Data processing method and system for an agent

Abstract

This disclosure provides a data processing method and system for intelligent agents. The data processing method includes the following steps: acquiring functional information reported by at least one functional node, and abstracting the functional node into an operator based on the functional information; constructing a target task sequence based on the input-output matching relationship and functional semantics between operators; executing the target task sequence and obtaining execution feedback; determining the evaluation result of the target task sequence based on the execution feedback; and updating the scheduling decision authority of the intelligent agent in the task processing chain in response to determining that evolutionary conditions are met based on the evaluation result. This disclosure, through the closed-loop learning mechanism of construction, execution, feedback, evaluation, and evolutionary management, enables the intelligent agent to autonomously optimize its behavior during continuous operation, significantly improving the resource utilization of various functional nodes in the system, while reducing user intervention and enhancing the interactive experience and system adaptability.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a data processing method and system for intelligent agents. Background Technology

[0002] Currently, the convergence of artificial intelligence and Internet of Things (IoT) technologies is driving the development of intelligent systems towards service-oriented and autonomous directions. Mainstream intelligent systems, especially automation platforms for home, office, or industrial scenarios, generally adopt an intent-driven interaction model. In this model, the intelligent system plays the role of a passive responder, only invoking the corresponding devices to perform specific tasks after receiving explicit instructions from the user, such as voice commands to dim living room lights or set the air conditioner temperature to 26°C.

[0003] However, as the number of devices connected to intelligent systems increases, if the above intent-driven model is still used, a large number of resources will remain idle for a long time, resulting in low resource utilization and thus affecting the user experience. Summary of the Invention

[0004] The technical problem to be solved by this disclosure is to overcome the above-mentioned defects in the prior art and provide a data processing method and system for intelligent agents that can improve resource utilization and enhance user experience.

[0005] This disclosure solves the above-mentioned technical problems through the following technical solution:

[0006] The first aspect of this disclosure provides a data processing method for an intelligent agent, comprising the following steps:

[0007] Obtain functional information reported by at least one functional node, and abstract the functional node into an operator based on the functional information; wherein, the operator includes an input interface, an output interface, and functional semantics;

[0008] Based on the input-output matching relationship and functional semantics between operators, a sequence of target tasks is constructed.

[0009] Execute the target task sequence and obtain execution feedback;

[0010] The evaluation result of the target task sequence is determined based on the execution feedback;

[0011] In response to the determination that the evolutionary conditions are met based on the evaluation results, the scheduling decision authority of the agent in the task processing link is updated.

[0012] Optionally, the functional information includes the resource consumption attributes of the functional nodes; the construction of the target task sequence based on the input-output matching relationship and functional semantics between operators includes:

[0013] In the operators, operators whose output interfaces match the input interfaces of downstream operators and whose functional semantics match the scene semantics of the target scene are connected.

[0014] Based on the connected operators, a candidate task sequence corresponding to the target scenario is generated. The total resource consumption of each candidate task sequence is determined according to the resource consumption attributes of each operator. The candidate task sequence whose total resource consumption satisfies the preset resource constraints is determined as the target task sequence.

[0015] Optionally, constructing the target task sequence based on the input-output matching relationship and functional semantics between operators includes:

[0016] Select at least two historical task sequences from long-term memory, wherein the historical task sequences are pre-connected by operators according to the input-output matching relationship and functional semantics;

[0017] Advantage segments are extracted from the at least two historical task sequences, wherein the advantage segments are continuous subsequences whose execution performance meets preset conditions, and the execution performance is determined based on the local performance data of the continuous subsequences in historical execution.

[0018] At the splicing positions between the advantageous segments, if it is confirmed that the input and output interfaces of adjacent operators match and the functional semantics meet the semantic coherence condition, the advantageous segments are combined to generate the target task sequence.

[0019] Optionally, before executing the target task sequence, the method further includes:

[0020] Detect whether there are semantic conflicts between operators in the target task sequence;

[0021] In response to the detection of a semantic conflict, a substitute operator equivalent to the conflict operator is retrieved from the semantic graph;

[0022] The conflict operator is replaced with the substitution operator to generate a new target task sequence, which serves as the semantic variation result of the target task sequence.

[0023] Optionally, the step of updating the agent's scheduling decision authority in the task processing chain in response to determining that the evolutionary conditions are met based on the evaluation results includes:

[0024] The survival weight of the target task sequence is updated based on the evaluation results; wherein, in response to the evaluation results indicating a positive change in the performance index of the target task sequence, the survival weight is increased; in response to the evaluation results indicating a negative change in the performance index of the target task sequence, the survival weight is decreased, and the survival weight is used to determine the invocation priority of the target task sequence.

[0025] In response to the survival weight satisfying the evolutionary condition, the agent's scheduling decision authority in the task processing chain is updated.

[0026] Optionally, the data processing method further includes:

[0027] In response to determining that the scenario solidification conditions are met based on the evaluation results, the target task sequence is bound to the target scenario, and a solidified template for the target scenario is generated; wherein, the solidified template is used to characterize the target task sequence;

[0028] In response to the current scene matching the target scene, the target task sequence represented by the fixed template is invoked.

[0029] A second aspect of this disclosure provides a data processing system for an intelligent agent, comprising:

[0030] An acquisition module is used to acquire functional information reported by at least one functional node, and abstract the functional node into an operator based on the functional information; wherein, the operator includes an input interface, an output interface, and functional semantics;

[0031] The building module is used to construct the target task sequence based on the input-output matching relationship and functional semantics between operators;

[0032] An execution module is used to execute the target task sequence and obtain execution feedback;

[0033] An evaluation module is used to determine the evaluation result of the target task sequence based on the execution feedback;

[0034] An update module is used to update the agent's scheduling decision authority in the task processing chain in response to determining that the evolution conditions are met based on the evaluation results.

[0035] Optionally, the functional information includes the resource consumption attributes of the functional nodes; the construction module is specifically used to connect operators whose output interfaces match the input interfaces of downstream operators and whose functional semantics match the scene semantics of the target scene; and to generate a candidate task sequence corresponding to the target scene based on the connected operators, determine the total resource consumption of each candidate task sequence according to the resource consumption attributes of each operator, and determine the candidate task sequence whose total resource consumption satisfies the preset resource constraints as the target task sequence.

[0036] Optionally, the building module specifically includes:

[0037] A selection unit is used to select at least two historical task sequences from long-term memory, the historical task sequences being pre-connected by operators according to the input-output matching relationship and functional semantics;

[0038] An extraction unit is configured to extract advantageous segments from the at least two historical task sequences, wherein the advantageous segments are continuous subsequences whose execution performance meets preset conditions, and the execution performance is determined based on local performance data of the continuous subsequences in historical execution.

[0039] The generation unit is used to combine the advantageous segments at the splicing positions between the advantageous segments, provided that the input-output interfaces of adjacent operators match and the functional semantics satisfy the semantic coherence condition, to generate the target task sequence.

[0040] Optionally, the data processing system further includes:

[0041] The detection module is used to detect whether there are semantic conflicts between operators in the target task sequence;

[0042] The retrieval module is used to retrieve, in response to the detection of a semantic conflict, an alternative operator that is functionally equivalent to the conflict operator in the semantic graph;

[0043] The replacement module is used to replace the conflict operator with the substitution operator to generate a new target task sequence as a semantic variation result of the target task sequence.

[0044] Optionally, the update module specifically includes:

[0045] A weight update unit is used to update the survival weight of the target task sequence based on the evaluation result; wherein, in response to the evaluation result indicating that the performance index of the target task sequence is changing positively, the survival weight is increased; in response to the evaluation result indicating that the performance index of the target task sequence is changing negatively, the survival weight is decreased, and the survival weight is used to determine the calling priority of the target task sequence.

[0046] The permission update unit is used to update the agent's scheduling decision permissions in the task processing link in response to the survival weight satisfying the evolution conditions.

[0047] Optionally, the data processing system further includes:

[0048] A binding module is used to bind the target task sequence to a target scene in response to determining that the scene solidification conditions are met based on the evaluation results, and to generate a solidified template for the target scene; wherein the solidified template is used to characterize the target task sequence;

[0049] The calling module is used to call the target task sequence represented by the fixed template in response to the matching of the current scene with the target scene.

[0050] A third aspect of this disclosure provides an electronic device including a memory, a processor, and a computer program stored in the memory and for running on the processor, wherein the processor executes the computer program to implement the data processing method provided in the first aspect described above.

[0051] A fourth aspect of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the data processing method provided in the first aspect above.

[0052] The fifth aspect of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the data processing method provided in the first aspect.

[0053] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this disclosure.

[0054] The positive and progressive effects of this disclosure are as follows: This disclosure obtains the functional information of functional nodes and abstracts it into operators. Based on the input-output matching relationship and functional semantics between operators, it realizes the adaptive construction of the target task sequence. By executing the target task sequence and obtaining execution feedback, it determines the evaluation result of the target task sequence based on the execution feedback. In response to the determination that the evolution conditions are met based on the evaluation result, it updates the scheduling decision authority of the agent in the task processing link. This enables the agent to autonomously optimize its own behavior during continuous operation, significantly improves the resource utilization of various functional nodes in the system, reduces user intervention, and enhances the interactive experience and system adaptability. Attached Figure Description

[0055] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0056] Figure 1 A flowchart illustrating a data processing method for an intelligent agent, provided as an exemplary embodiment of this disclosure;

[0057] Figure 2 A flowchart illustrating a data processing method for an intelligent agent, provided as an exemplary embodiment of this disclosure;

[0058] Figure 3 A flowchart of step S12 provided for an exemplary embodiment of this disclosure;

[0059] Figure 4 A flowchart of step S12 provided for an exemplary embodiment of this disclosure;

[0060] Figure 5 A structural block diagram of a data processing system for an intelligent agent provided as an exemplary embodiment of this disclosure;

[0061] Figure 6 This is a schematic diagram of the structure of an electronic device provided as an exemplary embodiment of the present disclosure. Detailed Implementation

[0062] The present disclosure is further illustrated below by way of embodiments, but the present disclosure is not limited to the scope of the embodiments described herein.

[0063] The prefixes such as "first" and "second" used in this disclosure are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this disclosure does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not be construed as an unnecessary limitation. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0064] In this embodiment of the disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of user personal information comply with relevant laws and regulations and do not violate public order and good morals.

[0065] The intelligent agent involved in this disclosure refers to an autonomous computing entity capable of perceiving the environment, executing decisions, and completing specific tasks. The intelligent agent can exist as a software program, a hardware module, or a combination of both. The intelligent agent runs on an electronic device, including but not limited to: servers, personal computers, smart speakers, smartphones, tablets, IoT gateways, embedded devices, robots, and any other physical device with computing, storage, and communication capabilities.

[0066] The functional nodes involved in the embodiments of this disclosure refer to entities that can provide independent, callable, and composable specific functions or operations. The specific form of a functional node can be a physical device, such as a sensor, air conditioner, camera, or switch; it can also be a software service, such as a weather query service or calendar reminder service; or it can be an algorithm model, such as an image recognition model or a speech synthesis model.

[0067] The system disclosed in this embodiment is deployed on an electronic device, specifically referring to the overall architecture that provides an operating environment for intelligent agents and provides registration, management, scheduling and data storage for each functional node.

[0068] The task sequence involved in the embodiments of this disclosure can also be called a collaboration chain or workflow, which refers to a logical sequence composed of one or more functional nodes in the order of execution, used to represent a series of operation steps that an intelligent agent needs to perform to complete a certain goal.

[0069] This disclosure provides a closed-loop learning mechanism for an intelligent agent, specifically including: acquiring functional information reported by at least one functional node, and abstracting the functional node into an operator based on the functional information; wherein the operator includes an input interface, an output interface, and functional semantics; constructing a target task sequence based on the input-output matching relationship and functional semantics between operators; executing the target task sequence and obtaining execution feedback; determining the evaluation result of the target task sequence based on the execution feedback; and updating the scheduling decision authority of the intelligent agent in the task processing link in response to determining that evolutionary conditions are met based on the evaluation result. Through the above-described closed-loop learning mechanism of construction, execution, feedback, evaluation, and evolutionary management, the intelligent agent can autonomously optimize its own behavior during continuous operation, significantly improving the resource utilization of various functional nodes in the system, while reducing user intervention and enhancing the interactive experience and system adaptability.

[0070] In practice, intelligent agents utilize the computing, storage, and communication resources of electronic devices to construct, execute, evaluate, and manage the aforementioned sequence of target tasks.

[0071] Figure 1 This is a flowchart illustrating a data processing method for an intelligent agent, provided as an exemplary embodiment of the present disclosure. This data processing method can be executed by a data processing system, which can be implemented through software and / or hardware. The data processing system can be part or all of an electronic device. The data processing method for an intelligent agent provided in this embodiment is described below using an intelligent agent as the executing entity.

[0072] like Figure 1 As shown, the data processing method for an intelligent agent provided in this embodiment may include the following steps S11~S15:

[0073] Step S11: Obtain functional information reported by at least one functional node, and abstract the functional node into an operator based on the functional information. The operator includes an input interface, an output interface, and functional semantics. The functional information includes relevant information about the functions possessed by the at least one functional node. In the intelligent agent monitoring system, each functional node reports its own functional information. For example, functional node A provides the function of "obtaining the current room temperature," and its output is a temperature value; functional node B provides the function of "adjusting the air conditioner temperature," and its input is the set temperature.

[0074] Step S12: Based on the input-output matching relationship and functional semantics between operators, construct a target task sequence. The input interface of each operator corresponds to the input constraints of a functional node, and the output interface corresponds to the output constraints. The agent connects multiple operators sequentially by matching the output of one operator with the input of another to form the target task sequence. For example, connecting the output of the "obtain room temperature" operator to the input of the "adjust air conditioning" operator constructs the target task sequence "automatically adjust temperature based on room temperature".

[0075] In a specific example, in a smart home scenario, the target task sequence named "Away Mode" includes the following steps: Step a, call the door magnetic sensor to detect whether the door is closed; Step b, call the light controller to turn off all lights; Step c, call the air conditioner controller to turn off the air conditioner; Step d, call the security camera to activate security mode.

[0076] Step S13: Execute the target task sequence and obtain execution feedback. The agent calls the corresponding functional nodes to execute each action in the task sequence sequentially. During execution, the agent records execution feedback, which may include the success or failure status returned by each functional node, CPU time, network latency, etc. CPU time refers to the processor time consumed in executing the target task sequence, reflecting the computational resource usage of the target task sequence. Network latency refers to the time elapsed from sending a request to receiving a response when calling a remote functional node, such as a cloud service, reflecting the degree of dependence of the target task sequence on network resources.

[0077] In some examples, the execution feedback includes single-execution feedback, which records the success or failure status, CPU time, and network latency of each functional node during a single execution. In other examples, the execution feedback includes multiple-execution feedback, which statistically analyzes the success rate, average CPU time, and average network latency of the target task sequence after multiple executions.

[0078] Step S14: Determine the evaluation result of the target task sequence based on the execution feedback.

[0079] It should be noted that the purpose of evaluating the target task sequence is to determine its execution performance, providing a basis for subsequent evolutionary management. In some examples, the agent classifies the target task sequence into grades such as excellent, good, and poor based on different performance indicators in the execution feedback. In other examples, the agent scores the target task sequence based on different performance indicators in the execution feedback. These performance indicators may include success rate, energy consumption, latency penalty, etc. In a specific example, the score of the target task sequence is calculated according to the following formula:

[0080] Score = (Success Rate × Weight) − (Energy Consumption + Delay Penalty);

[0081] The success rate is determined based on the success or failure status of functional nodes. For example, if a target task sequence is executed 100 times, and all functional nodes return a success status 98 times, while 2 times it fails due to a functional node malfunction or timeout, the success rate is 98%. Weights can be dynamically adjusted based on user or environmental feedback. For instance, if the execution of the target task sequence is not interrupted by the user, or the execution result is not corrected by the user, it indicates that the target task sequence meets user expectations, and the weight can be appropriately increased. Otherwise, it indicates that the target task sequence deviates from the user's actual needs, and the weight can be appropriately decreased. Energy consumption is determined based on CPU time; the longer the CPU time, the greater the energy consumption. For example, if the mapping relationship between CPU time and energy consumption is set to 0.01 units of energy consumption per 1ms, and the CPU time for executing the target task sequence is 50ms, then the corresponding energy consumption is 0.5 units. Latency penalty is determined based on network latency; the longer the network latency, the longer the user waits and the slower the system response, thus resulting in a higher latency penalty. For example, the mapping relationship between network latency and latency penalty can be set as 0.005 latency penalty for every 1ms. If the network latency is 200ms, then the corresponding latency penalty is 1.

[0082] Step S15: In response to determining that the evolutionary conditions are met based on the evaluation results, update the agent's scheduling decision authority in the task processing link. Here, the agent's scheduling decision authority in the task processing link refers to the task flow control authority the agent possesses within the task processing link. The task processing link includes multiple agents and / or functional nodes, and tasks flow sequentially through each node along the link for processing.

[0083] In some examples, the evolution condition is determined to be met when the level of the target task sequence reaches a preset level. In some examples, the evolution condition is determined to be met when the score of the target task sequence reaches a preset score.

[0084] When the evaluation results determine that the evolutionary conditions are met, the agent's scheduling decision authority is updated. The direction of this update is correlated with the trend of execution capability changes reflected in the evaluation results. In some examples, when the evaluation results indicate that the agent has higher execution accuracy and processing efficiency for a specific type of task, the agent's task acquisition priority is increased, enabling the agent to prioritize tasks matching its expertise during subsequent task distribution. In some examples, when the evaluation results indicate that the agent has the ability to successfully coordinate multiple functional nodes to complete complex tasks, the agent is granted orchestration authority to decompose the target task into subtasks and issue execution instructions to other agents or functional nodes.

[0085] In this embodiment, the role of the agent in the task processing chain is dynamically adjusted by updating the agent's scheduling decision authority. This allows the agent to adaptively change its position and function in the task processing chain according to the evolution of its execution capabilities, thereby optimizing the efficiency and adaptability of the overall task processing.

[0086] In one of the optional implementation methods, such as Figure 2 As shown, before step S13, the following steps S131~S133 are also included:

[0087] Step S131: Detect whether there is a semantic conflict among the operators in the target task sequence. If yes, proceed to step S132; otherwise, proceed to step S13. The agent analyzes the logical relationships between the operators in the target task sequence to determine whether there is a semantic conflict. In some examples, semantic conflicts include mismatched input and output types of adjacent operators, such as the output type of the previous operator being text, and the input type of the next operator being an image. In some examples, semantic conflicts include mutually exclusive side effects between operators, such as one operator requiring cooling, and another operator requiring heating.

[0088] Step S132: Retrieve alternative operators in the semantic graph that are functionally equivalent to the conflicting operator. Functional equivalence means that the alternative operator is substitutable for the conflicting operator in terms of task processing capability; that is, the alternative operator can assume the functional role of the conflicting operator in the target task sequence, and the substitution does not disrupt the overall execution logic and semantic coherence of the target task sequence. For example, if the original conflicting operator is "turn on strong lighting," but the subsequent operator requires "create a sleep environment," then "turn on night light mode" is retrieved as the alternative.

[0089] The semantic graph is a knowledge structure used to describe operators and their semantic relationships. The semantic graph organizes data in the form of a graph, where nodes represent abstract operators and edges represent semantic associations between operators. The semantic graph can be stored in the system's long-term memory, supporting efficient vector retrieval and graph traversal operations.

[0090] Step S133: Replace the conflict operator with the substitution operator to generate a new target task sequence, which serves as the semantic variation result of the target task sequence. The agent replaces the original conflict operator with the substitution operator to generate a new target task sequence without semantic conflict, and then executes the new target task sequence.

[0091] In this implementation, a semantic conflict is detected before the execution of the target task sequence, and the conflict operator is replaced if a semantic conflict exists. In essence, this is a semantic mutation operation. Semantic mutation can effectively improve the execution success rate of the target task sequence and reduce resource waste and user experience degradation caused by logical conflicts.

[0092] In one optional implementation of step S12, the functional information includes the resource consumption attributes of the functional node. These resource consumption attributes can characterize the system resources occupied by the functional node when it is invoked for execution, such as processor utilization or computational load, storage resource consumption, communication resource consumption, and time consumption. Figure 3 As shown, the specific steps include S211~S212:

[0093] Step S211: In the operators, connect operators whose output interfaces match the input interfaces of downstream operators and whose functional semantics match the scene semantics of the target scene. Through dual verification of interface matching and semantic matching, it can be ensured that the constructed operator connection topology is a valid task execution link at both the syntactic and semantic levels.

[0094] Step S212: Generate candidate task sequences corresponding to the target scenario based on the connected operators. Determine the total resource consumption of each candidate task sequence according to the resource consumption attributes of each operator, and determine the candidate task sequence whose total resource consumption satisfies the preset resource constraints as the target task sequence. Here, the candidate task sequence refers to a task execution path formed by arranging the connected operators in the connection order, capable of realizing the functions required by the target scenario. After generating the candidate task sequences, determine the total resource consumption of each candidate task sequence based on the resource consumption attributes of each operator in the candidate task sequence.

[0095] The determination of the total resource consumption depends on the execution dependencies between the operators in the candidate task sequence. Specifically, when the operators in the candidate task sequence are executed serially, the total resource consumption is the sum of the resource consumption attribute values ​​of each operator; when the operators in the candidate task sequence are executed in parallel, the total resource consumption is the maximum value among the resource consumption attribute values ​​of each operator in the parallel branch; when the candidate task sequence is a mixed serial and parallel execution topology, the total resource consumption is determined based on a combination of serial accumulation and parallel maximum value taking.

[0096] In step S212, after determining the total resource consumption of each candidate task sequence, the total resource consumption of each candidate task sequence is compared with preset resource constraints, and the candidate task sequence whose total resource consumption meets the preset resource constraints is determined as the final target task sequence. The preset resource constraints can be pre-configured based on available resource capacity or a user-defined resource limit. For example, the preset resource constraints can be set to a total execution time not exceeding a preset duration, or a total computing resource consumption not exceeding the current available computing resource limit.

[0097] In this implementation, by introducing resource consumption attributes as the screening criteria for candidate task sequences, the final determined target task sequence can meet the actual carrying capacity of the system in terms of resource consumption, thereby avoiding execution failure or system overload caused by resource overload.

[0098] In another optional implementation of step S12, such as Figure 4 As shown, the specific steps include S221~S223:

[0099] Step S221: Select at least two task sequences from the historical task sequences, which are pre-connected by operators according to the input-output matching relationship and functional semantics. The historical task sequences refer to task sequences that have been previously executed and are stored in long-term memory. Each historical task sequence may include its evaluation result. In some examples, the agent selects two task sequences with better evaluation results from the historical task sequences stored in long-term memory as parents.

[0100] Step S222: Extract advantageous segments from the at least two task sequences respectively. The advantageous segments are continuous subsequences whose execution performance meets preset conditions. The execution performance is determined based on local performance data of the continuous subsequences during historical execution. The local performance data refers to the performance indicators recorded during the historical execution of the continuous subsequences as independent execution units, specifically including execution success rate, resource consumption, and execution time. Correspondingly, the preset conditions can be that the execution success rate of the continuous subsequence is higher than a first preset threshold, the resource consumption of the continuous subsequence is lower than a second preset threshold, and the execution time of the continuous subsequence is lower than a third preset threshold. In a specific example, the "lights dimming" segment with an execution success rate higher than the first preset threshold is extracted from the "cinema mode" task sequence, and the "detected sleep triggering" segment with a resource consumption lower than the second preset threshold is extracted from the "sleep monitoring" task sequence.

[0101] Step S223: At the splicing positions between the advantageous segments, if it is confirmed that the input-output interfaces of adjacent operators match and the functional semantics satisfy the semantic coherence condition, the advantageous segments are combined to generate the target task sequence. In some examples, the semantic coherence condition is that the semantic similarity is higher than a preset similarity. In some examples, the semantic coherence condition is that there is a direct semantic association edge in the semantic graph. The agent combines the extracted advantageous segments in a way that matches interfaces and is semantically compatible to form a new target task sequence, such as "automatically dimming the lights after detecting that the user has fallen asleep". It should be noted that the generated target task sequence integrates the advantageous segments in the parent sequence and usually has better overall performance than any parent sequence.

[0102] The extraction and combination of advantageous segments in this implementation is essentially a semantic cross-operation. Through semantic cross-operation, verified advantageous segments can be reused, improving the quality and efficiency of generating new task sequences, and potentially generating new cross-domain collaborative logic, thereby enhancing the system's innovation capabilities.

[0103] In one optional implementation, step S15 above includes the following steps S151~S152:

[0104] Step S151: Update the survival weight of the target task sequence based on the evaluation result; wherein, in response to the evaluation result indicating a positive change in the performance index of the target task sequence, the survival weight is increased; in response to the evaluation result indicating a negative change in the performance index of the target task sequence, the survival weight is decreased. The survival weight is used to determine the priority of the target task sequence being invoked. The higher the survival weight of the target task sequence, the better its performance in terms of execution effect, resource consumption, or user satisfaction. In the future, when faced with multiple candidate task sequences, the agent will prioritize invoking sequences with high survival weights. Understandably, task sequences with good execution effects will be repeatedly invoked by the agent, and their survival weights will increase; task sequences with poor execution effects will gradually be marginalized, and their survival weights will decrease. The survival weight-based priority mechanism provided in this example enables the agent to automatically filter out efficient task sequences and give them priority in subsequent invocations, thereby continuously optimizing system behavior.

[0105] Step S152: In response to the survival weight satisfying the evolution condition, update the agent's scheduling decision authority in the task processing chain. In some examples, the evolution condition is that the survival weight reaches a preset evolution threshold. Specifically, when the survival weight of a target task sequence continuously accumulates and exceeds the evolution threshold, it is determined that the survival weight satisfies the evolution condition, and at this time, the update of the agent's scheduling decision authority is triggered.

[0106] In one optional implementation, the above data processing method further includes the following steps S16-S17:

[0107] Step S16: In response to determining that the scene solidification condition is met based on the evaluation result, the target task sequence is bound to the target scene, and a solidified template of the target scene is generated; wherein, the solidified template is used to characterize the target task sequence. In some examples, the preset condition is that the score of the target task sequence is higher than a certain threshold, for example, a score ≥ 0.5.

[0108] Step S17: In response to the matching of the current scene with the target scene, invoke the target task sequence represented by the solidified template.

[0109] In this example, a sequence of target tasks is bound to a specific target scenario and stored in long-term memory to achieve the solidification of the target task sequence. When a target task sequence is repeatedly verified to have high adaptability in a target scenario, such as automatically activating security monitoring when a user is at home, the agent generates a solidified template for that target scenario by binding the target task sequence to that scenario. Subsequently, whenever the agent detects the same scenario through sensor data or user behavior, it directly invokes the solidified template without user intervention or re-evaluation. In this way, the agent can transform verified efficient behaviors into instinctive responses, thereby improving response speed and user experience.

[0110] Based on the above method embodiments, an exemplary embodiment of this disclosure also provides a data processing system for an intelligent agent, such as... Figure 5 As shown, the system includes an acquisition module 501, a construction module 502, an execution module 503, an evaluation module 504, and an update module 505. The acquisition module 501 acquires functional information reported by at least one functional node and abstracts the functional node into an operator based on the functional information; wherein the operator includes an input interface, an output interface, and functional semantics. The construction module 502 constructs a target task sequence based on the input-output matching relationship and functional semantics between operators. The execution module 503 executes the target task sequence and obtains execution feedback. The evaluation module 504 determines the evaluation result of the target task sequence based on the execution feedback. The update module 505 updates the agent's scheduling decision authority in the task processing chain in response to the determination that evolutionary conditions are met based on the evaluation result.

[0111] In one optional implementation, the functional information includes the resource consumption attributes of the functional nodes; the construction module 502 is specifically used to connect operators whose output interfaces match the input interfaces of downstream operators and whose functional semantics match the scene semantics of the target scene; and to generate a candidate task sequence corresponding to the target scene based on the connected operators, determine the total resource consumption of each candidate task sequence according to the resource consumption attributes of each operator, and determine the candidate task sequence whose total resource consumption satisfies the preset resource constraints as the target task sequence.

[0112] In one optional implementation, the construction module 502 specifically includes a selection unit, an extraction unit, and a generation unit. The selection unit selects at least two historical task sequences from long-term memory, the historical task sequences being pre-connected by operators according to the input-output matching relationship and functional semantics. The extraction unit extracts advantageous segments from the at least two historical task sequences, wherein the advantageous segments are continuous subsequences whose execution performance meets preset conditions, and the execution performance is determined based on local performance data of the continuous subsequences in historical execution. The generation unit combines the advantageous segments at the splicing positions between them, provided that the input-output interfaces of adjacent operators match and the functional semantics meet the semantic coherence condition, to generate the target task sequence.

[0113] In one optional implementation, the data processing system further includes a detection module, a retrieval module, and a replacement module. The detection module detects whether semantic conflicts exist between operators in the target task sequence. The retrieval module, in response to the detection of a semantic conflict, retrieves a substitute operator functionally equivalent to the conflicting operator from the semantic graph. The replacement module replaces the conflicting operator with the substitute operator, generating a new target task sequence as a semantic variation result of the target task sequence.

[0114] In one optional implementation, the update module specifically includes: a weight update unit, configured to update the survival weight of the target task sequence based on the evaluation result; wherein, in response to the evaluation result indicating a positive change in the performance index of the target task sequence, the survival weight is increased; in response to the evaluation result indicating a negative change in the performance index of the target task sequence, the survival weight is decreased, and the survival weight is used to determine the calling priority of the target task sequence; and a permission update unit, configured to update the agent's scheduling decision permission in the task processing chain in response to the survival weight satisfying the evolution condition.

[0115] In one optional implementation, the data processing system further includes: a binding module, configured to bind the target task sequence to a target scene in response to determining that the scene solidification conditions are met based on the evaluation results, and generate a solidified template for the target scene; wherein the solidified template is used to characterize the target task sequence; and a calling module, configured to call the target task sequence characterized by the solidified template in response to the current scene matching the target scene.

[0116] It should be noted that the data processing system in this embodiment can be a separate chip, chip module, or electronic device, or it can be a chip or chip module integrated into an electronic device.

[0117] Regarding the various modules / units included in the data processing system described in this embodiment, they may be software modules / units, hardware modules / units, or a combination of both.

[0118] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0119] Figure 6 This is a schematic diagram of an electronic device provided as an exemplary embodiment of the present disclosure. The electronic device includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the steps of the data processing method of the intelligent agent described above. Figure 6 The electronic device 3 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0120] The components of the electronic device 3 may include, but are not limited to: at least one processor 4, at least one memory 5, and a bus 6 connecting different system components (including memory 5 and processor 4).

[0121] Bus 6 includes a data bus, an address bus, and a control bus.

[0122] The memory 5 may include volatile memory, such as random access memory (RAM) 51 and / or cache memory 52, and may further include read-only memory (ROM) 53.

[0123] The memory 5 may also include a program tool 55 having a set (at least one) of program modules 54, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0124] Processor 4 executes various functional applications and data processing by running computer programs stored in memory 5, such as the data processing method of the aforementioned intelligent agent.

[0125] Electronic device 3 can also communicate with one or more external devices 7 (e.g., keyboard, pointing device, etc.). This communication can be performed through input / output (I / O) interface 8. Furthermore, electronic device 3 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 9. Figure 6 As shown, network adapter 9 communicates with other modules of electronic device 3 via bus 6. It should be understood that, although... Figure 6 Not shown, it can be combined with electronic device 3 to use other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0126] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0127] An exemplary embodiment of this disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the data processing method of the intelligent agent described above.

[0128] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0129] In possible implementations, this disclosure can also be implemented as a computer program product comprising a computer program that, when executed by a processor, implements the steps of the data processing method for the intelligent agent described above.

[0130] The computer program for executing the present disclosure can be written in any combination of one or more programming languages, and the computer program can be executed entirely on an electronic device, partially on an electronic device, as a stand-alone software package, partially on an electronic device and partially on a remote device, or entirely on a remote device.

[0131] While specific embodiments of this disclosure have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this disclosure is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this disclosure, but all such changes and modifications fall within the scope of protection of this disclosure.

Claims

1. A data processing method for an intelligent agent, characterized in that, Includes the following steps: Obtain functional information reported by at least one functional node, and abstract the functional node into an operator based on the functional information; wherein, the operator includes an input interface, an output interface, and functional semantics; Based on the input-output matching relationship and functional semantics between operators, a sequence of target tasks is constructed. Execute the target task sequence and obtain execution feedback; The evaluation result of the target task sequence is determined based on the execution feedback; In response to the determination that the evolutionary conditions are met based on the evaluation results, the scheduling decision authority of the agent in the task processing link is updated.

2. The data processing method as described in claim 1, characterized in that, The functional information includes the resource consumption attributes of the functional nodes; the construction of the target task sequence based on the input-output matching relationship and functional semantics between operators includes: In the operators, operators whose output interfaces match the input interfaces of downstream operators and whose functional semantics match the scene semantics of the target scene are connected. Based on the connected operators, a candidate task sequence corresponding to the target scenario is generated. The total resource consumption of each candidate task sequence is determined according to the resource consumption attributes of each operator. The candidate task sequence whose total resource consumption satisfies the preset resource constraints is determined as the target task sequence.

3. The data processing method as described in claim 1, characterized in that, The construction of the target task sequence based on the input-output matching relationship and functional semantics between operators includes: Select at least two historical task sequences from long-term memory, wherein the historical task sequences are pre-connected by operators according to the input-output matching relationship and functional semantics; Advantage segments are extracted from the at least two historical task sequences, wherein the advantage segments are continuous subsequences whose execution performance meets preset conditions, and the execution performance is determined based on the local performance data of the continuous subsequences in historical execution. At the splicing positions between the advantageous segments, if it is confirmed that the input and output interfaces of adjacent operators match and the functional semantics meet the semantic coherence condition, the advantageous segments are combined to generate the target task sequence.

4. The data processing method as described in claim 1, characterized in that, Before executing the target task sequence, the following is also included: Detect whether there are semantic conflicts between operators in the target task sequence; In response to the detection of a semantic conflict, a substitute operator equivalent to the conflict operator is retrieved from the semantic graph; The conflict operator is replaced with the substitution operator to generate a new target task sequence, which serves as the semantic variation result of the target task sequence.

5. The data processing method as described in claim 1, characterized in that, The response to determining that the evolutionary conditions are met based on the evaluation results, and updating the agent's scheduling decision authority in the task processing chain, includes: The survival weight of the target task sequence is updated based on the evaluation results; wherein, in response to the evaluation results indicating a positive change in the performance index of the target task sequence, the survival weight is increased; in response to the evaluation results indicating a negative change in the performance index of the target task sequence, the survival weight is decreased, and the survival weight is used to determine the invocation priority of the target task sequence. In response to the survival weight satisfying the evolutionary condition, the agent's scheduling decision authority in the task processing chain is updated.

6. The data processing method according to any one of claims 1-5, characterized in that, The data processing also includes: In response to determining that the scenario solidification conditions are met based on the evaluation results, the target task sequence is bound to the target scenario, and a solidified template for the target scenario is generated; wherein, the solidified template is used to characterize the target task sequence; In response to the current scene matching the target scene, the target task sequence represented by the fixed template is invoked.

7. A data processing system for an intelligent agent, characterized in that, include: An acquisition module is used to acquire functional information reported by at least one functional node, and abstract the functional node into an operator based on the functional information; wherein, the operator includes an input interface, an output interface, and functional semantics; The building module is used to construct the target task sequence based on the input-output matching relationship and functional semantics between operators; An execution module is used to execute the target task sequence and obtain execution feedback; An evaluation module is used to determine the evaluation result of the target task sequence based on the execution feedback; An update module is used to update the agent's scheduling decision authority in the task processing chain in response to determining that the evolution conditions are met based on the evaluation results.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the data processing method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the data processing method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the data processing method as described in any one of claims 1-6.

Images