Agent execution method and device based on dynamic optimization knowledge base, equipment and medium

Abstract

This application discloses a proxy execution method, apparatus, device, and medium based on a dynamically optimized knowledge base. The method includes: converting natural language task information into standardized task target information; generating task context information based on local system information; generating a success case retrieval request based on the standardized task target information and task context information, and sending the success case retrieval request to a cloud server; receiving target success cases returned by the cloud server, generating task execution plan information based on the target success cases, and executing corresponding operations according to the task execution plan information; generating a latest success case based on the corresponding operations, and sending the latest success case to the cloud server for the cloud server to add the latest success case to the dynamically optimized knowledge base. This technical solution can learn from historical successful experiences, quickly retrieve and adapt existing efficient execution solutions, and achieve continuous and autonomous improvement in task execution efficiency.

Description

Technical Field

[0001] This application belongs to the field of intelligent agent technology, specifically relating to agent execution methods, devices, equipment and media based on dynamically optimized knowledge bases. Background Technology

[0002] Proxy execution systems are a type of intelligent software system with automated task processing capabilities. They can autonomously complete computer-side operation tasks based on user natural language commands, such as file editing, software operation, and process execution. They are widely used in conversational assistants, office automation, and other scenarios. Their core objective is to replace manual labor in performing repetitive and standardized computer operations, thereby improving user work efficiency.

[0003] Currently, proxy execution systems mostly rely on the general reasoning capabilities of pre-trained large models or fixed instruction templates to complete task planning. This means that when faced with repetitive or similar tasks, the proxy execution system needs to restart the entire planning process, which not only causes redundant consumption of computing resources, but also makes it difficult to continuously improve task execution efficiency, and cannot meet the actual needs for efficient and accurate task processing in complex scenarios. Summary of the Invention

[0004] This application provides a proxy execution method, apparatus, device, and medium based on a dynamically optimized knowledge base. The purpose is to enable the proxy execution system to learn from historical successes and quickly retrieve and adapt existing efficient execution schemes when faced with repetitive or similar tasks, thereby significantly reducing the computational overhead of repetitive planning and achieving continuous and autonomous improvement in task execution efficiency.

[0005] In a first aspect, embodiments of this application provide a proxy execution method based on a dynamically optimized knowledge base, the method being executed by a proxy execution client; the method includes: The system receives natural language task information and converts it into standardized task target information; it also collects local system information and generates task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution, and a list of installed applications; A success case retrieval request is generated based on the standardized task objective information and the task context information, and the success case retrieval request is sent to the cloud server. The cloud server determines the target success case based on the success case retrieval request and the dynamically optimized knowledge base. The dynamically optimized knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases. Receive the target success case returned by the cloud server, generate task execution plan information based on the target success case, and execute the corresponding operation according to the task execution plan information; The latest successful case is generated based on the corresponding operation, and the latest successful case is sent to the cloud server, so that the cloud server can add the latest successful case to the dynamic optimization knowledge base.

[0006] Optionally, the method further includes: During the execution of the corresponding operation according to the task execution plan information, the operation screenshots, mouse click coordinates and step description texts corresponding to each execution step in the task execution plan information are obtained; Based on the operation screenshot, the mouse click coordinates, and the step description text, determine the area boundary and type label of the interactive component; Based on the region boundary and type label of the interactive component, generate the icon template image of the interactive component; The icon template image is sent to the cloud server, whereby the cloud server stores the icon template image in the dynamic optimization knowledge base.

[0007] Secondly, embodiments of this application provide a dynamically optimized knowledge base management method for agent execution, wherein the method is executed by a cloud server; the method includes: Upon receiving a success case retrieval request from the agent execution client, extract the standardized task target information and task context information from the success case retrieval request; Based on the standardized task objective information and the task context information, target success cases are retrieved from the dynamic optimization knowledge base, and the target success cases are returned to the agent execution client; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; Upon receiving the latest success case from the agent execution client, a target success case set is determined in the dynamic optimization knowledge base based on the latest success case, and the latest success case is added to the target success case set; Calculate the dynamic recall evaluation score for each successful case in the target successful case set, and optimize the target successful case set based on the dynamic recall evaluation score.

[0008] Optionally, the step of calculating the dynamic recall evaluation score of each successful case in the target successful case set, and optimizing the target successful case set based on the dynamic recall evaluation score, includes: Obtain the success rate, number of execution steps, number of recalls, and average search similarity of each successful case in the target successful case set; The average number of execution steps for the target successful case set is calculated based on the number of execution steps, and the relative recall frequency of each successful case is calculated based on the number of recalls. Based on the success rate, the number of execution steps, the average number of execution steps, the relative recall frequency, and the average retrieval similarity, calculate the dynamic recall evaluation score for each successful case in the target successful case set; The successful cases in the target successful case set are sorted in descending order according to the dynamic recall evaluation score to obtain the sorting result; If the number of successful cases in the sorting results exceeds a preset threshold, successful cases to be deleted are determined in the sorting results based on the number of successful cases and the preset threshold, and the successful cases to be deleted are deleted.

[0009] Optionally, retrieving successful cases of the target from the dynamic optimization knowledge base based on the standardized task target information and the task context information includes: Based on the standardized task objective information, calculate the semantic similarity and tag similarity of each successful case in the dynamic optimization knowledge base; Calculate the environmental similarity of each successful case based on the task context information; Based on the semantic similarity, the tag similarity, and the environmental similarity, the comprehensive similarity of each successful case is calculated; Based on the comprehensive similarity, a target successful case is selected from all successful cases.

[0010] Optionally, after calculating the comprehensive similarity of each successful case based on the semantic similarity, the tag similarity, and the environmental similarity, the method further includes: Based on the comprehensive similarity and dynamic recall evaluation score of each successful case, the target similarity of each successful case is calculated. Accordingly, the selection of a target successful case from among the successful cases based on the comprehensive similarity includes: Based on the target similarity, a target success case is selected from each success case.

[0011] Thirdly, embodiments of this application provide a proxy execution device based on a dynamically optimized knowledge base, the device being configured on a proxy execution client; the device includes: The information acquisition module is used to receive natural language task information and convert it into standardized task target information, collect local system information and generate task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution and a list of installed applications; The retrieval request module is used to generate a success case retrieval request based on the standardized task target information and the task context information, and send the success case retrieval request to the cloud server, so that the cloud server can determine the target success case based on the success case retrieval request and the dynamic optimization knowledge base; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation score of the success cases; The task execution module is used to receive the target success case returned by the cloud server, generate task execution plan information based on the target success case, and execute the corresponding operation according to the task execution plan information; The case generation module is used to generate the latest successful case based on the corresponding operation, and send the latest successful case to the cloud server, so that the cloud server can add the latest successful case to the dynamic optimization knowledge base.

[0012] Optionally, the device is also used for: During the execution of the corresponding operation according to the task execution plan information, the operation screenshots, mouse click coordinates and step description texts corresponding to each execution step in the task execution plan information are obtained; Based on the operation screenshot, the mouse click coordinates, and the step description text, determine the area boundary and type label of the interactive component; Based on the region boundary and type label of the interactive component, generate the icon template image of the interactive component; The icon template image is sent to the cloud server, whereby the cloud server stores the icon template image in the dynamic optimization knowledge base.

[0013] Fourthly, embodiments of this application provide a dynamically optimized knowledge base management device for agent execution, the device being configured on a cloud server; the device includes: The request parsing module is used to extract standardized task target information and task context information from the successful case retrieval request when it receives a successful case retrieval request from the proxy execution client. The case retrieval module is used to retrieve target success cases from the dynamic optimization knowledge base based on the standardized task target information and the task context information, and return the target success cases to the agent execution client; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; The case addition module is used to, upon receiving the latest successful case from the agent execution client, determine the target successful case set in the dynamic optimization knowledge base based on the latest successful case, and add the latest successful case to the target successful case set; The case optimization module is used to calculate the dynamic recall evaluation score of each successful case in the target successful case set, and optimize the target successful case set based on the dynamic recall evaluation score.

[0014] Optionally, the case optimization module is specifically used for: Obtain the success rate, number of execution steps, number of recalls, and average search similarity of each successful case in the target successful case set; The average number of execution steps for the target successful case set is calculated based on the number of execution steps, and the relative recall frequency of each successful case is calculated based on the number of recalls. Based on the success rate, the number of execution steps, the average number of execution steps, the relative recall frequency, and the average retrieval similarity, calculate the dynamic recall evaluation score for each successful case in the target successful case set; The successful cases in the target successful case set are sorted in descending order according to the dynamic recall evaluation score to obtain the sorting result; If the number of successful cases in the sorting results exceeds a preset threshold, successful cases to be deleted are determined in the sorting results based on the number of successful cases and the preset threshold, and the successful cases to be deleted are deleted.

[0015] Optionally, the case retrieval module is specifically used for: Based on the standardized task objective information, calculate the semantic similarity and tag similarity of each successful case in the dynamic optimization knowledge base; Calculate the environmental similarity of each successful case based on the task context information; Based on the semantic similarity, the tag similarity, and the environmental similarity, the comprehensive similarity of each successful case is calculated; Based on the comprehensive similarity, a target successful case is selected from all successful cases.

[0016] Optionally, the case retrieval module is further configured to: Based on the comprehensive similarity and dynamic recall evaluation score of each successful case, the target similarity of each successful case is calculated. Based on the target similarity, a target success case is selected from each success case.

[0017] Fifthly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the methods described in the first and second aspects.

[0018] In a sixth aspect, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the methods described in the first and second aspects.

[0019] In this embodiment, natural language task information is received and converted into standardized task target information; local system information is collected and task context information is generated based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution, and a list of installed applications; a success case retrieval request is generated based on the standardized task target information and the task context information, and the success case retrieval request is sent to a cloud server, so that the cloud server determines target success cases based on the success case retrieval request and a dynamically optimized knowledge base; wherein, the dynamically optimized knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; the target success cases returned by the cloud server are received, and task execution plan information is generated based on the target success cases, and corresponding operations are executed according to the task execution plan information; a latest success case is generated based on the corresponding operations, and the latest success case is sent to the cloud server, so that the cloud server adds the latest success case to the dynamically optimized knowledge base. The aforementioned proxy execution method based on a dynamically optimized knowledge base enables the proxy execution system to learn from past successful experiences. When faced with repetitive or similar tasks, it can quickly retrieve and adapt existing efficient execution solutions, thereby significantly reducing the computational overhead of repetitive planning and achieving continuous and autonomous improvement in task execution efficiency. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the proxy execution method based on a dynamically optimized knowledge base provided in Embodiment 1 of this application; Figure 2 This is a flowchart illustrating the dynamically optimized knowledge base management method for agent execution provided in Embodiment 2 of this application; Figure 3This is a schematic diagram of the agent execution device based on a dynamically optimized knowledge base provided in Embodiment 3 of this application; Figure 4 This is a schematic diagram of the structure of the dynamically optimized knowledge base management device for agent execution provided in Embodiment 4 of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in Embodiment 5 of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0022] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0023] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0024] The following description, in conjunction with the accompanying drawings, details the agent execution method, apparatus, device, and medium based on a dynamically optimized knowledge base provided in this application, through specific embodiments and application scenarios.

[0025] First, this application applies to scenarios where an agent execution system is used to automate operations.

[0026] A proxy execution system can be an intelligent, automated system capable of autonomously receiving and parsing task instructions, generating and executing optimal operation plans. Correspondingly, the proxy execution client can be a user-side terminal device used to receive task instructions and perform specific operations, such as a personal computer or laptop.

[0027] A cloud server can be a remote service cluster with massive data storage, high-performance computing, and intelligent retrieval capabilities, and it carries a dynamically optimized knowledge base.

[0028] Example 1 Figure 1 This is a flowchart illustrating the proxy execution method based on a dynamically optimized knowledge base provided in Embodiment 1 of this application. Figure 1 As shown, the specific steps include the following: S101, receive natural language task information and convert the natural language task information into standardized task target information, collect local system information and generate task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution and installed application list; Based on the above usage scenarios, it is understood that the execution entity in this embodiment can be a proxy execution client.

[0029] Natural language task information can be a description of the user's operation requirements input in unstructured natural language, such as "use Excel to count the file modification records of drive C in the past 3 months and generate a bar chart". Natural language task information can be received in several ways: manually entered by the user through a text input box; voice commands received by a speech recognition module and converted into text; or natural language task commands pushed by a third-party application via an API interface.

[0030] Standardized task objective information can be a set of core task elements presented in a structured manner according to a preset data format (such as JSON or XML), which may include fields such as task type, target object, operation actions, output requirements, and time constraints. The method for converting natural language task information into standardized task objective information can be to input the natural language task information into an LLM (Large Language Model) and obtain the standardized task objective information output by the LLM.

[0031] Local system information can be attribute information describing the agent execution client, including operating system type, system language type, display resolution, and a list of installed applications. Specifically, the operating system type can be the core system software type of the agent execution client, such as Windows 10 / 11, macOS Ventura, Linux Ubuntu 22.04, Android 14, iOS 17, etc.; the system language type can be the language currently enabled for interface display, command parsing, and data interaction by the agent execution client, such as Simplified Chinese, English, Japanese, German, etc.; the display resolution can be the pixel array size parameter of the display device associated with the agent execution client, such as 1920×1080, 2560×1440, 3840×2160, 1366×768, etc.; the list of installed applications can be the names and version numbers of all executable applications installed on the agent execution client. Local system information can be collected by calling the agent execution client's system API interface.

[0032] Task context information can be a structured dataset that integrates local system information and task-related environment information. One method for generating task context information based on local system information is to obtain task-related environment information from standardized task objective information, perform format standardization processing on both local system information and task-related environment information, and construct a hierarchical data structure according to a pre-defined classification logic to obtain the task context information.

[0033] S102, a success case retrieval request is generated based on the standardized task target information and the task context information, and the success case retrieval request is sent to the cloud server, so that the cloud server can determine the target success case based on the success case retrieval request and the dynamic optimization knowledge base; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation score of the success cases; Successful cases can be a collection of historically successfully executed task data, which may include case identifiers, standardized task objective information, task context information, execution step sequences, execution result feedback, and applicable scenario tags. A successful case retrieval request can be a structured request data initiated by the proxy execution client to the cloud server to retrieve matching successful cases.

[0034] The method for generating a success case retrieval request based on standardized task objective information and task context information can be achieved by encapsulating the standardized task objective information and task context information into a standardized request according to a communication data format supported by the cloud server (such as Protobuf or JSON), and adding a request timestamp and the client identifier of the proxy execution client. The success case retrieval request can be sent to the cloud server by establishing an encrypted communication link between the proxy execution client and the cloud server based on the HTTPS / TLS protocol, and then sending the success case retrieval request to the cloud server through this encrypted communication link.

[0035] A dynamically optimized knowledge base can be a knowledge storage system deployed on a cloud server with intelligent retrieval capabilities. The dynamic recall evaluation score of successful cases can be a quantitative indicator used to measure the recall value of successful cases.

[0036] The target success case can be a success case retrieved by the cloud server from a dynamically optimized knowledge base that matches the current success case retrieval request. The cloud server determines the target success case based on the success case retrieval request and the dynamically optimized knowledge base by extracting standardized task objective information and task context information from the success case retrieval request, and then retrieving the target success case from the dynamically optimized knowledge base based on this standardized task objective information and task context information.

[0037] S103, receive the target success case returned by the cloud server, generate task execution plan information based on the target success case, and perform corresponding operations according to the task execution plan information; To receive the target success case returned by the cloud server, one can use an agent to execute an encrypted communication link between the client and the cloud server to receive the target success case returned by the cloud server.

[0038] Task execution plan information can be a sequence of execution steps generated based on target success cases. One way to generate task execution plan information based on target success cases is to input the target success cases into an LLM (Large Language Model) and obtain the task execution plan information output by the LLM.

[0039] The corresponding operation can be a specific execution action directly related to each execution step in the execution step sequence. To execute the corresponding operation according to the task execution plan information, an automated execution script can be generated based on the task execution plan information. This script can then be executed by calling the built-in automated execution engine of the proxy execution client.

[0040] S104, Generate a latest success case based on the corresponding operation, and send the latest success case to the cloud server, so that the cloud server can add the latest success case to the dynamic optimization knowledge base.

[0041] The latest success case can be the success case corresponding to the current task. Generating the latest success case based on the corresponding operation can be achieved by assigning applicable scenario tags to the latest success case, recording the actual execution time, operation parameters, whether exceptions were triggered and their handling methods for each execution step in the task execution plan information, and obtaining the output file path, interface screenshot, and system status feedback information after executing the corresponding operation. The standardized task target information, task context information, and the above-mentioned collected information are then linked and bound together, and each field is automatically populated according to the data structure format supported by the dynamically optimized knowledge base to obtain the latest success case.

[0042] The latest success stories can be sent to the cloud server via an encrypted communication link between the client and the cloud server through a proxy.

[0043] The cloud server adds the latest success cases to the dynamic optimization knowledge base by determining the target success case set in the dynamic optimization knowledge base based on the latest success cases, and then adding the latest success cases to the target success case set.

[0044] Optionally, the method further includes: During the execution of the corresponding operation according to the task execution plan information, the operation screenshots, mouse click coordinates and step description texts corresponding to each execution step in the task execution plan information are obtained; Based on the operation screenshot, the mouse click coordinates, and the step description text, determine the area boundary and type label of the interactive component; Based on the region boundary and type label of the interactive component, generate the icon template image of the interactive component; The icon template image is sent to the cloud server, whereby the cloud server stores the icon template image in the dynamic optimization knowledge base.

[0045] Operation screenshots can be image data of the currently displayed interface captured at key moments of the operation corresponding to the execution step (such as mouse click, button trigger, or interface jump); mouse click coordinates can be the two-dimensional pixel coordinates of the mouse relative to the upper left corner of the current display interface when the operation corresponding to the execution step is a mouse click; step description text can be a structured textual description of the core operation logic and target object of the operation corresponding to the execution step, such as "click the plus button". The methods for obtaining operation screenshots, mouse click coordinates, and step description text corresponding to each execution step in the task execution plan information can be as follows: capturing the operation screenshot by calling the system screen capture API when the operation corresponding to the execution step is triggered; registering a mouse hook function to listen for mouse click events sent by the automated execution engine and obtain the mouse click coordinates; and extracting the description field from the execution step as the step description text.

[0046] Interactive components can be visual elements that support user / system-triggered interactive behaviors; area boundaries can be rectangular pixel ranges selected by framing interactive components in operation screenshots; type labels can be standardized classification identifiers based on the functionality and form of interactive components.

[0047] The method of determining the area boundary and type label of interactive components based on operation screenshots, mouse click coordinates, and step description text can be achieved by inputting the operation screenshots, mouse click coordinates, and step description text into VLM (Visual-Language Model) to obtain the area boundary and type label of the interactive components output by VLM.

[0048] Icon template images can be standardized image templates extracted from screenshots of operations, with background interference removed and core visual features of the interactive components retained. One method for generating icon template images for interactive components based on their region boundaries and type labels is to crop an image region containing only the interactive component from the screenshot according to its region boundaries. Then, threshold segmentation or semantic segmentation algorithms are used to separate the foreground and background of the component, and the background is set to transparent. Finally, the component image is processed to grayscale and contrast enhancement to obtain the icon template image.

[0049] One way to send icon template images to a cloud server is to use an agent to execute an encrypted communication link between the client and the cloud server.

[0050] The cloud server stores icon template images in a dynamic optimization knowledge base by binding the icon template images with corresponding successful cases in the dynamic optimization knowledge base.

[0051] The advantage of this approach is that it accumulates standardized interactive component icon templates, providing more accurate interface recognition support for the execution of similar tasks in the future.

[0052] In this embodiment, natural language task information is received and converted into standardized task target information; local system information is collected and task context information is generated based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution, and a list of installed applications; a success case retrieval request is generated based on the standardized task target information and the task context information, and the success case retrieval request is sent to a cloud server, so that the cloud server determines target success cases based on the success case retrieval request and a dynamically optimized knowledge base; wherein, the dynamically optimized knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; the target success cases returned by the cloud server are received, and task execution plan information is generated based on the target success cases, and corresponding operations are executed according to the task execution plan information; a latest success case is generated based on the corresponding operations, and the latest success case is sent to the cloud server, so that the cloud server adds the latest success case to the dynamically optimized knowledge base. The aforementioned proxy execution method based on a dynamically optimized knowledge base enables the proxy execution system to learn from past successful experiences. When faced with repetitive or similar tasks, it can quickly retrieve and adapt existing efficient execution solutions, thereby significantly reducing the computational overhead of repetitive planning and achieving continuous and autonomous improvement in task execution efficiency.

[0053] Example 2 Figure 2 This is a flowchart illustrating the dynamically optimized knowledge base management method for agent execution provided in Embodiment 2 of this application. Figure 2 As shown, the specific steps include the following: S201, Upon receiving a successful case retrieval request from the agent execution client, extract the standardized task target information and task context information from the successful case retrieval request; The method to extract standardized task target information and task context information from successful case retrieval requests can be to identify the data format of the successful case retrieval request, call the corresponding format parser to deconstruct the request data packet, and split out the standardized task target information and task context information.

[0054] S202, based on the standardized task target information and the task context information, retrieve the target success cases from the dynamic optimization knowledge base, and return the target success cases to the agent execution client; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; The dynamically optimized knowledge base stores standardized task objectives and task context information corresponding to each successful case.

[0055] Based on the standardized task objective information and task context information, the target success cases are retrieved from the dynamic optimization knowledge base. The similarity between the current standardized task objective information and task context information and the standardized task objective information and task context information in each success case in the dynamic optimization knowledge base can be used as the similarity of each success case. The success cases with the highest preset number of similarities are determined as the target success cases.

[0056] Optionally, retrieving successful cases of the target from the dynamic optimization knowledge base based on the standardized task target information and the task context information includes: Based on the standardized task objective information, calculate the semantic similarity and tag similarity of each successful case in the dynamic optimization knowledge base; Calculate the environmental similarity of each successful case based on the task context information; Based on the semantic similarity, the tag similarity, and the environmental similarity, the comprehensive similarity of each successful case is calculated; Based on the comprehensive similarity, a target successful case is selected from all successful cases.

[0057] Semantic similarity can be the degree of matching between the core semantic intent of the current standardized task objective information and the standardized task objective information of successful cases in the dynamically optimized knowledge base; tag similarity can be the degree of overlap and matching between the applicable scenario tag corresponding to the current standardized task objective information and the applicable scenario tag of the successful cases.

[0058] The method of calculating the semantic similarity and label similarity of each successful case in the dynamic optimization knowledge base based on standardized task target information can be achieved by converting the current standardized task target information and the standardized task target information of successful cases in the dynamic optimization knowledge base into high-dimensional semantic vectors, calculating the cosine similarity between the high-dimensional semantic vectors as semantic similarity, and using the Jaccard similarity algorithm to calculate the similarity between the applicable scenario label corresponding to the current standardized task target information and the applicable scenario label of successful cases in the dynamic optimization knowledge base as label similarity.

[0059] Environment similarity can be defined as the degree of matching between the execution environment conditions of the current task context information and the task context information of successful cases in the dynamic optimization knowledge base. One method for calculating the environment similarity of each successful case based on task context information is to convert the current task context information and the task context information of successful cases in the dynamic optimization knowledge base into vector format to obtain environment vectors, and then calculate the cosine similarity between these environment vectors as the environment similarity.

[0060] The overall similarity can be a quantitative indicator of the degree of matching across multiple dimensions, including semantic similarity, tag similarity, and environmental similarity (optionally including semantic similarity of historical task trajectories). The overall similarity of each successful case can be calculated by weighting and summing the semantic similarity, tag similarity, and environmental similarity according to preset weights.

[0061] Optionally, the comprehensive similarity of each successful case can be calculated based on semantic similarity, label similarity, and environmental similarity. Alternatively, the semantic similarity of historical task trajectories can be obtained, and the semantic similarity, label similarity, environmental similarity, and historical task trajectory semantic similarity can be weighted and summed according to preset weights to obtain the comprehensive similarity.

[0062] The method of selecting target success cases from among various success cases based on comprehensive similarity can be to select a preset number of success cases as target success cases in descending order of comprehensive similarity.

[0063] Optionally, after calculating the comprehensive similarity of each successful case based on the semantic similarity, the tag similarity, and the environmental similarity, the method further includes: Based on the comprehensive similarity and dynamic recall evaluation score of each successful case, the target similarity of each successful case is calculated. Accordingly, the selection of a target successful case from among the successful cases based on the comprehensive similarity includes: Based on the target similarity, a target success case is selected from each success case.

[0064] Target similarity can be a final quantitative matching indicator that integrates overall similarity and the recall value (dynamic recall evaluation score) of successful cases. The target similarity of each successful case can be calculated by multiplying the overall similarity by the dynamic recall evaluation score.

[0065] The method of selecting target success cases from among the success cases based on target similarity can be to select a preset number of success cases as target success cases in descending order of target similarity.

[0066] The advantage of this approach is that, compared to filtering based solely on overall similarity, target similarity is also incorporated into dynamic recall evaluation scores, amplifying the recall probability of high-quality cases and preventing suitable but inefficient cases from being prioritized for recall.

[0067] The advantage of this approach is that by calculating semantic similarity, tag similarity, and environmental similarity from multiple dimensions, it provides a more comprehensive search compared to a single-dimensional search, avoiding issues such as semantic matching but environmental incompatibility or environmental adaptation but deviation from task intent.

[0068] The method to return the target success case to the agent execution client can be achieved by using an encrypted communication link between the agent execution client and the cloud server.

[0069] S203, Upon receiving the latest success case from the agent execution client, determine the target success case set in the dynamic optimization knowledge base based on the latest success case, and add the latest success case to the target success case set; The target success case set can be a set of success cases in the dynamically optimized knowledge base that share similar scenarios with the latest success cases. One way to determine the target success case set based on the latest success cases in the dynamically optimized knowledge base is to select success cases with the same applicable scenario tags from the dynamic optimization knowledge base, based on the applicable scenario tags found in the latest success cases.

[0070] Adding the latest success stories to the target success story set can be done by adding the metadata of the latest success stories to the target success story set.

[0071] S204, calculate the dynamic recall evaluation score of each successful case in the target successful case set, and optimize the target successful case set based on the dynamic recall evaluation score.

[0072] One way to calculate the dynamic recall evaluation score for each successful case in the target successful case set is to obtain the recall count of each successful case in the target successful case set, and then calculate the relative recall frequency of each successful case as the dynamic recall evaluation score based on the recall count.

[0073] One way to optimize the target success case set based on dynamic recall evaluation scores is to sort the success cases in the target success case set in descending order according to the dynamic recall evaluation scores.

[0074] Optionally, the step of calculating the dynamic recall evaluation score of each successful case in the target successful case set, and optimizing the target successful case set based on the dynamic recall evaluation score, includes: Obtain the success rate, number of execution steps, number of recalls, and average search similarity of each successful case in the target successful case set; The average number of execution steps for the target successful case set is calculated based on the number of execution steps, and the relative recall frequency of each successful case is calculated based on the number of recalls. Based on the success rate, the number of execution steps, the average number of execution steps, the relative recall frequency, and the average retrieval similarity, calculate the dynamic recall evaluation score for each successful case in the target successful case set; The successful cases in the target successful case set are sorted in descending order according to the dynamic recall evaluation score to obtain the sorting result; If the number of successful cases in the sorting results exceeds a preset threshold, successful cases to be deleted are determined in the sorting results based on the number of successful cases and the preset threshold, and the successful cases to be deleted are deleted.

[0075] The success rate can be defined as the proportion of successful cases that are actually executed and achieve the expected goal after being recalled by the proxy execution client; the number of execution steps can be defined as the total number of execution steps in the task execution plan corresponding to the successful case; the number of recalls can be defined as the cumulative number of times a successful case has been identified as a target successful case in response to a successful case retrieval request from the proxy execution client since the successful case was added to the database; the average retrieval similarity can be defined as the average similarity between a successful case and the standardized task target information and task context information in the corresponding successful case retrieval request each time the successful case is recalled. The success rate, number of execution steps, number of recalls, and average retrieval similarity of each successful case in the target successful case set can be obtained through collaborative querying and statistical calculation by the log system and metadata management module of the dynamically optimized knowledge base.

[0076] The average number of execution steps can be the arithmetic mean of the number of execution steps for all successful cases in the target success case set. One way to calculate the average number of execution steps for the target success case set is to sum the number of execution steps for all successful cases in the target success case set and then divide the sum by the number of successful cases in the target success case set to obtain the average number of execution steps.

[0077] Relative recall frequency can be the ratio of the number of recalls for successful cases to the total number of recalls for all successful cases in the target successful case set. One way to calculate the relative recall frequency for each successful case based on the recall count is to calculate the total number of recalls for all successful cases in the target successful case set, and then divide the recall count of each successful case by the total number of recalls to obtain the relative recall frequency.

[0078] The dynamic recall evaluation score for each successful case in the target successful case set is calculated based on success rate, number of execution steps, average number of execution steps, relative recall frequency, and average retrieval similarity. This can be achieved by calculating the ratio of the number of execution steps to the average number of execution steps and subtracting this ratio from 1 to obtain the relative execution complexity. The success rate, relative execution complexity, relative recall frequency, and average retrieval similarity are then weighted and summed according to preset weights to obtain the dynamic recall evaluation score.

[0079] Optionally, after calculating the dynamic recall evaluation score of each successful case in the target successful case set based on success rate, number of execution steps, average number of execution steps, relative recall frequency, and average retrieval similarity, the last recall time interval of each successful case can also be obtained. The last recall time interval and the dynamic recall evaluation score are substituted into the dynamic decay function to obtain the final dynamic recall evaluation score of each successful case.

[0080] The successful cases in the target successful case set are sorted in descending order according to their dynamic recall evaluation scores. This can be done using sorting algorithms such as bubble sort, quick sort, merge sort, or heap sort, which sort the successful cases in the target successful case set in descending order of their dynamic recall evaluation scores.

[0081] The preset threshold can be considered as the maximum number of successful cases allowed to be retained in the target successful case set. If the number of successful cases in the sorted results exceeds the preset threshold, it indicates that there are too many successful cases in the target successful case set (same or similar scenarios). Retaining all of them would increase the storage pressure on the dynamic optimization knowledge base and the time required for subsequent retrieval. Therefore, it is necessary to optimize the size of the set by replacing inefficient cases. Successful cases to be deleted are all successful cases in the sorted results that are ranked after the preset threshold. Therefore, these successful cases can be deleted.

[0082] The advantages of this approach are that it comprehensively measures the practicality, simplicity, and adaptability of cases through four core indicators: success rate, number of execution steps, recall frequency, and retrieval similarity, making it more objective than traditional single-dimensional scoring. Based on a preset quantity threshold, the replacement strategy directly eliminates inefficient cases ranked lower, quickly reducing the size of the collection and balancing the richness of the knowledge base with retrieval efficiency.

[0083] In this embodiment, upon receiving a successful case retrieval request from the proxy execution client, standardized task target information and task context information are extracted from the successful case retrieval request. Based on the standardized task target information and the task context information, target successful cases are retrieved from the dynamic optimization knowledge base, and the target successful cases are returned to the proxy execution client. The dynamic optimization knowledge base is used to dynamically store successful cases and their dynamic recall evaluation scores. Upon receiving the latest successful case from the proxy execution client, a target successful case set is determined from the dynamic optimization knowledge base based on the latest successful case, and the latest successful case is added to the target successful case set. The dynamic recall evaluation score of each successful case in the target successful case set is calculated, and the target successful case set is optimized based on the dynamic recall evaluation score. The above-described dynamic optimization knowledge base management method for proxy execution, based on score calculation and optimization of similar case sets, avoids inefficient cases consuming resources, while strengthening the recall priority of high-quality cases and continuously improving the matching accuracy of the knowledge base.

[0084] Example 3 Figure 3 This is a schematic diagram of the agent execution device based on a dynamically optimized knowledge base provided in Embodiment 3 of this application. Figure 3 As shown, the device includes: The information acquisition module 310 is used to receive natural language task information and convert the natural language task information into standardized task target information, collect local system information and generate task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution and installed application list; The retrieval request module 320 is used to generate a success case retrieval request based on the standardized task target information and the task context information, and send the success case retrieval request to the cloud server, so that the cloud server can determine the target success case based on the success case retrieval request and the dynamic optimization knowledge base; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation score of the success cases; The task execution module 330 is used to receive the target success case returned by the cloud server, generate task execution plan information based on the target success case, and execute corresponding operations according to the task execution plan information; The case generation module 340 is used to generate the latest successful case based on the corresponding operation and send the latest successful case to the cloud server, so that the cloud server can add the latest successful case to the dynamic optimization knowledge base.

[0085] Optionally, the device is also used for: During the execution of the corresponding operation according to the task execution plan information, the operation screenshots, mouse click coordinates and step description texts corresponding to each execution step in the task execution plan information are obtained; Based on the operation screenshot, the mouse click coordinates, and the step description text, determine the area boundary and type label of the interactive component; Based on the region boundary and type label of the interactive component, generate the icon template image of the interactive component; The icon template image is sent to the cloud server, whereby the cloud server stores the icon template image in the dynamic optimization knowledge base.

[0086] In this embodiment, the information acquisition module is used to receive natural language task information and convert it into standardized task target information, collect local system information and generate task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution and installed application list; the retrieval request module is used to generate a success case retrieval request based on the standardized task target information and the task context information, and send the success case retrieval request to the cloud server, so that the cloud server determines target success cases based on the success case retrieval request and a dynamically optimized knowledge base; wherein, the dynamically optimized knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; the task execution module is used to receive the target success cases returned by the cloud server, generate task execution plan information based on the target success cases, and execute corresponding operations according to the task execution plan information; the case generation module is used to generate the latest success case based on the corresponding operation, and send the latest success case to the cloud server, so that the cloud server adds the latest success case to the dynamically optimized knowledge base. The aforementioned proxy execution device based on a dynamically optimized knowledge base enables the proxy execution system to learn from past successful experiences. When faced with repetitive or similar tasks, it can quickly retrieve and adapt existing efficient execution solutions, thereby significantly reducing the computational overhead of repetitive planning and achieving continuous and autonomous improvement in task execution efficiency.

[0087] The agent execution device based on the dynamically optimized knowledge base in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0088] The agent execution device based on the dynamically optimized knowledge base in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0089] The proxy execution device based on a dynamically optimized knowledge base provided in this application embodiment can implement all the processes implemented in Embodiment 1 above. To avoid repetition, it will not be described again here.

[0090] Example 4 Figure 4 This is a schematic diagram of the structure of the dynamically optimized knowledge base management device for agent execution provided in Embodiment 4 of this application. Figure 4 As shown, the device includes: The request parsing module 410 is used to extract standardized task target information and task context information from the successful case retrieval request when it receives a successful case retrieval request from the proxy execution client. The case retrieval module 420 is used to retrieve target success cases from the dynamic optimization knowledge base based on the standardized task target information and the task context information, and return the target success cases to the agent execution client; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; The case addition module 430 is used to, upon receiving the latest successful case from the agent execution client, determine the target successful case set in the dynamic optimization knowledge base based on the latest successful case, and add the latest successful case to the target successful case set; The case optimization module 440 is used to calculate the dynamic recall evaluation score of each successful case in the target successful case set, and optimize the target successful case set based on the dynamic recall evaluation score.

[0091] Optionally, the case optimization module 440 is specifically used for: Obtain the success rate, number of execution steps, number of recalls, and average search similarity of each successful case in the target successful case set; The average number of execution steps for the target successful case set is calculated based on the number of execution steps, and the relative recall frequency of each successful case is calculated based on the number of recalls. Based on the success rate, the number of execution steps, the average number of execution steps, the relative recall frequency, and the average retrieval similarity, calculate the dynamic recall evaluation score for each successful case in the target successful case set; The successful cases in the target successful case set are sorted in descending order according to the dynamic recall evaluation score to obtain the sorting result; If the number of successful cases in the sorting results exceeds a preset threshold, successful cases to be deleted are determined in the sorting results based on the number of successful cases and the preset threshold, and the successful cases to be deleted are deleted.

[0092] Optionally, the case retrieval module 420 is specifically used for: Based on the standardized task objective information, calculate the semantic similarity and tag similarity of each successful case in the dynamic optimization knowledge base; Calculate the environmental similarity of each successful case based on the task context information; Based on the semantic similarity, the tag similarity, and the environmental similarity, the comprehensive similarity of each successful case is calculated; Based on the comprehensive similarity, a target successful case is selected from all successful cases.

[0093] Optionally, the case retrieval module 420 is further configured to: Based on the comprehensive similarity and dynamic recall evaluation score of each successful case, the target similarity of each successful case is calculated. Based on the target similarity, a target success case is selected from each success case.

[0094] In this embodiment, the request parsing module is used to extract standardized task target information and task context information from a successful case retrieval request received from a proxy execution client. The case retrieval module is used to retrieve target successful cases from a dynamic optimization knowledge base based on the standardized task target information and the task context information, and return the target successful cases to the proxy execution client. The dynamic optimization knowledge base dynamically stores successful cases and their dynamic recall evaluation scores. The case adding module is used to determine a target successful case set from the dynamic optimization knowledge base based on the latest successful case received from the proxy execution client, and add the latest successful case to the target successful case set. The case optimization module calculates the dynamic recall evaluation score of each successful case in the target successful case set and optimizes the target successful case set based on the dynamic recall evaluation score. This dynamic optimization knowledge base management device for proxy execution, based on score calculation and optimization of similar case sets, avoids inefficient cases consuming resources, while strengthening the recall priority of high-quality cases and continuously improving the matching accuracy of the knowledge base.

[0095] The dynamically optimized knowledge base management device for agent execution in this application embodiment can be a device, or a component, integrated circuit, or chip in a terminal. The device can be a mobile electronic device or a non-mobile electronic device. For example, mobile electronic devices can be mobile phones, tablets, laptops, PDAs, in-vehicle electronic devices, wearable devices, ultra-mobile personal computers (UMPCs), netbooks, or personal digital assistants (PDAs), etc., while non-mobile electronic devices can be servers, network-attached storage (NAS), personal computers (PCs), televisions (TVs), ATMs, or self-service machines, etc. This application embodiment does not impose specific limitations.

[0096] The dynamically optimized knowledge base management device for agent execution in this application embodiment can be a device with an operating system. This operating system can be Android, iOS, or other possible operating systems; this application embodiment does not specifically limit it.

[0097] The proxy execution device based on a dynamically optimized knowledge base provided in this application embodiment can implement the various processes implemented in Embodiment 2 above. To avoid repetition, it will not be described again here.

[0098] Example 5 Figure 5 This is a schematic diagram of the structure of the electronic device provided in Embodiment 5 of this application. Figure 5 As shown, this application embodiment also provides an electronic device 500, including a processor 501, a memory 502, and a program or instructions stored in the memory 502 and executable on the processor 501. When the program or instructions are executed by the processor 501, they implement the various processes of the above-described proxy execution method embodiment based on dynamic optimization knowledge base and the dynamic optimization knowledge base management embodiment for proxy execution, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0099] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0100] Example 6 This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described proxy execution method embodiment based on dynamic optimization knowledge base and the dynamic optimization knowledge base management embodiment for proxy execution, and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0101] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0102] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0103] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0104] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

[0105] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A proxy execution method based on a dynamically optimized knowledge base, characterized in that, The method is executed by the proxy client; the method includes: The system receives natural language task information and converts it into standardized task target information; it also collects local system information and generates task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution, and a list of installed applications; A success case retrieval request is generated based on the standardized task objective information and the task context information, and the success case retrieval request is sent to the cloud server. The cloud server determines the target success case based on the success case retrieval request and the dynamically optimized knowledge base. The dynamically optimized knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases. Receive the target success case returned by the cloud server, generate task execution plan information based on the target success case, and execute the corresponding operation according to the task execution plan information; The latest successful case is generated based on the corresponding operation, and the latest successful case is sent to the cloud server, so that the cloud server can add the latest successful case to the dynamic optimization knowledge base.

2. The proxy execution method based on a dynamically optimized knowledge base according to claim 1, characterized in that, The method further includes: During the execution of the corresponding operation according to the task execution plan information, the operation screenshots, mouse click coordinates and step description texts corresponding to each execution step in the task execution plan information are obtained; Based on the operation screenshot, the mouse click coordinates, and the step description text, determine the area boundary and type label of the interactive component; Based on the region boundary and type label of the interactive component, generate the icon template image of the interactive component; The icon template image is sent to the cloud server, whereby the cloud server stores the icon template image in the dynamic optimization knowledge base.

3. A dynamically optimized knowledge base management method for agent execution, characterized in that, The method is executed by a cloud server; the method includes: Upon receiving a success case retrieval request from the agent execution client, extract the standardized task target information and task context information from the success case retrieval request; Based on the standardized task objective information and the task context information, target success cases are retrieved from the dynamic optimization knowledge base, and the target success cases are returned to the agent execution client; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; Upon receiving the latest success case from the agent execution client, a target success case set is determined in the dynamic optimization knowledge base based on the latest success case, and the latest success case is added to the target success case set; Calculate the dynamic recall evaluation score for each successful case in the target successful case set, and optimize the target successful case set based on the dynamic recall evaluation score.

4. The dynamically optimized knowledge base management method for agent-oriented execution according to claim 3, characterized in that, The step of calculating the dynamic recall evaluation score of each successful case in the target successful case set, and optimizing the target successful case set based on the dynamic recall evaluation score, includes: Obtain the success rate, number of execution steps, number of recalls, and average search similarity of each successful case in the target successful case set; The average number of execution steps for the target successful case set is calculated based on the number of execution steps, and the relative recall frequency of each successful case is calculated based on the number of recalls. Based on the success rate, the number of execution steps, the average number of execution steps, the relative recall frequency, and the average retrieval similarity, calculate the dynamic recall evaluation score for each successful case in the target successful case set; The successful cases in the target successful case set are sorted in descending order according to the dynamic recall evaluation score to obtain the sorting result; If the number of successful cases in the sorting results exceeds a preset threshold, successful cases to be deleted are determined in the sorting results based on the number of successful cases and the preset threshold, and the successful cases to be deleted are deleted.

5. The dynamically optimized knowledge base management method for agent-oriented execution according to claim 3, characterized in that, The process of retrieving target success cases from the dynamic optimization knowledge base based on the standardized task objective information and the task context information includes: Based on the standardized task objective information, calculate the semantic similarity and tag similarity of each successful case in the dynamic optimization knowledge base; Calculate the environmental similarity of each successful case based on the task context information; Based on the semantic similarity, the tag similarity, and the environmental similarity, the comprehensive similarity of each successful case is calculated; Based on the comprehensive similarity, a target successful case is selected from all successful cases.

6. The dynamically optimized knowledge base management method for agent-oriented execution according to claim 5, characterized in that, After calculating the comprehensive similarity of each successful case based on the semantic similarity, the tag similarity, and the environmental similarity, the method further includes: Based on the comprehensive similarity and dynamic recall evaluation score of each successful case, the target similarity of each successful case is calculated. Accordingly, the selection of a target successful case from among the successful cases based on the comprehensive similarity includes: Based on the target similarity, a target success case is selected from each success case.

7. A proxy execution device based on a dynamically optimized knowledge base, characterized in that, The device is configured on the agent execution client; the device includes: The information acquisition module is used to receive natural language task information and convert it into standardized task target information, collect local system information and generate task context information based on the local system information; wherein, the local system information includes operating system type, system language type, display resolution and a list of installed applications; The retrieval request module is used to generate a success case retrieval request based on the standardized task target information and the task context information, and send the success case retrieval request to the cloud server, so that the cloud server can determine the target success case based on the success case retrieval request and the dynamic optimization knowledge base; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation score of the success cases; The task execution module is used to receive the target success case returned by the cloud server, generate task execution plan information based on the target success case, and execute the corresponding operation according to the task execution plan information; The case generation module is used to generate the latest successful case based on the corresponding operation, and send the latest successful case to the cloud server, so that the cloud server can add the latest successful case to the dynamic optimization knowledge base.

8. A dynamically optimized knowledge base management device for agent execution, characterized in that, The device is configured on a cloud server; the device includes: The request parsing module is used to extract standardized task target information and task context information from the successful case retrieval request when it receives a successful case retrieval request from the proxy execution client. The case retrieval module is used to retrieve target success cases from the dynamic optimization knowledge base based on the standardized task target information and the task context information, and return the target success cases to the agent execution client; wherein, the dynamic optimization knowledge base is used to dynamically store success cases and the dynamic recall evaluation scores of the success cases; The case addition module is used to, upon receiving the latest successful case from the agent execution client, determine the target successful case set in the dynamic optimization knowledge base based on the latest successful case, and add the latest successful case to the target successful case set; The case optimization module is used to calculate the dynamic recall evaluation score of each successful case in the target successful case set, and optimize the target successful case set based on the dynamic recall evaluation score.

9. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor. When the program or instructions are executed by the processor, they implement the agent execution method based on a dynamic optimization knowledge base as described in any one of claims 1-2 and the agent-oriented dynamic optimization knowledge base management method as described in any one of claims 3-6.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a processor, implement the proxy execution method based on a dynamically optimized knowledge base as described in any one of claims 1-2 and the dynamically optimized knowledge base management method for proxy execution as described in any one of claims 3-6.

Images