A task execution method and device, electronic equipment, storage medium and program product

Abstract

Embodiments of the present application relate to the technical field of artificial intelligence, and in particular to a task execution method and device, electronic equipment, storage medium and program product. The method is applied to an agent, and the method comprises: in response to a target task, determining a target reasoning template, wherein the target reasoning template is provided with a process layer and a description layer, the execution process of the target task is recorded in the process layer, and a plurality of descriptive knowledge candidates for implementing the execution process are recorded in the description layer; for an execution step in the execution process, scene information corresponding to the execution step is acquired, and target knowledge is determined from the plurality of descriptive knowledge according to the scene information; and the execution step is implemented by calling the target knowledge to execute the target task. The technical scheme of the embodiments of the present application can improve the generalization ability of the agent in the task execution process.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence technology, and in particular to a task execution method, apparatus, electronic device, storage medium and program product. Background Technology

[0002] In recent years, groundbreaking developments in Large Language Models (LLMs) have driven the evolution of AI Agent (hereinafter referred to as Agent) technology, enabling it to perform various tasks such as question answering, data analysis, and enterprise assistance based on the reasoning capabilities of LLMs.

[0003] In the process of realizing this invention, the inventors discovered the following technical problem in the prior art: the generalization ability of the Agent in the task execution process is insufficient, which urgently needs to be solved. Summary of the Invention

[0004] This invention provides a task execution method, apparatus, electronic device, storage medium, and program product, which solves the problem of insufficient generalization ability of the Agent during task execution.

[0005] According to one aspect of the present invention, a task execution method is provided, the method being applied to an intelligent agent, the method comprising:

[0006] In response to the target task, a target reasoning template is determined, which includes a process layer and a description layer. The process layer records the execution process of the target task, and the description layer records multiple descriptive knowledge that can be called for the implementation of the execution process.

[0007] For each execution step in the execution process, the corresponding scenario information is obtained, and the target knowledge is determined from multiple descriptive knowledge sources based on the scenario information. The execution steps are implemented by calling the target knowledge to execute the target task.

[0008] According to another aspect of the present invention, a task execution device is provided, which is configured on an intelligent agent and may include:

[0009] The reasoning template determination module is used to determine the target reasoning template in response to the target task. The target reasoning template has a process layer and a description layer. The process layer records the execution process of the target task, and the description layer records multiple descriptive knowledge that can be called for the implementation of the execution process.

[0010] The task execution module is used to obtain the scene information corresponding to the execution steps in the execution process, determine the target knowledge from multiple descriptive knowledge based on the scene information, and execute the execution steps by calling the target knowledge to execute the target task.

[0011] According to another aspect of the present invention, an electronic device is provided, which may include:

[0012] At least one processor; and

[0013] A memory that is communicatively connected to at least one processor; wherein,

[0014] The memory stores a computer program that can be executed by at least one processor, such that when the at least one processor executes the program, it implements the task execution method provided in any embodiment of the present invention.

[0015] According to another aspect of the present invention, a computer-readable storage medium is provided having computer instructions stored thereon for causing a processor to execute the task execution method provided in any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer program product is provided, on which a computer program is stored, which, when executed by a processor, implements the task execution method provided in any embodiment of the present invention.

[0017] The technical solution of this invention involves an agent determining a target inference template in response to a target task. This template includes a process layer and a description layer. The process layer records the execution process of the target task, while the description layer records multiple descriptive knowledge items that can be invoked to perform the execution process. For each execution step, the agent obtains the corresponding scenario information, then determines the target knowledge from the multiple descriptive knowledge items based on the scenario information, and invokes the target knowledge to perform the execution steps, thereby executing the target task. By introducing a target inference template with a process layer and a description layer, the agent can dynamically invoke appropriate descriptive knowledge (i.e., target knowledge) from the description layer based on the scenario information corresponding to the execution steps in the process to perform the execution steps and execute the target task. This allows the agent to adapt to diverse scenarios, thereby improving its generalization ability during task execution.

[0018] It should be understood that the description in this section is not intended to identify key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

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

[0020] Figure 1 This is a flowchart of a task execution method provided according to an embodiment of the present invention;

[0021] Figure 2 This is a flowchart of another task execution method provided by an embodiment of the present invention;

[0022] Figure 3 This is a flowchart of another task execution method provided by an embodiment of the present invention;

[0023] Figure 4 This is a schematic diagram illustrating an example of candidate inference template construction in another task execution method provided by an embodiment of the present invention;

[0024] Figure 5 This is a schematic diagram illustrating an example of automated iteration in another task execution method provided by an embodiment of the present invention;

[0025] Figure 6 This is a flowchart of another task execution method provided according to an embodiment of the present invention;

[0026] Figure 7 This is a schematic diagram illustrating a systematic invocation and automated iteration example of an inference template in another task execution method provided according to an embodiment of the present invention;

[0027] Figure 8 This is a structural block diagram of a task execution device according to an embodiment of the present invention;

[0028] Figure 9 This is a schematic diagram of the structure of an electronic device that implements the task execution method of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The same applies to "target," "original," etc., and will not be repeated here. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0031] It should be noted that the collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in the technical solution of this invention all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to maintain user personal information security and network security.

[0032] Before introducing the embodiments of the present invention, the application scenarios of the embodiments of the present invention will be described by way of example, so as to better understand the reason why the Agent has insufficient generalization ability during task execution, and thus better understand how the embodiments of the present invention solve this problem.

[0033] For example, to improve the accuracy of the agent's task execution, a corresponding execution process (i.e., execution flow) can be constructed for the task. This execution process includes multiple execution steps, which the agent can implement sequentially to execute the task. Furthermore, to further improve the agent's accuracy in executing the task, when implementing any execution step, the agent can invoke descriptive knowledge such as corresponding rule-based knowledge and / or conceptual knowledge to constrain the agent and prevent it from making arbitrary assumptions or creating illusions during implementation. It should be noted that in practical applications, the same execution step can be implemented for different scenarios. For example, the execution step of indicator analysis can be performed for different indicators; another example is the execution step of indicator analysis [GVM] (GVM is the abbreviation for Gross Merchandise Volume), which can be performed for the GVM during the Spring Festival shopping season or for the GVM during a group-buying festival. However, the descriptive knowledge invoked by the agent for this execution step has been solidified, specifically to descriptive knowledge adapted to a certain scenario. This results in the agent being unable to adapt to diverse scenarios and having insufficient generalization ability.

[0034] In this embodiment of the invention, when the Agent reaches any execution step, it can determine which scenario the execution step is for by obtaining the corresponding scenario information, and then execute it by calling descriptive knowledge adapted to that scenario, thus ensuring the Agent's generalization ability. This will be explained in detail below.

[0035] Figure 1 This is a flowchart of a task execution method provided by an embodiment of the present invention. This embodiment is applicable to situations where an agent performs tasks, and is particularly applicable to situations where an agent performs data analysis tasks. The method can be executed by the task execution device provided by this embodiment of the present invention. This device can be implemented in software and / or hardware, and can be configured on an agent. The agent can be integrated into an electronic device, which can be various user terminals or servers.

[0036] See Figure 1 The method of this invention specifically includes the following steps:

[0037] S110. In response to the target task, determine the target reasoning template;

[0038] The target reasoning template includes a process layer and a description layer. The process layer records the execution process of the target task, while the description layer records multiple descriptive knowledge items that can be called to implement the execution process.

[0039] The target task can be understood as the task that the Agent is instructed to perform, and more particularly as the task that the Agent is instructed to perform in what scenario. For example, it could be instructing the Agent to analyze the reasons for the GVM decline, or more specifically, instructing the Agent to analyze the reasons for the GVM decline during the Chinese New Year shopping festival.

[0040] In this embodiment of the invention, optionally, the target task can be represented by a user question. In this case, a user interface can be displayed, and the user question input based on the user interface is used as the target task. Here, receiving user questions through the user interface to trigger the target task allows the Agent to directly provide services to the needs of end users, improving the Agent's agility and accuracy in responding to user needs. Alternatively, the target task can also be triggered by time scheduling (i.e., timed triggering of the target task) or by internal events of the business system, etc., which depends on the actual situation and is not specifically limited here.

[0041] A target reasoning template can be understood as a pre-built structured framework for a target task, especially a task of this kind, to guide the agent in executing the target task and improve the accuracy of the target task execution. The target reasoning template is obtained through a layered design, which includes a process layer and a description layer.

[0042] Based on this, the process layer can be understood as the skeleton part of the target reasoning template. It contains the execution process consisting of one or more execution steps that needs to be followed to complete the target task. In addition, it can also contain the input and output of each execution step and / or the descriptive knowledge that needs to be triggered by each execution step. This is related to the actual situation and is not specifically limited here.

[0043] This description layer can be understood as the knowledge base portion of the target reasoning template, centrally storing multiple descriptive knowledge items that can be selected during the execution process, specifically during the execution of each step. This descriptive knowledge can be understood as knowledge used to describe the content of the execution steps to assist the Agent in performing those steps. In this embodiment of the invention, optionally, the description layer may include a rule layer and / or a concept layer. The descriptive knowledge recorded in the rule layer may include rule-based knowledge, which can be understood as knowledge that characterizes logical rules and can be used for reasoning. These logical rules may include logical rules applicable to different scenarios to define a reasoning mechanism, facilitating the Agent's reasoning based on this knowledge (i.e., implementing corresponding execution steps). The descriptive knowledge recorded in the concept layer may include conceptual knowledge that serves as factual basis during the reasoning process, such as entities, concepts, and terms, which are related to the actual situation and are not specifically limited here.

[0044] This layered design reduces maintenance difficulty and allows for separate maintenance by relevant professionals. For example, professionals familiar with the execution process can maintain the execution process without descriptive knowledge, while professionals familiar with descriptive knowledge can maintain the descriptive knowledge without the execution process. This also provides the Agent with more accurate and comprehensive constraints, thereby further improving the Agent's accuracy in task execution.

[0045] After detecting the target task, determine the target inference template that matches it. This can be determined based on the task type and / or task intent of the target task, and can be set according to requirements. No specific limitation is made here.

[0046] In this step, a target reasoning template with built-in process and description layers is defined for the target task. This separates the fixed execution process from the descriptive knowledge that can be called in the candidate, so that the Agent can flexibly call the descriptive knowledge to implement the execution process and complete the target task according to the scenario.

[0047] S120. For the execution steps in the execution process, obtain the scenario information corresponding to the execution steps, and determine the target knowledge from multiple descriptive knowledge based on the scenario information. Implement the execution steps by calling the target knowledge to execute the target task.

[0048] The execution steps can be understood as components of the execution process. In this embodiment of the invention, the processing method for each execution step is the same, so any execution step will be used as an example for explanation.

[0049] Scenario information can characterize the scenario in which the Agent performs the current execution step. Based on this, in the embodiments of the present invention, optionally, the scenario information can be business scenario information to characterize the business scenario (such as finance, healthcare, customer service, or supply chain) in which the Agent performs the execution step; it can be user scenario information to characterize the user scenario (such as enterprise users or individual users) in which the Agent performs the execution step; it can be technology scenario information to characterize the technology scenario (such as PC or mobile terminal) in which the Agent performs the execution step; it can be time scenario information to characterize the time scenario (such as weekdays or holidays) in which the Agent performs the execution step; it can also be geographic scenario information to characterize the geographic scenario (such as first-tier cities or second-tier cities) in which the Agent performs the execution step; it can also be implementation scenario information to characterize the implementation scenario (such as the implementation result of the previous execution step) in which the Agent performs the execution step; and so on. This is related to the actual situation and is not specifically limited here.

[0050] Target knowledge can be understood as descriptive knowledge that the Agent dynamically selects from multiple descriptive knowledge in the description layer based on scenario information, and that is adapted to the current specific scenario. The number of such descriptive knowledge can be one or more, and when there are multiple adapted descriptive knowledge, they can be assembled for application.

[0051] The process involves acquiring scenario information and determining target knowledge from multiple descriptive knowledge sets. This target knowledge can then be invoked from the description layer to execute the process steps and achieve the target task. For example, when executing the step of "calculating discounts," if the scenario information "the user belongs to the VIP level" is acquired, rule-based knowledge describing VIP user discount rules can be used as the target knowledge and invoked. Similarly, when executing the step of "defining a problem," if the scenario information "medical consultation" is acquired, a medical terminology dictionary can be selected from the concept layer as the target knowledge and invoked to accurately understand the user's terminology.

[0052] Based on this, optionally, descriptive information adapted to different scenario information can be directly labeled in the execution steps, or descriptive information adapted to different scenario information can be labeled in other modules. This can be set according to the actual situation, and no specific limitation is made here.

[0053] In this step, by dynamically acquiring scenario information and selecting appropriate descriptive knowledge accordingly, the limitations of traditional fixed descriptive knowledge are broken. This allows the agent to call upon the most suitable descriptive knowledge to guide the implementation of the current execution steps, significantly improving the agent's generalization ability to perform tasks in diverse scenarios.

[0054] Based on this, to more vividly understand the task execution process described in the embodiments of the present invention, an example of a reasoning template and an example of an Agent executing a corresponding task based on it is provided here. This example implements each execution step in the process layer in sequence, and implements it by calling adapted descriptive knowledge. For example, in step 1, the GMV metric is parsed by obtaining the GMV definition.

[0055]

[0056] The technical solution of this invention involves an agent determining a target inference template in response to a target task. This template includes a process layer and a description layer. The process layer records the execution process of the target task, while the description layer records multiple descriptive knowledge items that can be invoked to perform the execution process. For each execution step, the agent obtains the corresponding scenario information, then determines the target knowledge from the multiple descriptive knowledge items based on the scenario information, and invokes the target knowledge to perform the execution steps, thereby executing the target task. By introducing a target inference template with a process layer and a description layer, the agent can dynamically invoke appropriate descriptive knowledge (i.e., target knowledge) from the description layer based on the scenario information corresponding to the execution steps in the process to perform the execution steps and execute the target task. This allows the agent to adapt to diverse scenarios, thereby improving its generalization ability during task execution.

[0057] Figure 2 This is a flowchart of another task execution method provided by an embodiment of the present invention. This embodiment is based on and optimized from the above-described technical solutions. In this embodiment, optionally, determining the target inference template includes: performing intent recognition on the target task to obtain the target intent; in response to the existence of a candidate inference template that matches the target intent among a plurality of pre-constructed candidate inference templates, using the matched candidate inference template as the target inference template. The explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here.

[0058] See Figure 2 The method in this embodiment may specifically include the following steps:

[0059] S210. In response to the target task, perform intent recognition on the target task to obtain the target intent.

[0060] Among them, the intention recognition of the target task can be achieved, for example, by semantic abstraction and analysis of the target task, so as to obtain the target intention that can represent the core purpose of the target task.

[0061] In this step, by transforming the raw input into structured intent labels, a semantic basis is provided for the subsequent accurate matching inference template, avoiding the ambiguity and bias that may result from direct matching.

[0062] S220. In response to the existence of a candidate reasoning template that matches the target intent among the pre-constructed multiple candidate reasoning templates, the matched candidate reasoning template is used as the target reasoning template;

[0063] The target reasoning template includes a process layer and a description layer. The process layer records the execution process of the target task, while the description layer records multiple descriptive knowledge items that can be called to implement the execution process.

[0064] Among them, the candidate reasoning template can be understood as a reasoning template that is pre-constructed for various intentions (such as historical intentions as described later) and includes a process layer and a description layer.

[0065] If a candidate inference template matches the target intent among multiple candidate inference templates, that candidate inference template can be used as the target inference template to guide the execution of the target task. For example, multiple candidate inference templates such as replenishment strategy, promotion strategy, and clearance strategy are pre-built. After identifying the target intent as replenishment strategy formulation, the candidate inference template of the corresponding replenishment strategy can be used as the target inference template to guide the execution of the replenishment strategy formulation task.

[0066] S230. For the execution steps in the execution process, obtain the scenario information of the execution steps, determine the target knowledge from multiple descriptive knowledge based on the scenario information, and implement the execution steps by calling the target knowledge to execute the target task.

[0067] The technical solution of this invention realizes automated mapping from task input to structured execution framework through intent recognition and inference template matching mechanism. This ensures that the Agent can quickly obtain the inference template most suitable for the task purpose, avoiding the blindness and arbitrariness of inference template selection, thus laying a good foundation for subsequent dynamic invocation of descriptive knowledge based on scenario information.

[0068] Figure 3 This is a flowchart of another task execution method provided by an embodiment of the present invention. This embodiment is based on and optimized from the above-described technical solutions. In this embodiment, optionally, the multiple candidate reasoning templates are pre-constructed in the following manner: acquiring multiple historical tasks, and performing intent recognition on each of the multiple historical tasks to obtain historical intents, wherein each of the historical intents contains a historical intent that matches the target intent; clustering the multiple historical tasks according to each historical intent to obtain task domains; and constructing the candidate reasoning template corresponding to each task domain. The explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here.

[0069] See Figure 3 The method in this embodiment may specifically include the following steps:

[0070] S310. Acquire multiple historical tasks and perform intent recognition on each historical task to obtain the historical intent.

[0071] In this context, historical tasks can be understood as tasks that precede the target task, serving as the input foundation for constructing candidate reasoning templates. For each historical task, intent identification is performed to obtain historical intents that characterize the core purpose of that historical task. For example, multiple historical tasks that occurred within the past year are retrieved, and intent identification is performed on each of them to obtain multiple historical intents such as restocking reminders, handling slow-moving inventory, promotional stocking, retirement processing, medical insurance reimbursement, and housing provident fund withdrawal.

[0072] In this step, by identifying the intent of historical tasks, the raw historical data is transformed into structured intent labels, providing standardized input for subsequent semantic-based clustering analysis, thereby ensuring that historical experience can be effectively mined and reused.

[0073] S320. Based on each historical intent, cluster multiple historical tasks to obtain the task domain.

[0074] This involves clustering multiple historical tasks based on their respective historical intentions, thereby categorizing historical tasks with similar semantics to obtain corresponding task domains. For example, based on the semantic similarity of historical intentions such as replenishment reminders, handling slow-moving inventory, and promotional stocking, the corresponding historical tasks are clustered into the inventory strategy task domain; and further, based on the semantic similarity of historical intentions such as retirement processing, medical insurance reimbursement, and housing provident fund withdrawal, the corresponding historical tasks are clustered into the social security business task domain.

[0075] In this step, a clustering mechanism based on historical intent is used to summarize scattered historical tasks into structured task domains, thereby clearly defining the capability boundaries of the Agent and providing a classification basis for the subsequent construction of candidate inference templates by task domain, thus maximizing the alignment between business expectations and the Agent's deliverable capabilities.

[0076] S330. For each task domain, construct a candidate inference template corresponding to the task domain.

[0077] For each task domain, a corresponding candidate inference template can be constructed based on the historical tasks within that domain. For example, the common execution logic of these historical tasks can be analyzed to construct the corresponding candidate inference template, particularly the process layer within the candidate inference template. Of course, candidate inference templates can also be constructed using other methods, which will be detailed below.

[0078] In this step, by constructing corresponding candidate reasoning templates for each task domain, historical experience is transformed into a reusable structured execution framework, realizing the knowledge transformation from historical tasks to standardized reasoning templates, thereby providing rich candidate resources for subsequent intent matching.

[0079] S340. In response to the target task, perform intent recognition on the target task to obtain the target intent, wherein there are historical intents that match the target intent among the historical intents.

[0080] Based on this, optionally, a large language model can be used to identify the intent of multiple historical tasks separately to obtain the historical intent. Then, the same large language model can be used to identify the intent of the target task to obtain the target intent. The advantage of this setup is that a consistent intent identification mechanism is established by using a unified large language model throughout the entire process of historical intent identification and target intent identification. This ensures the reliability and accuracy of inference template matching and lays a solid foundation for the agent to accurately select the appropriate inference template.

[0081] S350. In response to the existence of a candidate reasoning template that matches the target intent among multiple candidate reasoning templates, the matched candidate reasoning template is used as the target reasoning template;

[0082] The target reasoning template includes a process layer and a description layer. The process layer records the execution process of the target task, while the description layer records multiple descriptive knowledge items that can be called to implement the execution process.

[0083] S360. For the execution steps in the execution process, obtain the scenario information of the execution steps, determine the target knowledge from multiple descriptive knowledge based on the scenario information, and implement the execution steps by calling the target knowledge to execute the target task.

[0084] The technical solution of this invention constructs an automated generation mechanism from historical experience to candidate reasoning templates by identifying historical task intent, performing cluster analysis, and establishing candidate reasoning templates according to task domains. This enables the candidate reasoning templates to truly reflect the inherent laws and business boundaries of historical tasks, avoiding the subjectivity and one-sidedness that may be caused by manual construction. It provides the Agent with multiple candidate reasoning templates that are comprehensive and reasonably classified, further improving the generalization ability of task execution.

[0085] An optional technical solution involves constructing candidate inference templates corresponding to the task domain of the target inference template, including:

[0086] Obtain task knowledge represented by the task domain, divide the task knowledge into procedural knowledge and descriptive knowledge, and perform logical abstraction on the procedural knowledge to obtain the execution process;

[0087] Obtain the initial reasoning template with a process layer and a description layer, record the execution process in the process layer of the initial reasoning template and record the descriptive knowledge in the description layer of the initial reasoning template to construct the target reasoning template.

[0088] In this context, task knowledge can be understood as all knowledge contained within the task domain. This knowledge is divided into procedural knowledge and the descriptive knowledge described above. The procedural knowledge represents the execution process of the task represented by the task domain. For example, for the inventory strategy task domain, a complete analytical case of replenishment decision-making within it is obtained as procedural knowledge, while inventory calculation formulas and safety stock definitions are obtained as descriptive knowledge. Similarly, for the social security business task domain, a complete analytical case of retirement processing within it can be obtained as procedural knowledge, while social security policies, regulations, and terminology definitions are obtained as descriptive knowledge. This hierarchical decomposition of task knowledge, by dividing the task knowledge represented by the task domain into procedural and descriptive categories, provides a clear classification of materials for the subsequent construction of the procedural and descriptive layers, ensuring that the two levels of the corresponding candidate reasoning templates each have their own sources and are not confused with each other.

[0089] Furthermore, procedural knowledge is logically abstracted to extract a general and reusable sequence of execution steps, upon which the execution process is constructed. Taking the inventory strategy task domain as an example, multiple complete analysis cases of replenishment decisions are logically abstracted to extract the execution process of inventory counting → demand forecasting → replenishment calculation → order generation → supplier confirmation. Through logical abstraction, personalized procedural knowledge is transformed into a general execution process, forming the skeleton of the agent's task execution. This ensures that the agent does not deviate from the core process when executing similar tasks, improving the standardization and stability of task execution.

[0090] Based on this, an initial reasoning template is obtained, which has a process layer and a description layer, but these two layers do not contain any content. At this point, the execution process can be recorded in the process layer and the descriptive knowledge can be recorded in the description layer to construct the corresponding candidate reasoning template.

[0091] The above technical solution achieves refined construction from task domain to candidate reasoning templates through task knowledge layering, procedural knowledge logical abstraction, and candidate reasoning template integration.

[0092] Based on this, in order to better understand the construction process of the candidate inference template as described in the embodiments of the present invention, specific examples are provided below. For example, see... Figure 4The process involves acquiring a set of business problems (i.e., multiple historical tasks), and then using LLM (Local Management Model) to cluster them based on the semantics and analytical approach of the business problems. This results in multiple problem domains (i.e., task domains), defining the capability boundaries of the types of business problems that the Agent can solve. This approach maximizes the alignment between business expectations and the Agent's deliverable capabilities. Optionally, to improve clustering accuracy, when setting prompts for the LLM, the following key elements should be considered: accurate goal setting, explanation of the business background, and definition of classification criteria. Further, business problems can be input offline by business personnel during the cold start phase, and continuously accumulated based on user online interactions after online operation, thereby achieving automated iteration of problem domains.

[0093] For each problem domain, a corresponding candidate reasoning template is constructed. Taking any problem domain as an example, the business knowledge (i.e., task knowledge) represented by this problem domain is obtained. This business knowledge is divided into procedural knowledge and descriptive knowledge, including business rules and a business dictionary. Based on this, the procedural knowledge is logically abstracted to obtain the execution process, which is then populated into the process layer; rule-based knowledge from the business rules is referenced and populated into the rule layer; and conceptual knowledge from the business dictionary is referenced and populated into the concept layer. At this point, the corresponding candidate reasoning template is constructed.

[0094] The above example demonstrates how the candidate reasoning templates are layered and business knowledge is layered, and how business knowledge and candidate reasoning templates are hierarchically mapped, thereby achieving accurate construction of candidate reasoning templates.

[0095] Before introducing the following technical solutions, we will first illustrate their application scenarios. Most current systems construct the cold start logic of agents using offline knowledge and offline LLM. However, offline knowledge is disconnected from online applications, lacking a closed-loop feedback mechanism. This makes it difficult to use human experience to correct errors and supplement knowledge in a timely manner, and it cannot support the continuous upgrading of agent capabilities, thus affecting the accuracy and generalization of agents.

[0096] An optional technical solution, the above task execution method, further includes:

[0097] In response to the absence of a matching candidate inference template among multiple candidate inference templates, the target task is recorded, and a new task domain is clustered based on the recorded target task.

[0098] If the target inference template cannot be matched from multiple candidate inference templates, the target task is recorded as input data for subsequent iterative optimization. By recording the target task when no matching inference template is found, an automatic capture mechanism for capability gaps is established during online operation, thus providing a real and timely data source for the agent's continuous optimization.

[0099] Furthermore, the recorded target tasks are clustered to obtain new task domains not covered by existing inference templates (i.e., existing candidate inference templates), thus defining new boundaries of the Agent's capabilities. This achieves automated discovery and definition of new task domains, forming a closed-loop feedback mechanism from online execution to offline iteration, enabling the Agent's capability boundaries to continuously expand as task requirements change.

[0100] For example, see Figure 5 For the problem domain, during the cold start phase, multiple problem domains can be constructed by semantic abstraction based on the set of business problems input offline by business personnel. After online operation, user problems that cannot be covered by the existing inference template can be found through intent recognition, and new problem domains can be constructed based on these user problems, thereby realizing the automated iteration of the problem domain.

[0101] The above technical solution constructs an automated iteration mechanism for task domains by recording unmatched target tasks and clustering new task domains. This solves the problems of the disconnect between offline knowledge and online applications and the lack of closed-loop feedback, enabling the agent to capture new requirements that existing inference templates cannot cover in a timely manner, and continuously expand its capability boundaries based on real online data to improve the agent's accuracy and generalization in task execution.

[0102] Figure 6 This is a flowchart of another task execution method provided by an embodiment of the present invention. This embodiment is based on and optimized from the above-described technical solutions. In this embodiment, optionally, after the execution step is implemented by calling the target knowledge, the method further includes: outputting the implementation status of the execution step; and recording the correction information in response to the correction information fed back for the implementation status, so as to correct the execution step based on the recorded correction information. The explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here.

[0103] See Figure 6 The method in this embodiment may specifically include the following steps:

[0104] S410. In response to the target task, determine the target reasoning template;

[0105] The target reasoning template includes a process layer and a description layer. The process layer records the execution process of the target task, while the description layer records multiple descriptive knowledge items that can be called to implement the execution process.

[0106] S420. For the execution steps in the execution process, obtain the scenario information of the execution steps, determine the target knowledge from multiple descriptive knowledge based on the scenario information, implement the execution steps by calling the target knowledge to execute the target task, and output the implementation status of the execution steps.

[0107] After a certain execution step is completed, the implementation status of the execution step can be output. The implementation status can characterize the agent's implementation process and / or implementation result for the execution step. In the embodiments of the present invention, the implementation status may optionally include at least one of the target knowledge invoked, intermediate results generated during the implementation process, and final results generated after the implementation is completed, etc., without specific limitations.

[0108] In this step, by outputting the implementation status, the implementation process and / or results of the Agent are presented transparently, providing relevant personnel with an implementation record that can be reviewed, understood and evaluated, thus laying the foundation for subsequent manual intervention and knowledge correction.

[0109] S430. In response to corrective information provided in response to implementation status feedback, the corrective information is recorded, and the execution steps are corrected based on the recorded corrective information.

[0110] The correction information can be understood as information provided by relevant personnel regarding the implementation status for correcting the execution steps. In the embodiments of the present invention, this information may optionally be corrective opinions, optimization suggestions, or correct execution examples, which are related to the actual situation and are not specifically limited here.

[0111] After receiving the correction information, it can be recorded, and then the execution step can be corrected based on the multiple correction information recorded for the same execution step. For example, the execution logic represented by the execution step can be corrected and / or the adaptation relationship directly marked in the execution step can be corrected. The adaptation relationship is the adaptation relationship between each scenario information and each descriptive information, so as to transform human experience into the driving force for continuous optimization of agent capabilities.

[0112] For example, see Figure 5 For execution steps (or execution process), during the cold start phase, the execution process can be constructed based on the procedural knowledge input offline by business personnel. After running online, it can receive correction information from user feedback and, based on this, sort out good analysis cases to update the execution process, thereby realizing the automated iteration of the execution process (or reasoning template).

[0113] The technical solution of this invention constructs an automated iteration mechanism for reasoning templates by recording and updating the execution steps with implementation status output and correction information. This solves the problems of the disconnect between offline knowledge and online applications and the lack of closed-loop feedback, and enables human experience to continuously guide the upgrade of agent capabilities, thereby improving the agent's sustainable iteration efficiency and long-term generalization ability.

[0114] Based on this, in order to better understand the above technical solutions as a whole, the following will combine... Figure 7 The systematic invocation and automated iteration example of the inference template shown is illustrated by example.

[0115] For example, as explained above, in an offline process, LLM can be used to semantically abstract the set of business questions to achieve intent recognition and obtain the question domain. Then, by logically abstracting procedural knowledge, the execution process can be obtained and descriptive knowledge can be referenced to construct a reasoning template (i.e., a candidate reasoning template).

[0116] In the online process, users input their questions through the user interface. The agent then uses LLM (Limited Language Management) to identify the intent behind the questions, matching candidate inference templates to obtain the target inference template. Based on the target inference template, the agent collects and provides feedback, and then executes the target task triggered by the user's question. Furthermore, during the execution of the target task (i.e., the implementation of execution steps), the agent can output the implementation status of the execution steps. If the user does not approve of the implementation status, they can provide feedback for correction.

[0117] Furthermore, in the closed-loop feedback stage, multiple experts can be organized to review the correction information and user issues that failed to match, thereby obtaining good and bad analysis cases. Based on this, new problem domains and new analysis processes can be constructed to achieve automated iteration.

[0118] The above example constructs the inference template by layering the problem domain, inference template, and business knowledge, and executes the target task through the inference template. By combining offline and online iteration, the problem domain and inference template are automatically iterated, thereby improving the accuracy and generalization of the agent in the task execution process and providing efficient automated iteration capabilities.

[0119] Figure 8 This is a structural block diagram of a task execution apparatus provided in an embodiment of the present invention. This apparatus is used to execute the task execution method provided in any of the above embodiments. This apparatus and the task execution methods of the above embodiments belong to the same inventive concept. Details not described in detail in the embodiments of the task execution apparatus can be found in the embodiments of the above task execution methods. See also... Figure 8The device is configured on an intelligent agent and may specifically include: a reasoning template determination module 510 and a task execution module 520.

[0120] Among them, the reasoning template determination module 510 is used to determine the target reasoning template in response to the target task. The target reasoning template is provided with a process layer and a description layer. The process layer records the execution process of the target task, and the description layer records multiple descriptive knowledge that can be called for implementing the execution process.

[0121] The task execution module 520 is used to obtain the scenario information corresponding to the execution steps in the execution process, determine the target knowledge from multiple descriptive knowledge based on the scenario information, and execute the execution steps by calling the target knowledge to execute the target task.

[0122] Optionally, the inference template determination module 510 may include:

[0123] The target intent acquisition unit is used to identify the intent of the target task and obtain the target intent.

[0124] The reasoning template matching unit is used to select the matching candidate reasoning template as the target reasoning template in response to the existence of a candidate reasoning template that matches the target intent among a plurality of pre-built candidate reasoning templates.

[0125] Based on this, optional multiple candidate inference templates are pre-built through the following modules:

[0126] The historical intent acquisition module is used to acquire multiple historical tasks and perform intent recognition on each of the multiple historical tasks to obtain historical intents, among which there are historical intents that match the target intent;

[0127] The task domain acquisition module is used to cluster multiple historical tasks based on their respective historical intentions to obtain the task domain.

[0128] The inference template building module is used to build candidate inference templates corresponding to each task domain.

[0129] Based on this, optionally, for the task domain corresponding to the target inference template, the inference template construction module may include:

[0130] The execution process obtains units, which are used to acquire task knowledge represented by the task domain, divide the task knowledge into procedural knowledge and descriptive knowledge, and perform logical abstraction on the procedural knowledge to obtain the execution process;

[0131] The reasoning template construction unit is used to obtain an initial reasoning template with a process layer and a description layer, and to record the execution process in the process layer of the initial reasoning template and the descriptive knowledge in the description layer of the initial reasoning template to construct the target reasoning template.

[0132] Alternatively, the aforementioned task execution device may further include:

[0133] The task domain construction module is used to record the target task in response to the absence of a matching candidate inference template among multiple candidate inference templates, and to cluster new task domains based on the recorded target tasks.

[0134] Another optional historical intent retrieval module may include:

[0135] The historical intent acquisition unit is used to perform intent recognition on multiple historical tasks using a large language model to obtain the historical intent.

[0136] Accordingly, the target intent obtaining unit may include:

[0137] The target intent sub-unit is used to perform intent recognition on the target task using a large language model to obtain the target intent.

[0138] Optionally, the description layer includes a rules layer and / or a concept layer; wherein,

[0139] The descriptive knowledge recorded in the rule layer includes rule-based knowledge used for reasoning, and / or, the descriptive knowledge recorded in the concept layer includes concept-based knowledge that serves as factual basis in the reasoning process.

[0140] Optionally, the above-mentioned task execution device may further include:

[0141] The implementation status output module is used to output the implementation status of the execution steps after the execution steps are implemented by calling the target knowledge;

[0142] The execution step correction module is used to respond to correction information fed back regarding the implementation status, record the correction information, and correct the execution steps based on the recorded correction information.

[0143] Optionally, the target task is represented by a user question, and the inference template determination module 510 may include:

[0144] User interface display unit, used to display the user interface;

[0145] The reasoning template determination unit is used to determine a target reasoning template in response to a user question based on user interface input.

[0146] The task execution device provided in this embodiment of the invention allows the Agent to determine a target inference template in response to a target task via an inference template determination module. This target inference template includes a process layer and a description layer. The process layer records the execution process of the target task, and the description layer records multiple descriptive knowledge items that can be invoked to perform the execution process. The task execution module obtains scenario information corresponding to each execution step in the execution process, determines target knowledge from the multiple descriptive knowledge items based on the scenario information, and executes the target task by invoking the target knowledge. By introducing a target inference template with a process layer and a description layer, the Agent can dynamically invoke appropriate descriptive knowledge (i.e., target knowledge) from the description layer based on the scenario information corresponding to each execution step in the process to execute the target task. This allows the Agent to adapt to diverse scenarios, thereby improving its generalization ability during task execution.

[0147] The task execution device provided in the embodiments of the present invention can execute the task execution method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0148] It is worth noting that in the above embodiments of the task execution device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0149] Figure 9 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0150] like Figure 9As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0151] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0152] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as task execution methods.

[0153] In some embodiments, the task execution method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the task execution method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to execute the task execution method by any other suitable means (e.g., by means of firmware).

[0154] The various implementations of the systems and technologies described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chips or system-on-a-chips (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0155] Computer programs used to implement the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0156] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0157] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0158] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with the implementation of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0159] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0160] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication unit 19, or installed from storage unit 18, or installed from ROM 12. When the computer program is executed by processor 11, it performs the functions defined in the methods of the embodiments of the present invention.

[0161] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0162] The specific implementation described above does not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A task execution method, characterized in that, Applied to intelligent agents, the method includes: In response to a target task, a target reasoning template is determined, wherein the target reasoning template includes a process layer and a description layer. The process layer records the execution process of the target task, and the description layer records multiple descriptive knowledge items that can be invoked to implement the execution process. For each execution step in the execution process, scenario information corresponding to the execution step is obtained, and target knowledge is determined from multiple descriptive knowledge sources based on the scenario information. The execution step is then implemented by calling the target knowledge to execute the target task.

2. The method according to claim 1, characterized in that, The determination of the target reasoning template includes: The target task is subjected to intent recognition to obtain the target intent; In response to the existence of a candidate reasoning template that matches the target intent among a plurality of pre-constructed candidate reasoning templates, the matched candidate reasoning template is used as the target reasoning template.

3. The method according to claim 2, characterized in that, The candidate inference templates are pre-constructed in the following manner: Multiple historical tasks are acquired, and intent recognition is performed on each of the multiple historical tasks to obtain historical intents, wherein there is a historical intent that matches the target intent among the historical intents; Based on the stated historical intent, the multiple stated historical tasks are clustered to obtain the task domain; For each task domain, a candidate inference template corresponding to the task domain is constructed.

4. The method according to claim 3, characterized in that, For the task domain corresponding to the target inference template, constructing the candidate inference template corresponding to the task domain includes: Obtain the task knowledge represented by the task domain, divide the task knowledge into procedural knowledge and descriptive knowledge, and perform logical abstraction on the procedural knowledge to obtain the execution process; An initial reasoning template with the process layer and the description layer is obtained, and the execution process is recorded in the process layer of the initial reasoning template and the descriptive knowledge is recorded in the description layer of the initial reasoning template to construct the target reasoning template.

5. The method according to claim 3, characterized in that, Also includes: In response to the absence of a matching candidate inference template among the multiple candidate inference templates, the target task is recorded, and a new task domain is clustered based on the recorded target task.

6. The method according to claim 3, characterized in that, The step of performing intent recognition on multiple historical tasks to obtain historical intent includes: Using a large language model, intent recognition is performed on multiple historical tasks to obtain historical intent; The step of performing intent recognition on the target task to obtain the target intent includes: Using the large language model, the target task is identified to obtain the target intent.

7. The method according to claim 1, characterized in that, The description layer includes a rule layer and / or a concept layer; wherein... The descriptive knowledge recorded in the rule layer includes rule-based knowledge used for reasoning, and / or the descriptive knowledge recorded in the concept layer includes concept-based knowledge that serves as factual basis during the reasoning process.

8. The method according to claim 1, characterized in that, After performing the execution step by invoking the target knowledge, the method further includes: Output the implementation status of the execution steps; In response to correction information provided in response to the implementation, the correction information is recorded, and the execution steps are corrected based on the recorded correction information.

9. The method according to claim 1, characterized in that, The target task is represented by a user question, and the step of determining the target reasoning template in response to the target task includes: Display the user interface; In response to the user question input based on the user interface, a target reasoning template is determined.

10. A task execution device, characterized in that, Configured in an intelligent agent, the device includes: The reasoning template determination module is used to determine a target reasoning template in response to a target task. The target reasoning template includes a process layer and a description layer. The process layer records the execution process of the target task, and the description layer records multiple descriptive knowledge items that can be invoked to implement the execution process. The task execution module is used to obtain the scene information corresponding to the execution step in the execution process, determine the target knowledge from multiple descriptive knowledge according to the scene information, and execute the execution step by calling the target knowledge to execute the target task.

11. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to cause the at least one processor to perform the task execution method as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the task execution method as described in any one of claims 1-9.

13. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the task execution method as described in any one of claims 1-9.

Images