Task processing method and device

Abstract

The invention provides a task processing method and device, and the method comprises the steps: determining a plurality of to-be-processed tasks and the input and output demands of each task based on the semantic information of a task description text; determining a resource demand of each task based on the input and output demand of each task; determining a target resource of each task from idle resources based on the resource demand of each task; and executing each task based on the target resource of each task to obtain a processing result of each task. According to the task processing method and device provided by the invention, the task can be automatically processed in the whole process, the dependence on manpower is reduced, the labor cost is reduced, and the processing speed is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a task processing method and device. Background Art

[0002] Task processing refers to the process by which a computer or artificial intelligence system performs a series of operations or calculations to complete a task for a specific goal or requirement. In the context of content review, task processing specifically refers to the system identifying, classifying or filtering input text, images and other content according to predefined rules or models to achieve the purpose of review. This process involves multiple links such as data reception, preprocessing, feature extraction, model reasoning and output of results.

[0003] At present, the audit system built on traditional artificial intelligence (such as LSTM, CNN, etc.) requires a large amount of pre-training data to adapt to different scenarios, such as text, pictures, etc. The data requirements for each scenario are different. For example, text audits need to cover different languages ​​and styles, and image audits need to include diverse visual elements. The core of this audit system is to train on large-scale annotated data through supervised learning models, but because the models are mainly designed for specific tasks, they can often only master a single capability. For example, after training a CNN model for text auditing, it can only effectively process text content of a specific category and is powerless for other data types. Summary of the invention

[0004] The present invention provides a task processing method and device to solve the defects in the prior art.

[0005] The present invention provides a task processing method, comprising the following steps: Based on the semantic information of the task description text, multiple tasks to be processed and the input and output requirements of each task are determined; Determine the resource requirements of each task based on the input and output requirements of each task; Based on the resource requirements of each task, determine the target resources for each task from the idle resources; Based on the target resources of each task, each task is executed to obtain the processing results of each task.

[0006] According to a task processing method provided by the present invention, the method of determining a plurality of tasks to be processed and the input and output requirements of each task based on the semantic information of the task description text includes: Determining a task scenario based on semantic information of the task description text; Based on the task scenario, extracting multiple key elements associated with the task scenario and the logical relationship between the key elements from the task description text; Based on the key elements and the logical relationships between the key elements, the plurality of tasks to be processed and the input and output requirements of each task are determined.

[0007] According to a task processing method provided by the present invention, the determining of the plurality of tasks to be processed and the input and output requirements of each task based on the key elements and the logical relationships between the key elements includes: Based on the key elements and the logical relationships between the key elements, determining the plurality of tasks to be processed and a processing target of each task; Based on the processing objectives of each task, determine the input and output requirements of each task.

[0008] According to a task processing method provided by the present invention, executing each task based on the target resource of each task to obtain the processing result of each task includes: Determine the execution order of each task based on the input and output requirements of each task; According to the execution order of each task, each task is executed in turn based on the target resources of each task to obtain the processing results of each task.

[0009] According to a task processing method provided by the present invention, in the process of executing each task based on the target resource of each task, the method further includes: If there is a newly added idle resource whose performance is better than the performance of the target resource of any task, the newly added idle resource is allocated to the corresponding task.

[0010] According to a task processing method provided by the present invention, the target resource of each task includes multiple modules; The executing of each task based on the target resource of each task and obtaining the processing result of each task include: Optimize the parameters of the next module based on the accuracy of the processing results of the current module; Based on the optimized next module, the processing result of the current module is applied to obtain the processing result of the optimized next module.

[0011] According to a task processing method provided by the present invention, the step of obtaining the processing result of each task further includes: Based on the difference between the processing results of each task and the corresponding actual results, the target resources of each task are optimized.

[0012] The present invention also provides a task processing device, comprising the following modules: A task determination unit, used for determining a plurality of tasks to be processed and input and output requirements of each task based on semantic information of the task description text; A resource determination unit, used to determine the resource requirements of each task based on the input and output requirements of each task; A resource screening unit, used to determine the target resource of each task from the idle resources based on the resource requirements of each task; The task processing unit is used to execute each task based on the target resources of each task and obtain the processing results of each task.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the task processing methods described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the task processing method described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned task processing methods.

[0016] The task processing method and device provided by the present invention determine multiple tasks and the input and output requirements of each task through the semantic information of the task description text to ensure that each task is correctly understood. At the same time, according to the input and output requirements of each task, a suitable resource is dynamically selected from the idle resources as the target resource to ensure that the resources are fully utilized and avoid resource waste. At the same time, the present invention realizes full-process automated task processing, reduces dependence on manual labor, reduces labor costs, and improves processing speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flowchart of the task processing method provided by the present invention.

[0019] Figure 2 It is a structural schematic diagram of the intelligent orchestration system provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the execution logic of the orchestration unit provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of the task processing device provided by the present invention.

[0022] Figure 5It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] At present, the audit system built on traditional artificial intelligence (such as LSTM, CNN, etc.) requires a large amount of pre-training data to adapt to different scenarios, such as text, pictures, etc. The data requirements for each scenario are different. For example, text audits need to cover different languages ​​and styles, and image audits need to include diverse visual elements. The core of this audit system is to train on large-scale annotated data through supervised learning models, but because the models are mainly designed for specific tasks, they can often only master a single capability. For example, after training a CNN model for text auditing, it can only effectively process text content of a specific category and is powerless for other data types.

[0025] To adapt to different audit scenarios, it is necessary to optimize the model parameters based on the data sets of different audit scenarios. However, each optimization iteration takes a long time and is inefficient. In addition, the model has very high requirements for the quality of the data set. Labeling errors or uneven data distribution will significantly affect the performance of the model. In addition, the trained model is usually highly specialized and lacks general capabilities. Therefore, retraining or transfer learning may be required when dealing with complex and changing scenarios, further extending the development cycle.

[0026] In addition, traditional methods also rely on systems built with rules and expert knowledge for auditing. However, since rules and expert knowledge are based on existing experience and specific business logic, when faced with changing business needs or new application scenarios, the above systems need to be adjusted and updated a lot. Each time a new business rule is introduced or an existing rule is improved, it may involve complex logic adjustments and testing, which will greatly increase the maintenance cost and implementation difficulty of the system.

[0027] Secondly, rules and expert knowledge bases are usually derived from the long-term experience accumulation of domain experts and manually written in a fixed format. When business processes or regulatory policies change, experts must review all relevant rules and update them manually, which is not only prone to errors, but also requires a lot of manpower investment, greatly limiting the flexibility and maintainability of the system.

[0028] In addition, when the system constructed by rules and expert knowledge is first designed, it is often customized for specific business scenarios to solve single problems. However, as the complexity of the business increases, solutions based on single-point problem solving are difficult to adapt to complex and changing global needs. For example, in the fields of medicine and finance, rules may involve multiple different processes, departments or systems. It is difficult to achieve comprehensive coverage across departments and scenarios relying solely on expert knowledge and fixed rules. Therefore, in actual applications, the above systems can often only solve local problems and cannot achieve global automated review or decision support.

[0029] Furthermore, the updating and expansion of the system requires continuous investment of a large amount of manpower, and due to the complex relationship between the rules in the system, when adding new rules, it is also necessary to evaluate their impact on existing rules to avoid conflicts and contradictions, which makes the rule system unsustainable when faced with dynamically changing business needs and complex scenarios.

[0030] In this regard, the present invention provides a task processing method, which determines multiple tasks and the input and output requirements of each task through the semantic information of the task description text to ensure that each task is correctly understood. At the same time, according to the input and output requirements of each task, a suitable resource is dynamically selected from the idle resources as the target resource to ensure that the resources are fully utilized and avoid resource waste.

[0031] The method can be applied to task processing in different fields, such as medical record quality inspection tasks, risk control tasks, auxiliary diagnosis and treatment tasks, medical review tasks, automated testing tasks, image processing tasks, text processing tasks, etc. In order to facilitate understanding of the technical solution of the present invention, the following embodiments are all described by taking the application in medical review as an example.

[0032] Figure 1 is a flowchart of the task processing method provided by the present invention, such as Figure 1 As shown, the method includes step 110 , step 120 , step 130 and step 140 .

[0033] Step 110: Based on the semantic information of the task description text, determine a plurality of tasks to be processed and the input and output requirements of each task.

[0034] Specifically, the task description text refers to the text information describing the task to be performed, and the task description text usually includes the purpose and requirements of the task to be performed. Considering that the task to be performed pointed to by the task description text is usually complex, and may involve a large amount of data and complex logic, complex tasks to be performed may be more prone to errors during execution, such as any small error may lead to the failure of the entire task. In this regard, the embodiment of the present invention decomposes the task to be performed into multiple simple tasks based on the semantic information of the task description text. Among these simple tasks, some tasks can be executed in parallel, thereby improving the task processing speed; there are dependencies between some tasks, so they need to be executed in sequence, that is, not all tasks may need to occupy resources in the same time period, thereby reducing resource occupation. Furthermore, each task has a smaller amount of data processing involved than the task to be performed, and thus requires fewer resources, which can maximize the utilization of resources under limited resources.

[0035] For example, if the resources required for the task to be executed are greater than the current idle resources, it means that the current idle resources cannot execute the task to be executed, while the resources required for the decomposed task are less than the current idle resources, it means that the current idle resources can execute the decomposed task, thereby achieving full utilization of the current idle resources.

[0036] Since the execution logic corresponding to each task to be processed may be different, each task to be processed may correspond to different input and output requirements, which are used to characterize the input and output forms, such as input data type and output result form. In medical scenarios, input data may be in various forms, including structured data such as laboratory test results and pathology reports, as well as unstructured data such as patient medical history, doctor's symptom description, and imaging data (such as CT scans, MRI images, etc.).

[0037] For example, in the medical review scenario, the task description text is "review all patients' medical records and determine whether they meet the reimbursement conditions". Based on the task description text, it can be decomposed into multiple tasks, such as "extracting the patient's basic information", "identifying key information in the medical record", "determining whether it meets specific reimbursement conditions", etc. The input corresponding to the task "extracting the patient's basic information" is the patient's medical record file, and the output is the patient's basic information; the input corresponding to the task "identifying key information in the medical record" is the patient's complete medical record, and the output is the key information; the input of the task "determining whether it meets specific reimbursement conditions" is the patient's medical record, and the output is the judgment result used to characterize whether the reimbursement conditions are met.

[0038] Exemplarily, the task description text may be input into a large model, which usually uses natural language processing (NLP) technology to parse the task description text based on semantic information to determine multiple tasks to be processed and the input and output requirements of each task.

[0039] Step 120: Determine the resource requirements of each task based on the input and output requirements of each task.

[0040] Specifically, resource requirements refer to the resources needed to complete the task. The resources here can include computing resources (such as CPU, GPU, memory), storage resources (such as hard disk space, database capacity), human resources (such as processing time, professional skills and time), software resources (such as specific algorithms, tools or platforms), etc.

[0041] Consider that the size and complexity of the input data will affect the time and computing resources required to process the data. For example, if the input data is a large number of medical records, then more storage space and stronger computing power may be required to process the data efficiently. In addition, the complexity and format requirements of the output data may also affect the resource requirements. For example, if the output needs to be highly structured data, then more time may be required to ensure the accuracy and completeness of the data.

[0042] In addition, input and output requirements often imply the processing logic and algorithms that the task needs to execute, and different logic and algorithms have different resource requirements. For example, some complex natural language processing tasks may require a high-performance GPU to accelerate calculations.

[0043] For example, the input of the task is the patient's medical record document, and the output is the key information to be identified. Based on the input and output requirements, the computing resources (such as CPU, GPU, memory), storage resources (such as hard disk space, database capacity), etc. required for the model to process the medical records are determined, and then the appropriate model is selected to perform the corresponding task.

[0044] Step 130: Based on the resource requirements of each task, determine the target resources of each task from the idle resources.

[0045] Specifically, idle resources refer to resources that are not currently occupied and can be used to perform tasks. Here, resources can be understood as models that can be used to perform tasks. Target resources refer to resources in idle resources that can successfully perform corresponding tasks.

[0046] Considering that each task has different resource requirements, different resource requirements may correspond to different resource types and resource amounts. For example, in a medical review scenario, extracting basic patient information may rely mainly on database access and data processing capabilities, while identifying key information in medical records may require high-performance computing resources and memory support.

[0047] Therefore, according to the resource requirements of each task, a target resource matching the task can be determined from the idle resources. The target resource can meet the resource type requirement and resource quantity requirement of the corresponding task, and thus the target resource can ensure the smooth execution of the task.

[0048] Among them, the above resources can be machine learning models, such as deep neural network models, reinforcement learning models, etc.

[0049] Step 140: Execute each task based on the target resource of each task to obtain the processing result of each task.

[0050] Specifically, in step 130, the target resources required for each task have been identified, and then the corresponding target resources have been allocated to the corresponding tasks. Based on the allocated target resources, each task is executed to obtain the processing result of each task.

[0051] Exemplarily, target resources may be allocated to each task in the order in which they are executed, and after each stage of task execution is completed, the corresponding target resources may be released so that the target resources can be used as idle resources for adaptation of the tasks of the next stage.

[0052] For example, Task 1 and Task 2 are executed in the first stage, Task 1 corresponds to target resource a, and Task 2 corresponds to target resource b. After Task 1 and Task 2 are completed, target resource a and target resource b are released, so that target resource a and target resource b can be used as idle resources for task adaptation in the second stage, thereby maximizing resource utilization.

[0053] The task processing method provided by the embodiment of the present invention determines multiple tasks and the input and output requirements of each task through the semantic information of the task description text to ensure that each task is correctly understood. At the same time, according to the input and output requirements of each task, a suitable resource is dynamically selected from the idle resources as the target resource to ensure that the resources are fully utilized and avoid waste of resources. At the same time, the embodiment of the present invention realizes full-process automated processing of tasks, reduces dependence on manual labor, reduces labor costs, and improves processing speed.

[0054] Compared with the audit system built on traditional artificial intelligence, the embodiment of the present invention does not need to iteratively optimize the audit system for different scenarios, and is more efficient. Compared with the traditional system built on rules and expert knowledge, when business processes or regulations and policies change, the embodiment of the present invention does not require experts to re-examine all relevant rules and manually update them, saving labor costs.

[0055] Based on the above embodiment, based on the semantic information of the task description text, a plurality of tasks to be processed and the input and output requirements of each task are determined, including: Determine the task scenario based on the semantic information of the task description text; Based on the task scenario, multiple key elements associated with the task scenario and the logical relationship between the key elements are extracted from the task description text; Based on the key elements and the logical relationships between the key elements, multiple tasks to be processed and the input and output requirements of each task are determined.

[0056] Specifically, considering that the task description text is usually concise and may only contain basic information and objectives of the task, the task scenario provides the specific environment and background of the task, which helps to understand the task more comprehensively. By combining the task scenario and the task description text, multiple key elements closely related to the task can be extracted.

[0057] Taking the medical review scenario as an example, if the task scenario is not determined, when extracting key elements, we may only focus on the semantic information of the task description text "reviewing medical expense reimbursement applications" itself, that is, the extracted key elements may all come from the task description text, and the deep semantic information of the task description text cannot be extracted. However, according to "reviewing medical expense reimbursement applications", the corresponding task scenario is the medical review scenario, which also involves multiple key elements such as the patient's identity information, medical expense details, diagnosis certificates, and the logical relationship between these key elements (such as verifying identity information first, then checking expense details and diagnosis certificates, and finally judging the rationality of the expenses).

[0058] It can be seen that the embodiment of the present invention determines the task scenario based on the semantic information of the task description text, so that based on the task scenario, it can mine the deep semantic information of the task description text, and then accurately and comprehensively extract multiple key elements associated with the task scenario and the logical relationship between the key elements.

[0059] After obtaining multiple key elements associated with the task scenario and the logical relationships between the key elements, the complex task to be executed can be decomposed into multiple smaller and more specific tasks based on the key elements and the logical relationships between the key elements, as well as the input and output requirements of each task.

[0060] For example, in a medical audit, key elements may include the patient's identity information, medical expense details, diagnosis certificate, etc. The logical relationship between key elements may be to verify the identity information first, and then check the medical expense details and diagnosis certificate. Based on the key elements and the logical relationship between key elements, the tasks to be performed can be decomposed into multiple tasks such as "verify identity information", "check medical expense details", and "check diagnosis certificate". For the task of "verifying identity information", the input requirement may be the patient's identity information data, and the output requirement may be the result of passing or failing the verification. For the task of "checking medical expense details", the input requirement may be the medical expense details data, and the output requirement may be the confirmed expense amount after verification or an abnormal expense prompt.

[0061] For example, the task description text can be input into the big model, and the big model can extract multiple key elements associated with the task scenario and the logical relationship between the key elements based on the contextual semantics of the task description text. Unlike traditional rule-based or template-based parsing methods, the use of big models can understand the deep meaning in the text, which is particularly suitable for scenarios that process massive data or multilingual text. In the fields of medicine, law, etc., the core content that needs further analysis can be extracted from reports, medical records or legal documents.

[0062] In other words, the embodiment of the present invention does not rely on traditional expert systems, but uses the self-learning ability of large models to dynamically analyze the semantic associations and logical chains in the text. Through in-depth analysis of the context, the core issues and priorities that need to be addressed can be quickly extracted. For example, in a medical scenario, the patient's main symptoms, disease development trends, and potential risk points can be identified, thereby providing accurate input for subsequent analysis steps.

[0063] Based on any of the above embodiments, based on the key elements and the logical relationships between the key elements, a plurality of tasks to be processed and input and output requirements of each task are determined, including: Based on the key elements and the logical relationships between the key elements, multiple tasks to be processed and the processing objectives of each task are determined; Based on the processing objectives of each task, determine the input and output requirements of each task.

[0064] Specifically, processing objectives are usually used to clarify the specific results or standards that the task needs to achieve. The core content of each task is composed of the corresponding key elements. Based on the key elements and the logical relationship between the key elements, the core content and requirements of the task can be identified to decompose the complex tasks to be performed into multiple smaller and more specific tasks and the processing objectives of each task.

[0065] In order to achieve the processing goal, the corresponding data, information or resources need to be input during the task execution, that is, the input requirements can be clearly defined based on the processing goal. After the task is completed, it is necessary to produce output results that meet the processing goal, that is, the output requirements can be clearly defined based on the processing goal. The input requirements are set to meet the needs of the processing goal, while the output requirements are the specific manifestations after the processing goal is achieved.

[0066] For example, in a medical review scenario, for the task "verify identity information", the processing goal is to ensure the authenticity and accuracy of the patient's identity information. According to the processing goal, the input requirement is the patient's identity information data, and the output requirement is the result of verification passed or failed.

[0067] Among them, the processing goals may include improving diagnostic accuracy, improving the efficiency of medical processes, reducing misdiagnosis and missed diagnoses, etc. Improving diagnostic accuracy can be reflected in the generation of more accurate diagnostic results by analyzing a large amount of historical data, medical literature, and the latest medical guidelines, which helps reduce doctors' misjudgment, especially in complex cases with multiple symptoms. Improving the efficiency of medical processes is reflected in the ability to quickly process and analyze large amounts of medical data and reduce doctors' repetitive work. For time-sensitive scenarios such as emergency departments, intelligent scenarios can quickly generate possible diagnostic conclusions, helping doctors prioritize critical cases and thus speed up the decision-making process. Reducing misdiagnosis and missed diagnoses is reflected in the ability to capture subtle symptoms that may be overlooked through a comprehensive analysis of the patient's symptoms and medical history, thereby reducing the risk of misdiagnosis. At the same time, it can provide different possible diagnostic results and provide doctors with probability-based disease recommendations.

[0068] Based on any of the above embodiments, each task is executed based on the target resource of each task to obtain the processing result of each task, including: Determine the execution order of each task based on the input and output requirements of each task; According to the execution order of each task, each task is executed in turn based on the target resources of each task to obtain the processing results of each task.

[0069] Specifically, considering that the output of a task may be the input of another task, that is, there is a dependency between the two tasks, this dependency determines that the tasks must be executed in a specific order. For example, a directed acyclic graph (DAG) can be constructed based on the input and output requirements of each task. Each node in the DAG represents a task, and the edge represents the dependency between the tasks. In other words, DAG can intuitively display the execution order of each task and the dependency between the tasks.

[0070] After determining the execution order of each task, each task is executed in turn based on the target resources of each task according to the execution order to obtain the processing result of each task.

[0071] For example, in the medical review scenario, it usually includes patient identity information verification tasks, medical expense details verification tasks, diagnosis certificate review tasks, etc. The inputs of the medical expense details verification tasks and the diagnosis certificate review tasks both include the output of the patient identity information verification task, and thus the medical expense details verification tasks and the diagnosis certificate review tasks depend on the output of the patient identity information verification task, that is, the medical expense details verification tasks and the diagnosis certificate review tasks need to be executed after the patient identity information verification task.

[0072] Based on any of the above embodiments, in the process of executing each task based on the target resource of each task, the method further includes: If there is a newly added idle resource whose performance is better than the target resource of any task, the newly added idle resource will be allocated to the corresponding task.

[0073] Exemplarily, in the process of executing each task, the resource status can be monitored in real time. If any task is completed, the corresponding target resource is released, and the released target resource is the newly added idle resource.

[0074] After monitoring the newly added idle resources, the performance of the newly added idle resources is compared with the performance of the target resources currently allocated to each task to determine whether the performance of the newly added idle resources exceeds the performance of the target resources of any task. The performance here can be determined based on parameters such as computing speed, response time, and computing power.

[0075] If so, the newly added idle resources will be allocated to the corresponding tasks, that is, the target resources originally allocated to the tasks will be replaced with the newly added idle resources with better performance, so as to realize dynamic resource allocation and ensure that the tasks can always run on the resources with the best performance, thereby improving the efficiency and quality of task execution.

[0076] Optionally, if there are multiple tasks matching the newly added idle resources, the newly added idle resources may be preferentially allocated to tasks with higher complexity.

[0077] Based on any of the above embodiments, the target resource of each task includes multiple modules; Based on the target resources of each task, each task is executed to obtain the processing results of each task, including: Optimize the parameters of the next module based on the accuracy of the processing results of the current module; Based on the optimized next module, the processing result of the current module is applied to obtain the processing result of the optimized next module.

[0078] Specifically, the target resources of each task include multiple modules, and the modules cooperate with each other to complete the task together. For example, the modules here can be agents. When processing a task, each agent will consider the processing results of its previous agent and optimize its own parameters based on these results to better complete the subsequent processing work.

[0079] In addition, the accuracy of the processing result of the current module refers to the consistency between the output result of the current module and the expected result or the actual result when the current module performs the corresponding task, which can be measured by evaluation indicators (such as accuracy, recall rate, F1 score, etc.). According to the accuracy of the processing result of the current module, the parameters of the next module are optimized.

[0080] For example, if the accuracy of the processing result of the current module is high, the parameters of the next module can be fine-tuned to avoid making drastic changes to prevent the introduction of new errors. If the accuracy of the processing result of the current module is low, the corresponding optimization algorithm can be determined, and the parameters of the next module can be optimized based on the optimization algorithm and the accuracy of the processing result of the current module.

[0081] Optionally, the behavior and feedback data of multiple modules can be integrated and optimized through the Cooperative Reinforcement Learning (CRL) mechanism to ensure that each link in the decision-making process is highly coordinated and consistent on key decision factors, that is, to achieve collaborative consistency.

[0082] In a complex and dynamic environment, multiple rounds of automated iterations not only improve the independent decision-making ability of each module, but also, through feedback interaction between modules, make the system composed of modules more adaptable, thus forming "group intelligence" at the system level. This group intelligence plays a decisive role in the phased objectives and tasks, ensuring that the system as a whole has higher flexibility and comprehensive response capabilities when dealing with complex problems, and achieving a more efficient decision-making process.

[0083] The embodiments of the present invention can efficiently process various tasks and maintain the stability and reliability of its performance in a constantly changing environment through the collaborative work among multiple modules, parameter optimization, continuous optimization of processing results, and adaptability and flexibility.

[0084] In addition, after optimizing the parameters of the next module, the next module continues to reason based on the output of the previous module. It can not only generate the best processing solution based on the current data, but also handle multiple constraints and hypothetical scenarios, and provide different alternatives for selection. Especially for complex multivariate decision-making scenarios, it can handle uncertainty and diverse data and generate customized solutions for each specific scenario.

[0085] Based on any of the above embodiments, the processing results of each task are obtained, and then the method further includes: Based on the difference between the processing results of each task and the corresponding actual results, the target resources of each task are optimized.

[0086] Specifically, after each task is completed, the processing result of each task is obtained, and the processing result of each task is compared and analyzed with the corresponding actual result to obtain an analysis result, and the target resource of each task is optimized based on the analysis result. Optimizing the target resource may include reallocating resources, adjusting resource parameters, etc.

[0087] The embodiment of the present invention optimizes the task target resources by comparing the difference between the task processing result and the actual result, and feeds back the audit result to the system in a closed-loop manner, thereby forming a continuously optimized decision-making process, which helps to improve resource utilization efficiency and improve the execution efficiency and effectiveness of subsequent tasks.

[0088] Based on any of the above embodiments, Figure 2 is a schematic diagram of the structure of the intelligent scheduling system provided by the present invention, and the intelligent scheduling system is used to implement the task processing method described in any of the above embodiments, such as Figure 2 As shown, the intelligent orchestration system includes a scene definition unit, an intelligent inspiration unit and an orchestration unit.

[0089] The scenario definition unit mainly uses the logical thinking ability of the large model to assist or inspire the generation of key characters. This unit combines natural language processing, logical reasoning, knowledge application and other technologies to determine multiple tasks to be processed, the input and output requirements of each task, and the processing goals of each task based on the semantic information of the task description text.

[0090] The intelligent heuristic unit is used to determine the dependencies between tasks based on the input and output requirements of each task, and to construct a DAG with each task as a node and the dependencies between tasks as edges. The DAG is used to intuitively display the execution order of each task and the dependencies between tasks. Each task is assigned to a different orchestration unit.

[0091] Figure 3 is a schematic diagram of the execution logic of the arrangement unit provided by the present invention, such as Figure 3 As shown, the orchestration unit is used to perform primary logic analysis and secondary logic analysis based on the semantic information of the task description text, determine the task intention pointed to by the task description text, and determine the resource requirements of each task based on the analysis results of the two, and allocate target resources to each task from idle resources according to the resource requirements of each task, so as to realize dynamic task allocation. In addition, if the performance of the newly added idle resources is greater than the performance of the target resources of any task, the newly added idle resources will be allocated to the corresponding tasks to realize resource optimization configuration. The resources here can be understood as the model used to execute tasks.

[0092] After allocating target resources to each task, decisions are generated and reasoned based on each target resource to obtain the processing results of each task. The processing results of each task are compared with the corresponding actual results to obtain analysis results, and the target resources of each task are optimized based on the analysis results.

[0093] In addition, in the process of executing each task based on the target resources of each task, the behavior and feedback data of multiple modules in the orchestration unit can be integrated and optimized through the collaborative reinforcement learning mechanism to ensure that each link in the decision-making process maintains a high degree of coordination and consistency on key decision factors, that is, to achieve collaborative consistency.

[0094] In addition, after optimizing the parameters of each module, the next module continues to reason based on the output of the previous module. It can not only generate the optimal processing solution based on the current data, but also achieve collaborative decision-making.

[0095] The task processing device provided by the present invention is described below. The task processing device described below and the task processing method described above can be referenced to each other.

[0096] Based on any of the above embodiments, Figure 4 is a schematic diagram of the structure of the task processing device provided by the present invention, such as Figure 4 As shown, the device comprises: A task determination unit 410 is used to determine a plurality of tasks to be processed and input and output requirements of each task based on semantic information of the task description text; A resource determination unit 420, configured to determine resource requirements of each task based on the input and output requirements of each task; A resource screening unit 430, configured to determine a target resource for each task from idle resources based on the resource requirements of each task; The task processing unit 440 is used to execute each task based on the target resources of each task and obtain the processing result of each task.

[0097] Based on any of the above embodiments, based on the semantic information of the task description text, a plurality of tasks to be processed and input and output requirements of each task are determined, including: Determine the task scenario based on the semantic information of the task description text; Based on the task scenario, multiple key elements associated with the task scenario and the logical relationship between the key elements are extracted from the task description text; Based on the key elements and the logical relationships between the key elements, multiple tasks to be processed and the input and output requirements of each task are determined.

[0098] Based on any of the above embodiments, based on the key elements and the logical relationships between the key elements, a plurality of tasks to be processed and input and output requirements of each task are determined, including: Based on the key elements and the logical relationships between the key elements, multiple tasks to be processed and the processing objectives of each task are determined; Based on the processing objectives of each task, determine the input and output requirements of each task.

[0099] Based on any of the above embodiments, each task is executed based on the target resource of each task to obtain the processing result of each task, including: Determine the execution order of each task based on the input and output requirements of each task; According to the execution order of each task, each task is executed in turn based on the target resources of each task to obtain the processing results of each task.

[0100] Based on any of the above embodiments, in the process of executing each task based on the target resource of each task, the process further includes: If there is a newly added idle resource whose performance is better than the target resource of any task, the newly added idle resource will be allocated to the corresponding task.

[0101] Based on any of the above embodiments, the target resource of each task includes multiple modules; Based on the target resources of each task, each task is executed to obtain the processing results of each task, including: Optimize the parameters of the next module based on the accuracy of the processing results of the current module; Based on the optimized next module, the processing result of the current module is applied to obtain the processing result of the optimized next module.

[0102] Based on any of the above embodiments, the processing results of each task are obtained, and then the method further includes: Based on the difference between the processing results of each task and the corresponding actual results, the target resources of each task are optimized.

[0103] Figure 5 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the task processing method, which includes: determining a plurality of tasks to be processed and the input and output requirements of each task based on the semantic information of the task description text; determining the resource requirements of each task based on the input and output requirements of each task; determining the target resources of each task from the idle resources based on the resource requirements of each task; and executing each task based on the target resources of each task to obtain the processing results of each task.

[0104] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the 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 a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0105] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the task processing method provided by the above-mentioned methods, which includes: based on the semantic information of the task description text, determining multiple tasks to be processed and the input and output requirements of each task; based on the input and output requirements of each task, determining the resource requirements of each task; based on the resource requirements of each task, determining the target resources of each task from the idle resources; based on the target resources of each task, executing each task to obtain the processing results of each task.

[0106] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the task processing method provided by the above-mentioned methods is implemented, and the method includes: determining multiple tasks to be processed and the input and output requirements of each task based on the semantic information of the task description text; determining the resource requirements of each task based on the input and output requirements of each task; determining the target resources of each task from the idle resources based on the resource requirements of each task; and executing each task based on the target resources of each task to obtain the processing results of each task.

[0107] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0108] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A task processing method, characterized in that: include: Based on the semantic information of the task description text, multiple tasks to be processed and the input and output requirements of each task are determined; Determine the resource requirements of each task based on the input and output requirements of each task; Based on the resource requirements of each task, determine the target resources for each task from the idle resources; Based on the target resources of each task, each task is executed to obtain the processing results of each task.

2. The task processing method according to claim 1, characterized in that: The step of determining a plurality of tasks to be processed and input and output requirements of each task based on the semantic information of the task description text includes: Determining a task scenario based on semantic information of the task description text; Based on the task scenario, extracting multiple key elements associated with the task scenario and the logical relationship between the key elements from the task description text; Based on the key elements and the logical relationships between the key elements, the plurality of tasks to be processed and the input and output requirements of each task are determined.

3. The task processing method according to claim 2, characterized in that: The determining of the plurality of tasks to be processed and the input and output requirements of each task based on the key elements and the logical relationships between the key elements includes: Based on the key elements and the logical relationships between the key elements, determining the plurality of tasks to be processed and a processing target of each task; Based on the processing objectives of each task, determine the input and output requirements of each task.

4. The task processing method according to any one of claims 1 to 3, characterized in that: The executing of each task based on the target resource of each task and obtaining the processing result of each task include: Determine the execution order of each task based on the input and output requirements of each task; According to the execution order of each task, each task is executed in turn based on the target resources of each task to obtain the processing results of each task.

5. The task processing method according to any one of claims 1 to 3, characterized in that: In the process of executing each task based on the target resource of each task, the method further includes: If there is a newly added idle resource whose performance is better than the performance of the target resource of any task, the newly added idle resource is allocated to the corresponding task.

6. The task processing method according to any one of claims 1 to 3, characterized in that: The target resources for each task include multiple modules; The executing of each task based on the target resource of each task and obtaining the processing result of each task include: Optimize the parameters of the next module based on the accuracy of the processing results of the current module; Based on the optimized next module, the processing result of the current module is applied to obtain the processing result of the optimized next module.

7. The task processing method according to any one of claims 1 to 3, characterized in that: The processing result of each task is obtained, and then the following is further included: Based on the difference between the processing results of each task and the corresponding actual results, the target resources of each task are optimized.

8. A task processing device, characterized in that: include: A task determination unit, used for determining a plurality of tasks to be processed and input and output requirements of each task based on semantic information of the task description text; A resource determination unit, used to determine the resource requirements of each task based on the input and output requirements of each task; A resource screening unit, used to determine the target resource of each task from the idle resources based on the resource requirements of each task; The task processing unit is used to execute each task based on the target resources of each task and obtain the processing results of each task.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the task processing method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the task processing method according to any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the task processing method according to any one of claims 1 to 7 is implemented.

Images