedback path data based on the recursive feedback optimization data; After the iteration end condition is reached, the optimization result of the preliminary feedback path data outputted in the last iteration is determined as the target feedback path data.

8. The method for determining a task classification label according to claim 1, wherein: Performing dependency matrix decomposition on the dependency data to generate a target dependency matrix includes: Split the dependencies between tasks into low-rank matrices using matrix decomposition formula; Using the low-rank matrix to capture explicit dependency data and initial implicit dependency data between tasks, wherein the explicit dependency data is used to reflect the significant association between tasks, and the initial implicit dependency data is used to reflect the hidden dependency structure between tasks; The initial implicit dependency matrix is ​​iteratively updated through the gradient descent algorithm to obtain the implicit dependency data of the target; The explicit dependency data and the implicit dependency data are weightedly fused to generate the target dependency matrix.

9. The method for determining a task classification label according to claim 1, wherein: After inputting the target classification input data into the target classification model to obtain a classification label for each task, the method further includes: Generate a classification confidence of the task, and determine a classification result of each task according to the classification label and the classification confidence of each task; The execution order of each corresponding task is adjusted according to the classification label and the classification confidence, so that the execution order of each task conforms to the target dependency matrix and the target feedback path data.

10. A device for determining a classification label of a task, characterized in that: include: A preprocessing unit is used to perform multimodal semantic preprocessing on the target knowledge graph data and target task association data of multiple tasks to obtain multimodal semantic feature data for each task; A clustering unit, configured to extract deep semantic features from the multimodal semantic feature data and perform clustering processing on the deep semantic features to obtain dynamic clustering data and task feature clusters; a dependency processing unit, configured to generate target feedback path data of the task feature cluster, perform dependency relationship modeling based on the target feedback path data to generate dependency data, and perform dependency matrix decomposition on the dependency data to generate a target dependency matrix, wherein the target feedback path data represents the execution order of the tasks; an input data generating unit, configured to combine the target dependency matrix, the target feedback path data, the task feature cluster, and the dynamic clustering data to generate target classification input data; The label generation unit is used to input the target classification input data into the target classification model to obtain a classification label for each task.

11. An electronic device, characterized in that: include: a memory storing an executable program; A processor, configured to run the program, wherein the program, when running, executes the method for determining the classification label of the task described in any one of claims 1 to 9.

Images