Interaction-driven agent skill generation and adaptive evolution method and device

Abstract

The application discloses an interactive driving-based skill generation and adaptive evolution method and device for an intelligent agent, which comprises the following steps: obtaining interactive data in the interactive process of a user and a human-computer interaction system, generating a behavior log set based on the interactive data and storing the behavior log set into a log database; if it is determined that an evolution trigger condition is met, obtaining a target operation sequence set based on the log database; obtaining a new skill corresponding to each target operation sequence based on the target operation sequence set and a pre-trained skill generation model; performing skill merging and updating on a skill library in the human-computer interaction system based on the new skill corresponding to each target operation sequence, so as to obtain an updated skill library; generating new knowledge based on the interactive data and the behavior log set, and updating a knowledge base in the human-computer interaction system by using the new knowledge, so as to obtain an updated knowledge base. The application can automatically capture, learn and update skills and knowledge in the natural human-computer interaction process of a user completing a task.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for generating and adaptively evolving intelligent agent skills based on interaction-driven processes. Background Technology

[0002] Current human-computer interaction systems rely heavily on users' tacit knowledge when handling complex tasks, especially knowledge-intensive ones. Despite continuous advancements in human-computer interaction technology, existing systems still face several critical technical challenges: First, expert knowledge cannot be effectively accumulated and reused. Systems cannot record valuable experiences such as user judgment rules, search strategies, and corrective behaviors during task processing, nor can they structure these experiences into reusable knowledge units. This leads to the loss of expert tacit knowledge upon task completion, resulting in the loss of experience as users leave. New users or new tasks are forced to repeat the same exploration and trial-and-error process, causing significant duplication of effort and hindering the effective accumulation and reuse of knowledge. Second, the human-computer collaboration process is unidirectionally fragmented, with a heavy interactive burden. AI assisting users in completing tasks (e.g., providing pre-labeled data) and users helping AI evolve (e.g., providing feedback data) are two separate stages. If users want system improvements, they must incur additional, deliberate calibration costs, such as specifically labeling a batch of data, filling out structured feedback forms, or participating in model retraining. However, this design disrupts the natural task processing workflow, treating system usage and system improvement as mutually exclusive activities. This forces users to choose between efficiency (using only AI assistance) or long-term benefits (paying extra costs to help AI evolve), making it difficult to achieve both. Third, system capabilities are rigid, lacking a continuous evolution mechanism. Once deployed, traditional models' capabilities remain fixed at the training level. While some research introduces active or online learning, dedicated retraining processes or model updates are still required, failing to achieve continuous, low-cost evolution through learning from daily interactions. The system cannot automatically accumulate experience and optimize performance as usage time increases or the user base expands; its assistance capabilities remain at the initial level. Fourth, the lack of a closed-loop collaborative mechanism for bidirectional reinforcement means AI cannot learn from natural human interactions and continuously evolve, and humans cannot benefit from the accumulated experience of predecessors, resulting in the potential of human-machine collaboration being far from being fully explored.

[0003] Therefore, existing human-computer interaction systems are unable to capture and structure users' valuable experiences during task processing, resulting in the inability to effectively accumulate and reuse knowledge. Summary of the Invention

[0004] This invention provides a method and apparatus for generating and adaptively evolving intelligent agent skills based on interaction-driven approaches. It aims to solve the problem that existing human-computer interaction systems cannot capture and structure users' valuable experiences during task processing, resulting in the inability to effectively accumulate and reuse knowledge.

[0005] In a first aspect, embodiments of the present invention provide a method for generating and adaptively evolving intelligent agent skills based on interaction-driven processes, comprising: In response to a task execution command, the system acquires interaction data during the interaction between the user and the human-computer interaction system, generates a behavior log set based on the interaction data, and stores it in a preset log database. If the preset evolution triggering conditions are met, the target operation sequence set is obtained based on the log database; Based on the target operation sequence set and the pre-trained skill generation model, a new skill corresponding to each target operation sequence in the target operation sequence set is obtained; The skill library in the human-computer interaction system is updated by merging skills based on the new skills corresponding to each target operation sequence, resulting in an updated skill library. New knowledge is generated based on the interaction data and the behavior log set, and the knowledge base in the human-computer interaction system is updated using the new knowledge to obtain the updated knowledge base.

[0006] Secondly, embodiments of the present invention also provide an interaction-driven intelligent agent skill generation and adaptive evolution apparatus, the apparatus being used to execute the interaction-driven intelligent agent skill generation and adaptive evolution method of the first aspect described above, the apparatus comprising: The behavior recording unit is used to respond to task execution instructions, acquire interaction data during the interaction between the user and the human-computer interaction system, generate a behavior log set based on the interaction data, and store it in a preset log database; The behavior mining unit is used to obtain a target operation sequence set based on the log database if it is determined that a preset evolution triggering condition is met. The skill generation unit is used to obtain a new skill corresponding to each target operation sequence in the target operation sequence set based on the target operation sequence set and the pre-trained skill generation model; The skill management unit is used to merge and update the skill library in the human-computer interaction system based on the new skill corresponding to each target operation sequence, so as to obtain the updated skill library. The knowledge update unit is used to generate new knowledge based on the interaction data and the behavior log set, and to update the knowledge base of the human-computer interaction system using the new knowledge to obtain the updated knowledge base.

[0007] Thirdly, embodiments of the present invention provide a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the interactive-driven intelligent agent skill generation and adaptive evolution method of the first aspect described above.

[0008] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the interaction-driven intelligent agent skill generation and adaptive evolution method described in the first aspect.

[0009] This invention provides a method and apparatus for intelligent agent skill generation and adaptive evolution based on interaction-driven approaches. The method acquires interaction data during user-human-computer interaction (HCI) processes, generates a behavior log set based on the interaction data, and stores it in a log database. If evolution triggering conditions are met, a target operation sequence set is obtained from the log database. Based on the target operation sequence set and a pre-trained skill generation model, a new skill corresponding to each target operation sequence is obtained. The skill library in the HCI system is merged and updated based on the new skills corresponding to each target operation sequence, resulting in an updated skill library. New knowledge is generated based on the interaction data and behavior log set, and this new knowledge is used to update the knowledge base in the HCI system, resulting in an updated knowledge base. This invention can automatically capture, learn, and update skills and knowledge during natural HCI interactions as users complete tasks. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an interaction-driven intelligent agent skill generation and adaptive evolution method according to an embodiment of the present invention. Figure 2 A schematic block diagram of an interactive-driven intelligent agent skill generation and adaptive evolution device provided in an embodiment of the present invention; Figure 3 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be noted that any AI models, software tools, or components not belonging to our company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. The collection, storage, use, processing, transmission, provision, and disclosure of information, data, and signals involved in the embodiments of this application all comply with relevant laws and regulations and do not violate public order and good morals.

[0014] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0015] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0017] like Figure 1 As shown, Figure 1 This is a flowchart illustrating an interaction-driven intelligent agent skill generation and adaptive evolution method according to an embodiment of the present invention. The interaction-driven intelligent agent skill generation and adaptive evolution method provided in this embodiment is applied to a human-computer interaction system and executed by an interaction-driven intelligent agent skill generation and adaptive evolution device corresponding to this method within the human-computer interaction system. The human-computer interaction system is configured with a skill base and a knowledge base, used to collaboratively complete tasks based on user interaction content, the skill base, and the knowledge base, ultimately outputting the task execution result to the user. The method includes the following steps S11-S15.

[0018] S11. In response to the task execution instruction, acquire the interaction data during the interaction between the user and the human-computer interaction system, generate a behavior log set based on the interaction data, and store it in a preset log database.

[0019] In this embodiment, in response to a task execution instruction—an instruction issued by the user containing specific task content—the human-computer interaction system begins to collaboratively execute the task corresponding to the instruction using a skill base and a knowledge base. During task execution, the system dynamically adjusts the calls to the skill base and knowledge base based on subsequent interactions with the user until the task corresponding to the instruction is completed. Simultaneously, interaction data between the user and the human-computer interaction system is acquired, specifically, the interaction data between the user and the system from the triggering of the task execution instruction until the completion of the task corresponding to that instruction. This interaction data includes at least one of the following: a sequence of user actions, records of user adoption or correction of system suggestions, explicit feedback information input by the user, the context of task execution, and the final task execution result. Based on the interaction data, a behavior log set corresponding to the task execution instruction is generated. Preferably, the interaction data is continuously monitored, and an event-driven recording mechanism is used to generate the logs, thereby forming a complete behavior log set reflecting the user's interaction trajectory. Specifically, the system captures user interaction events seamlessly in the background for each action, automatically generating a structured behavior log for each event. This log includes a timestamp, operation type, target, and content. All generated behavior logs are chronologically aggregated to form a complete behavior log set reflecting the user's interaction trajectory. Furthermore, the process of generating this log set is completely transparent to the user and does not interfere with the normal task processing workflow of the human-computer interaction system. Finally, the behavior log set is stored in a pre-defined log database for efficient subsequent querying and analysis.

[0020] S12. If it is determined that the preset evolution triggering conditions are met, then the target operation sequence set is obtained based on the log database.

[0021] In this embodiment, it is determined whether a preset evolution trigger condition is met. When the evolution trigger condition is met, pattern mining is performed on the log database to analyze effective interaction patterns, thereby obtaining a target operation sequence set, and subsequently, skill generation is performed based on the target operation sequence set. Furthermore, the evolution trigger condition can be set as a real-time trigger condition, such as detecting the completion of a task corresponding to the task execution instruction (i.e., a single task completion), so that pattern mining is performed immediately after each task is completed. The evolution trigger condition can also be set as a periodic trigger condition, such as detecting that the current system time has reached a preset periodic time point, or detecting that the cumulative number of tasks has reached a preset threshold, so that pattern mining is performed periodically. The evolution trigger condition can also be set as a manual trigger condition, such as detecting that a user can manually trigger an instruction generated by a corresponding button through the front-end interface.

[0022] In one embodiment, obtaining the target operation sequence set based on the log database includes: Retrieve multiple sets of behavior logs from the log database; For each of the multiple behavior log sets, multiple corresponding operation sequences are generated based on the behavior log set, and the multiple operation sequences are combined into an operation sequence set corresponding to the behavior log set. Merge the operation sequence sets corresponding to each log set to obtain the total operation sequence set; Calculate the support and sequence length of each operation sequence in the total operation sequence set, and determine the operation sequences with a support greater than or equal to a preset minimum support and a sequence length less than or equal to a preset maximum sequence length as target operation sequences; The determined target operation sequences are combined into the target operation sequence set.

[0023] In this embodiment, when the type of the evolution trigger condition indicates batch processing mode, multiple behavior log sets need to be obtained from the log database for pattern mining. The SPADE (Sequential Pattern Discovery using Equivalence classes) algorithm can be used for pattern mining. Specifically, firstly, multiple behavior log sets with the latest storage time are obtained from the log database. The number of behavior logs in each behavior log set can be limited according to specific application requirements. Then, for each behavior log set, multiple corresponding operation sequences are extracted from the behavior log set, and these multiple operation sequences are combined into an operation sequence set corresponding to that behavior log set. Each operation sequence consists of a series of operations with a temporal relationship, indicating a continuous user operation behavior. Next, the operation sequence sets corresponding to each behavior log set in the multiple behavior log sets are combined into a total operation sequence set. The support and sequence length of each operation sequence in the total operation sequence set are calculated. The support of an operation sequence refers to the number of operation sequence sets containing that operation sequence in the operation sequence sets corresponding to the multiple behavior log sets; the sequence length of an operation sequence refers to the number of operations contained in that operation sequence. Finally, operation sequences with support greater than or equal to the preset minimum support and sequence length less than or equal to the preset maximum sequence length are selected as target operation sequences, and all selected target operation sequences are combined into a target operation sequence set.

[0024] In another embodiment, obtaining the target operation sequence set based on the log database includes: The latest stored set of behavior logs is obtained from the log database, and a complete operation sequence is generated in chronological order based on the timestamp of each behavior log in the set. If the sequence length of the complete operation sequence is determined to be less than or equal to a preset length, then the complete operation sequence is determined as the target operation sequence. If it is determined that the sequence length of the complete operation sequence is greater than the preset length, then the complete operation sequence is divided into multiple sub-operation sequences, and each sub-operation sequence is determined as the corresponding target operation sequence. The target operation sequence set is constructed based on the determined target operation sequence.

[0025] In this embodiment, when the type of the evolution trigger condition indicates a single processing mode, it is necessary to retrieve the latest stored behavior log set from the log database for pattern mining. This ensures that pattern mining is performed on the behavior log set of the task immediately after the task corresponding to the current task execution instruction is completed. Specifically, first, the latest stored behavior log set is retrieved from the log database, i.e., the behavior log set with the most recent storage time. Then, based on the timestamp of each behavior log in the behavior log set, a complete operation sequence is extracted in chronological order. Next, the number of operations contained in the complete operation sequence is calculated to obtain the sequence length of the complete operation sequence. Finally, if it is determined that the sequence length of the complete operation sequence is less than or equal to a preset length, the complete operation sequence is identified as the target operation sequence to construct a target operation sequence set containing only the target operation sequence. If it is determined that the sequence length of the complete operation sequence is greater than the preset length, a preset sliding window can be used to divide the complete operation sequence into multiple sub-operation sequences, and each sub-operation sequence is identified as the corresponding target operation sequence to construct a target operation sequence set containing multiple target operation sequences.

[0026] S13. Based on the target operation sequence set and the pre-trained skill generation model, obtain the new skill corresponding to each target operation sequence in the target operation sequence set.

[0027] In this embodiment, a pre-trained skill generation model is used to generate skills for a set of target operation sequences, thereby obtaining new skills corresponding to each target operation sequence in the set. This transforms the mined effective interaction patterns into structured, executable, and reusable skills. A generative large language model (such as LLaMA or the GPT-J series) is used as the base model, and the base model is fine-tuned using a training dataset to enable it to generate structured skills, thus obtaining the pre-trained skill generation model.

[0028] More specifically, step S13 includes: for each target operation sequence in the target operation sequence set, inputting the target operation sequence and a preset structured representation into the skill generation model to obtain a structured description corresponding to each module (i.e., functional unit) in the structured representation; and combining the structured descriptions corresponding to each module into a new skill corresponding to the target operation sequence. A unified structured representation is used to guide the skill generation model in generating a new skill corresponding to each target operation sequence, resulting in a modular structure for the generated skills.

[0029] S14. Based on the new skill corresponding to each target operation sequence, the skill library in the human-computer interaction system is merged and updated to obtain the updated skill library.

[0030] In this embodiment, the skill library in the human-computer interaction system is merged and updated based on the generated new skills to obtain an updated skill library. The updated skill library is then used to update the skill library in the human-computer interaction system, realizing incremental updates and redundancy removal evolution of the skill library. This achieves the optimization and iterative updates of skills in the human-computer interaction system, improving the current situation where traditional system capabilities are fixed and cannot continuously evolve.

[0031] In one embodiment, step S14 includes: For each new skill corresponding to a target operation sequence, a set of candidate skills that are similar to the new skill is obtained from the skill library; For each candidate skill in the candidate skill set, the structural difference degree between the new skill and the candidate skill is calculated to obtain a structural difference degree set; If it is determined that there are structural differences in the set of structural differences that are less than a preset difference threshold, then the structural differences in the set of structural differences that are less than the difference threshold are grouped into a structural difference subset. Based on the new skill and the structural difference subset, the skill library is merged and updated to obtain the updated skill library. If it is determined that there is no structural difference less than the difference threshold in the set of structural differences, then the new skill is added to the skill library to obtain the updated skill library.

[0032] In this embodiment, for each new skill corresponding to a target operation sequence, firstly, the similarity between the new skill and each existing skill in the skill library is calculated. Existing skills with a similarity greater than or equal to a preset global similarity threshold are identified as candidate skills, and these candidate skills are grouped into a candidate skill set. Preferably, calculating the similarity between the new skill and each existing skill in the skill library specifically includes: extracting the global text content of the new skill, i.e., extracting the text content of each module in the new skill and concatenating it into a complete global text content; inputting the global text content of the new skill into a semantic encoding model to generate a global semantic vector for the new skill; for each existing skill in the skill library, extracting the global text content of the existing skill; inputting the global text content of the existing skill into a semantic encoding model to generate a global semantic vector for the existing skill; calculating the cosine similarity between the global semantic vector of the new skill and the global semantic vector of the existing skill; and using this cosine similarity as the similarity between the new skill and the existing skill.

[0033] Then, for each candidate skill in the candidate skill set, the structural difference degree between the new skill and the candidate skills is calculated. The structural difference degree is used to characterize the degree of difference between the new skill and the candidate skills in terms of module composition, parameter configuration, or inference path. The structural difference degrees corresponding to each candidate skill are combined into a structural difference degree set, and based on the structural difference degree set and a preset difference threshold, it is determined whether the new skill meets the skill merging condition.

[0034] Finally, if it is determined that there are structural differences in the set of structural differences less than the preset difference threshold, it indicates that the new skill meets the skill merging conditions and there are candidate skills that are structurally highly similar to the new skill. In this case, structural differences less than the difference threshold are selected from the set of structural differences, and these selected structural differences are grouped into a subset. The skill library is then merged and updated based on the new skill and the subset of structural differences, resulting in an updated skill library. This prevents skill redundancy and enables incremental evolution. Conversely, if it is determined that there are no structural differences less than the difference threshold in the set of structural differences, it indicates that the new skill does not meet the skill merging conditions and there are no candidate skills that are structurally highly similar to the new skill. In this case, a skill ID, version number, and creation time are generated, and the new skill, skill ID, version number, and creation time are grouped into a new entry and added to the skill library, resulting in an updated skill library.

[0035] In one embodiment, the step of calculating the structural difference degree between the new skill and the candidate skills for each candidate skill in the candidate skill set to obtain a structural difference degree set includes: For each candidate skill in the candidate skill set, the module structure of the new skill is aligned with that of the candidate skill to obtain the alignment result; Based on the alignment results, the degree of difference between the new skill and the candidate skill is calculated to obtain the structural difference degree corresponding to the candidate skill; The structural difference degree corresponding to each candidate skill is composed of the structural difference degree set.

[0036] In this embodiment, for each candidate skill in the candidate skill set, the module structure of the new skill and the candidate skills is aligned to obtain an alignment result. The module structure includes, but is not limited to, functional modules, semantic modules, and reasoning modules. The alignment result indicates the module-level correspondence between the new skill and the candidate skills. Based on the alignment result, the degree of difference between the new skill and the candidate skills is calculated to obtain the structural difference degree corresponding to the candidate skills. Specifically, multiple modules and their respective module type identifiers are extracted from both the new skill and the candidate skill. Based on these identifiers, each module in the new skill is paired one-to-one with modules in the candidate skill that share the same module type identifier (e.g., pairing the "workflow" module in the new skill with the "workflow" module in the candidate skill as a matching module pairing group). This results in multiple matching groups, which are then used as the alignment results. For each matching group, the text content of the two modules within the pairing group is extracted and input into a semantic encoding model to obtain the semantic vectors corresponding to the two modules. The cosine similarity between these semantic vectors is calculated and used as the similarity of the matching group. The difference between the matching groups is then determined based on this similarity, where the difference is the difference between 1 and the overall similarity. Finally, the difference of each matching group in the alignment results is weighted and summed to obtain the structural difference corresponding to the candidate skill, which can be understood as Δ = Σ(w i ×D i ), where Δ represents the structural dissimilarity corresponding to the candidate skill, D i w represents the difference between the pairings of the i-th module in the alignment results. i The preset weights are assigned to the pairing groups corresponding to the i-th module. Finally, the structural dissimilarity scores corresponding to each candidate skill in the candidate skill set are combined into a structural dissimilarity score set.

[0037] In one embodiment, the step of merging and updating the skill library based on the new skill and the structural difference subset to obtain the updated skill library includes: For each structural difference degree in the subset of structural difference degrees, the candidate skill corresponding to the structural difference degree is taken as the target candidate skill, and the conflicting modules in the new skill and the target candidate skill are identified based on the structural difference degree. The conflicting modules in the new skill and the target candidate skill are processed using a preset conflict resolution strategy to obtain a merged skill; The merged skills are used to replace the skills in the skill library that correspond to the target candidate skills, so as to obtain the updated skill library.

[0038] In this embodiment, each candidate skill corresponding to a structural difference degree in the structural difference degree subset is structurally highly similar to the new skill. For each structural difference degree in the structural difference degree subset, the following processing flow is executed: First, the candidate skill corresponding to the structural difference degree is taken as the target candidate skill. Conflicting modules in the new skill and the target candidate skill are identified based on the structural difference degree. Conflicting modules refer to inconsistent or mutually exclusive module content contained in the new skill and the target candidate skill. Then, a preset conflict resolution strategy is used to process the conflicting modules in the new skill and the target candidate skill. After the conflict is resolved, a consistent and better merged skill is generated. The merged skill is used to replace the skill in the skill library corresponding to the target candidate skill, thereby enabling the candidate skills in the skill library corresponding to the structural difference degree to complete the version update. When all candidate skills in the skill library corresponding to each structural difference degree have completed the version update, the updated skill library is obtained.

[0039] Furthermore, the conflict resolution strategy includes at least one of the following: a module fusion strategy, a module replacement strategy, and a module branching strategy. The module fusion strategy is used to merge functionally complementary conflicting modules from the new skill and the target candidate skill to integrate complementary information between the conflicting modules. The module replacement strategy is used to select the better-performing module from the conflicting modules of the new skill and the target candidate skill and replace the other module with the better-performing module. The module branching strategy is used to retain multiple forms of conflicting modules in the new skill and the target candidate skill based on contextual conditions.

[0040] S15. Generate new knowledge based on the interaction data and the behavior log set, and use the new knowledge to update the knowledge base in the human-computer interaction system to obtain the updated knowledge base.

[0041] In this embodiment, after each task execution instruction is completed, the knowledge base of the human-computer interaction system is updated based on the new knowledge extracted from the interaction data and behavior log set. This updated knowledge base transforms the user's natural interaction behavior into reusable knowledge, enabling the effective accumulation and reuse of user knowledge. It records the user's valuable experience in the task processing process, and the new knowledge is incrementally updated into the knowledge base for use by the human-computer interaction system when completing corresponding tasks. Furthermore, the accumulated new knowledge can enhance the accuracy of the task execution results given by the human-computer interaction system.

[0042] In one embodiment, generating new knowledge based on the interaction data and the behavior log set includes: Obtain the final task execution result from the interaction data; The new knowledge is generated based at least on the final task execution result and the behavior log set.

[0043] In this embodiment, new knowledge is generated at least based on the final task execution results and behavior log sets in the interaction data. Additionally, new knowledge can also be extracted from the final task execution results and behavior log sets, as well as user adoption or correction records of system suggestions, explicit user feedback, and the contextual state of task execution in the interaction data.

[0044] The interactive-driven intelligent agent skill generation and adaptive evolution method provided in this invention acquires interaction data during the interaction between the user and the human-computer interaction system, generates a behavior log set based on the interaction data, and stores it in a log database. If the evolution triggering condition is met, a target operation sequence set is obtained based on the log database. Based on the target operation sequence set and a pre-trained skill generation model, a new skill corresponding to each target operation sequence is obtained. Based on the new skills corresponding to each target operation sequence, the skill library in the human-computer interaction system is merged and updated to obtain an updated skill library. New knowledge is generated based on the interaction data and behavior log set, and the knowledge base in the human-computer interaction system is updated using the new knowledge to obtain an updated knowledge base. This invention can automatically capture, learn, and update skills and knowledge during the natural human-computer interaction process of users completing tasks. It can capture and structure the valuable experience of users in the task processing process, enabling user knowledge to be effectively accumulated and reused, improving the system's continuous evolution capability, enhancing the system's adaptability, reasoning ability, and cross-scenario generalization ability in complex tasks, and realizing a closed-loop collaboration with bidirectional enhancement of the human-computer collaboration process.

[0045] This invention also provides an interaction-driven agent skill generation and adaptive evolution apparatus. This apparatus is used to execute any embodiment of the aforementioned interaction-driven agent skill generation and adaptive evolution method. Specifically, please refer to... Figure 2 , Figure 2 This is a schematic block diagram of an interactive-driven intelligent agent skill generation and adaptive evolution device provided in an embodiment of the present invention. The interactive-driven intelligent agent skill generation and adaptive evolution device provided in this embodiment of the present invention includes: a behavior recording unit 11, a behavior mining unit 12, a skill generation unit 13, a skill management unit 14, and a knowledge updating unit 15.

[0046] The behavior recording unit 11 is used to respond to the task execution instruction, acquire the interaction data during the interaction between the user and the human-computer interaction system, generate a behavior log set based on the interaction data, and store it in a preset log database.

[0047] In this embodiment, in response to a task execution instruction—an instruction issued by the user containing specific task content—the human-computer interaction system begins to collaboratively execute the task corresponding to the instruction using a skill base and a knowledge base. During task execution, the system dynamically adjusts the calls to the skill base and knowledge base based on subsequent interactions with the user until the task corresponding to the instruction is completed. Simultaneously, interaction data between the user and the human-computer interaction system is acquired, specifically, the interaction data between the user and the system from the triggering of the task execution instruction until the completion of the task corresponding to that instruction. This interaction data includes at least one of the following: a sequence of user actions, records of user adoption or correction of system suggestions, explicit feedback information input by the user, the context of task execution, and the final task execution result. Based on the interaction data, a behavior log set corresponding to the task execution instruction is generated. Preferably, the interaction data is continuously monitored, and an event-driven recording mechanism is used to generate the logs, thereby forming a complete behavior log set reflecting the user's interaction trajectory. Specifically, the system captures user interaction events seamlessly in the background for each action, automatically generating a structured behavior log for each event. This log includes a timestamp, operation type, target, and content. All generated behavior logs are chronologically aggregated to form a complete behavior log set reflecting the user's interaction trajectory. Furthermore, the process of generating this log set is completely transparent to the user and does not interfere with the normal task processing workflow of the human-computer interaction system. Finally, the behavior log set is stored in a pre-defined log database for efficient subsequent querying and analysis.

[0048] The behavior mining unit 12 is used to obtain a target operation sequence set based on the log database if it is determined that the preset evolution triggering conditions are met.

[0049] In this embodiment, it is determined whether a preset evolution trigger condition is met. When the evolution trigger condition is met, pattern mining is performed on the log database to analyze effective interaction patterns, thereby obtaining a target operation sequence set, and subsequently, skill generation is performed based on the target operation sequence set. Furthermore, the evolution trigger condition can be set as a real-time trigger condition, such as detecting the completion of a task corresponding to the task execution instruction (i.e., a single task completion), so that pattern mining is performed immediately after each task is completed. The evolution trigger condition can also be set as a periodic trigger condition, such as detecting that the current system time has reached a preset periodic time point, or detecting that the cumulative number of tasks has reached a preset threshold, so that pattern mining is performed periodically. The evolution trigger condition can also be set as a manual trigger condition, such as detecting that a user can manually trigger an instruction generated by a corresponding button through the front-end interface.

[0050] In one embodiment, when the behavior mining unit 12 performs the step of obtaining the target operation sequence set based on the log database, it is specifically used for: Retrieve multiple sets of behavior logs from the log database; For each of the multiple behavior log sets, multiple corresponding operation sequences are generated based on the behavior log set, and the multiple operation sequences are combined into an operation sequence set corresponding to the behavior log set. Merge the operation sequence sets corresponding to each log set to obtain the total operation sequence set; Calculate the support and sequence length of each operation sequence in the total operation sequence set, and determine the operation sequences with a support greater than or equal to a preset minimum support and a sequence length less than or equal to a preset maximum sequence length as target operation sequences; The determined target operation sequences are combined into the target operation sequence set.

[0051] In this embodiment, when the type of the evolution trigger condition indicates batch processing mode, multiple behavior log sets need to be obtained from the log database for pattern mining. The SPADE (Sequential Pattern Discovery using Equivalence classes) algorithm can be used for pattern mining. Specifically, firstly, multiple behavior log sets with the latest storage time are obtained from the log database. The number of behavior logs in each behavior log set can be limited according to specific application requirements. Then, for each behavior log set, multiple corresponding operation sequences are extracted from the behavior log set, and these multiple operation sequences are combined into an operation sequence set corresponding to that behavior log set. Each operation sequence consists of a series of operations with a temporal relationship, indicating a continuous user operation behavior. Next, the operation sequence sets corresponding to each behavior log set in the multiple behavior log sets are combined into a total operation sequence set. The support and sequence length of each operation sequence in the total operation sequence set are calculated. The support of an operation sequence refers to the number of operation sequence sets containing that operation sequence in the operation sequence sets corresponding to the multiple behavior log sets; the sequence length of an operation sequence refers to the number of operations contained in that operation sequence. Finally, operation sequences with support greater than or equal to the preset minimum support and sequence length less than or equal to the preset maximum sequence length are selected as target operation sequences, and all selected target operation sequences are combined into a target operation sequence set.

[0052] In another embodiment, when performing the step of obtaining the target operation sequence set based on the log database, the behavior mining unit 12 is specifically used for: The latest stored set of behavior logs is obtained from the log database, and a complete operation sequence is generated in chronological order based on the timestamp of each behavior log in the set. If the sequence length of the complete operation sequence is determined to be less than or equal to a preset length, then the complete operation sequence is determined as the target operation sequence. If it is determined that the sequence length of the complete operation sequence is greater than the preset length, then the complete operation sequence is divided into multiple sub-operation sequences, and each sub-operation sequence is determined as the corresponding target operation sequence. The target operation sequence set is constructed based on the determined target operation sequence.

[0053] In this embodiment, when the type of the evolution trigger condition indicates a single processing mode, it is necessary to retrieve the latest stored behavior log set from the log database for pattern mining. This ensures that pattern mining is performed on the behavior log set of the task immediately after the task corresponding to the current task execution instruction is completed. Specifically, first, the latest stored behavior log set is retrieved from the log database, i.e., the behavior log set with the most recent storage time. Then, based on the timestamp of each behavior log in the behavior log set, a complete operation sequence is extracted in chronological order. Next, the number of operations contained in the complete operation sequence is calculated to obtain the sequence length of the complete operation sequence. Finally, if it is determined that the sequence length of the complete operation sequence is less than or equal to a preset length, the complete operation sequence is identified as the target operation sequence to construct a target operation sequence set containing only the target operation sequence. If it is determined that the sequence length of the complete operation sequence is greater than the preset length, a preset sliding window can be used to divide the complete operation sequence into multiple sub-operation sequences, and each sub-operation sequence is identified as the corresponding target operation sequence to construct a target operation sequence set containing multiple target operation sequences.

[0054] The skill generation unit 13 is used to obtain a new skill corresponding to each target operation sequence in the target operation sequence set based on the target operation sequence set and the pre-trained skill generation model.

[0055] In this embodiment, a pre-trained skill generation model is used to generate skills for a set of target operation sequences, thereby obtaining new skills corresponding to each target operation sequence in the set. This transforms the mined effective interaction patterns into structured, executable, and reusable skills. A generative large language model (such as LLaMA or the GPT-J series) is used as the base model, and the base model is fine-tuned using a training dataset to enable it to generate structured skills, thus obtaining the pre-trained skill generation model.

[0056] More specifically, the skill generation unit 13 is specifically used for: for each target operation sequence in the target operation sequence set, inputting the target operation sequence and a preset structured representation into the skill generation model to obtain a structured description corresponding to each module (i.e., functional unit) in the structured representation; and combining the structured descriptions corresponding to each module into a new skill corresponding to the target operation sequence. A unified structured representation is used to guide the skill generation model in generating a new skill corresponding to each target operation sequence, making the generated skills present a modular structure.

[0057] The skill management unit 14 is used to merge and update the skill library in the human-computer interaction system based on the new skill corresponding to each target operation sequence, so as to obtain the updated skill library.

[0058] In this embodiment, the skill library in the human-computer interaction system is merged and updated based on the generated new skills to obtain an updated skill library. The updated skill library is then used to update the skill library in the human-computer interaction system, realizing incremental updates and redundancy removal evolution of the skill library. This achieves the optimization and iterative updates of skills in the human-computer interaction system, improving the current situation where traditional system capabilities are fixed and cannot continuously evolve.

[0059] In one embodiment, the skill management unit 14 is specifically used for: For each new skill corresponding to a target operation sequence, a set of candidate skills that are similar to the new skill is obtained from the skill library; For each candidate skill in the candidate skill set, the structural difference degree between the new skill and the candidate skill is calculated to obtain a structural difference degree set; If it is determined that there are structural differences in the set of structural differences that are less than a preset difference threshold, then the structural differences in the set of structural differences that are less than the difference threshold are grouped into a structural difference subset. Based on the new skill and the structural difference subset, the skill library is merged and updated to obtain the updated skill library. If it is determined that there is no structural difference less than the difference threshold in the set of structural differences, then the new skill is added to the skill library to obtain the updated skill library.

[0060] In this embodiment, for each new skill corresponding to a target operation sequence, firstly, the similarity between the new skill and each existing skill in the skill library is calculated. Existing skills with a similarity greater than or equal to a preset global similarity threshold are identified as candidate skills, and these candidate skills are grouped into a candidate skill set. Preferably, calculating the similarity between the new skill and each existing skill in the skill library specifically includes: extracting the global text content of the new skill, i.e., extracting the text content of each module in the new skill and concatenating it into a complete global text content; inputting the global text content of the new skill into a semantic encoding model to generate a global semantic vector for the new skill; for each existing skill in the skill library, extracting the global text content of the existing skill; inputting the global text content of the existing skill into a semantic encoding model to generate a global semantic vector for the existing skill; calculating the cosine similarity between the global semantic vector of the new skill and the global semantic vector of the existing skill; and using this cosine similarity as the similarity between the new skill and the existing skill.

[0061] Then, for each candidate skill in the candidate skill set, the structural difference degree between the new skill and the candidate skills is calculated. The structural difference degree is used to characterize the degree of difference between the new skill and the candidate skills in terms of module composition, parameter configuration, or inference path. The structural difference degrees corresponding to each candidate skill are combined into a structural difference degree set, and based on the structural difference degree set and a preset difference threshold, it is determined whether the new skill meets the skill merging condition.

[0062] Finally, if it is determined that there are structural differences in the set of structural differences less than the preset difference threshold, it indicates that the new skill meets the skill merging conditions and there are candidate skills that are structurally highly similar to the new skill. In this case, structural differences less than the difference threshold are selected from the set of structural differences, and these selected structural differences are grouped into a subset. The skill library is then merged and updated based on the new skill and the subset of structural differences, resulting in an updated skill library. This prevents skill redundancy and enables incremental evolution. Conversely, if it is determined that there are no structural differences less than the difference threshold in the set of structural differences, it indicates that the new skill does not meet the skill merging conditions and there are no candidate skills that are structurally highly similar to the new skill. In this case, a skill ID, version number, and creation time are generated, and the new skill, skill ID, version number, and creation time are grouped into a new entry and added to the skill library, resulting in an updated skill library.

[0063] In one embodiment, when the skill management unit 14 performs the step of calculating the structural difference degree between the new skill and the candidate skills for each candidate skill in the candidate skill set to obtain a structural difference degree set, it is specifically used to: For each candidate skill in the candidate skill set, the module structure of the new skill is aligned with that of the candidate skill to obtain the alignment result; Based on the alignment results, the degree of difference between the new skill and the candidate skill is calculated to obtain the structural difference degree corresponding to the candidate skill; The structural difference degree corresponding to each candidate skill is composed of the structural difference degree set.

[0064] In this embodiment, for each candidate skill in the candidate skill set, the module structure of the new skill and the candidate skills is aligned to obtain an alignment result. The module structure includes, but is not limited to, functional modules, semantic modules, and reasoning modules. The alignment result indicates the module-level correspondence between the new skill and the candidate skills. Based on the alignment result, the degree of difference between the new skill and the candidate skills is calculated, i.e., the alignment result is quantified to obtain the structural difference degree corresponding to the candidate skills. Specifically, multiple modules and their respective module type identifiers are extracted from both the new skill and the candidate skill. Based on these identifiers, each module in the new skill is paired one-to-one with modules in the candidate skill that share the same module type identifier (e.g., pairing the "workflow" module in the new skill with the "workflow" module in the candidate skill as a matching module pairing group). This results in multiple matching groups, which are then used as the alignment results. For each matching group, the text content of the two modules within the pairing group is extracted and input into a semantic encoding model to obtain the semantic vectors corresponding to the two modules. The cosine similarity between these semantic vectors is calculated and used as the similarity of the matching group. The difference between the matching groups is then determined based on this similarity, where the difference is the difference between 1 and the overall similarity. Finally, the difference of each matching group in the alignment results is weighted and summed to obtain the structural difference corresponding to the candidate skill, which can be understood as Δ = Σ(w i ×D i ), where Δ represents the structural dissimilarity corresponding to the candidate skill, D i w represents the difference between the pairings of the i-th module in the alignment results. i The preset weights are assigned to the pairing groups corresponding to the i-th module. Finally, the structural dissimilarity scores corresponding to each candidate skill in the candidate skill set are combined into a structural dissimilarity score set.

[0065] In one embodiment, when the skill management unit 14 performs the step of merging and updating the skill library based on the new skill and the structural difference subset to obtain the updated skill library, it is specifically used for: For each structural difference degree in the subset of structural difference degrees, the candidate skill corresponding to the structural difference degree is taken as the target candidate skill, and the conflicting modules in the new skill and the target candidate skill are identified based on the structural difference degree. The conflicting modules in the new skill and the target candidate skill are processed using a preset conflict resolution strategy to obtain a merged skill; The merged skills are used to replace the skills in the skill library that correspond to the target candidate skills, so as to obtain the updated skill library.

[0066] In this embodiment, each candidate skill corresponding to a structural difference degree in the structural difference degree subset is structurally highly similar to the new skill. For each structural difference degree in the structural difference degree subset, the following processing flow is executed: First, the candidate skill corresponding to the structural difference degree is taken as the target candidate skill. Conflicting modules in the new skill and the target candidate skill are identified based on the structural difference degree. Conflicting modules refer to inconsistent or mutually exclusive module content contained in the new skill and the target candidate skill. Then, a preset conflict resolution strategy is used to process the conflicting modules in the new skill and the target candidate skill. After the conflict is resolved, a consistent and better merged skill is generated. The merged skill is used to replace the skill in the skill library corresponding to the target candidate skill, thereby enabling the candidate skills in the skill library corresponding to the structural difference degree to complete the version update. When all candidate skills in the skill library corresponding to each structural difference degree have completed the version update, the updated skill library is obtained.

[0067] Furthermore, the conflict resolution strategy includes at least one of the following: a module fusion strategy, a module replacement strategy, and a module branching strategy. The module fusion strategy is used to merge functionally complementary conflicting modules from the new skill and the target candidate skill to integrate complementary information between the conflicting modules. The module replacement strategy is used to select the better-performing module from the conflicting modules of the new skill and the target candidate skill and replace the other module with the better-performing module. The module branching strategy is used to retain multiple forms of conflicting modules in the new skill and the target candidate skill based on contextual conditions.

[0068] The knowledge update unit 15 is used to generate new knowledge based on the interaction data and the behavior log set, and to update the knowledge base of the human-computer interaction system using the new knowledge to obtain the updated knowledge base.

[0069] In this embodiment, after each task execution instruction is completed, the knowledge base of the human-computer interaction system is updated based on the new knowledge extracted from the interaction data and behavior log set. This updated knowledge base transforms the user's natural interaction behavior into reusable knowledge, enabling the effective accumulation and reuse of user knowledge. It records the user's valuable experience in the task processing process, and the new knowledge is incrementally updated into the knowledge base for use by the human-computer interaction system when completing corresponding tasks. Furthermore, the accumulated new knowledge can enhance the accuracy of the task execution results given by the human-computer interaction system.

[0070] In one embodiment, when performing the step of generating new knowledge based on the interaction data and the behavior log set, the knowledge update unit 15 is specifically used for: Obtain the final task execution result from the interaction data; The new knowledge is generated based at least on the final task execution result and the behavior log set.

[0071] In this embodiment, new knowledge is generated at least based on the final task execution results and behavior log sets in the interaction data. Additionally, new knowledge can also be extracted from the final task execution results and behavior log sets, as well as user adoption or correction records of system suggestions, explicit user feedback, and the contextual state of task execution in the interaction data.

[0072] The present invention discloses an interaction-driven intelligent agent skill generation and adaptive evolution device, which is used to execute any embodiment of the aforementioned interaction-driven intelligent agent skill generation and adaptive evolution method. It acquires interaction data during the interaction between the user and the human-computer interaction system, generates a behavior log set based on the interaction data, and stores it in a log database. If the evolution triggering condition is determined to be met, it acquires a target operation sequence set based on the log database. Based on the target operation sequence set and a pre-trained skill generation model, it obtains a new skill corresponding to each target operation sequence. Based on the new skill corresponding to each target operation sequence, it merges and updates the skill library in the human-computer interaction system to obtain an updated skill library. Finally, it generates new knowledge based on the interaction data and the behavior log set, and uses the new knowledge to update the knowledge base in the human-computer interaction system to obtain an updated knowledge base. This invention can automatically capture, learn, and update skills and knowledge during the natural human-computer interaction process of users completing tasks. It can capture and structure users' valuable experience in the task processing process, so that user knowledge can be effectively accumulated and reused, improve the system's continuous evolution capability, enhance the system's adaptability, reasoning ability and cross-scenario generalization ability in complex tasks, and realize a closed-loop collaboration with bidirectional enhancement of human-computer collaboration process.

[0073] The aforementioned interaction-driven agent skill generation and adaptive evolution method can be implemented as a computer program, which can be used in, for example... Figure 3 It runs on the computer device shown.

[0074] Please see Figure 3 , Figure 3 This is a schematic block diagram of a computer device provided in an embodiment of the present invention. The computer device 500 includes a processor 502, a memory, and a network interface 505 connected via a device bus 501, wherein the memory may include a storage medium 503 and internal memory 504.

[0075] The storage medium 503 can store the operating device 5031 and the computer program 5032. When the computer program 5032 is executed, it enables the processor 502 to execute an interaction-driven intelligent agent skill generation and adaptive evolution method.

[0076] The processor 502 provides computing and control capabilities to support the operation of the entire computer device 500.

[0077] The internal memory 504 provides an environment for the operation of the computer program 5032 in the storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an interactive-driven intelligent agent skill generation and adaptive evolution method.

[0078] This network interface 505 is used for network communication, such as providing data transmission. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device 500 to which the present invention is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0079] The processor 502 is used to run the computer program 5032 stored in the memory to implement the interaction-driven intelligent agent skill generation and adaptive evolution method disclosed in the embodiments of the present invention.

[0080] Those skilled in the art will understand that Figure 3 The embodiments of the computer device shown do not constitute a limitation on the specific configuration of the computer device. In other embodiments, the computer device may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. For example, in some embodiments, the computer device may include only memory and a processor. In such embodiments, the structure and function of the memory and processor are different from those shown. Figure 3 The embodiments shown are consistent and will not be described again here.

[0081] It should be understood that, in this embodiment of the invention, the processor 502 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0082] In another embodiment of the invention, a computer-readable storage medium is provided. This computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, wherein when executed by a processor, the computer program implements the interaction-driven intelligent agent skill generation and adaptive evolution method disclosed in this embodiment of the invention.

[0083] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the specific working processes of the devices and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0084] In the embodiments provided by this invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Units with the same function may be grouped into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, systems, or units, or it may be an electrical, mechanical, or other form of connection.

[0085] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0086] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0087] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a backend server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks.

[0088] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for generating and adaptively evolving intelligent agent skills based on interaction-driven processes, characterized in that, The method, applied to human-computer interaction systems, includes: In response to a task execution command, the system acquires interaction data during the interaction between the user and the human-computer interaction system, generates a behavior log set based on the interaction data, and stores it in a preset log database. If the preset evolution triggering conditions are met, the target operation sequence set is obtained based on the log database; Based on the target operation sequence set and the pre-trained skill generation model, a new skill corresponding to each target operation sequence in the target operation sequence set is obtained; The skill library in the human-computer interaction system is updated by merging skills based on the new skills corresponding to each target operation sequence, resulting in an updated skill library. New knowledge is generated based on the interaction data and the behavior log set, and the knowledge base in the human-computer interaction system is updated using the new knowledge to obtain the updated knowledge base.

2. The method for generating and adaptively evolving intelligent agent skills based on interaction-driven methods according to claim 1, characterized in that, The step of obtaining the target operation sequence set based on the log database includes: Retrieve multiple sets of behavior logs from the log database; For each of the multiple behavior log sets, multiple corresponding operation sequences are generated based on the behavior log set, and the multiple operation sequences are combined into an operation sequence set corresponding to the behavior log set. Merge the operation sequence sets corresponding to each log set to obtain the total operation sequence set; Calculate the support and sequence length of each operation sequence in the total operation sequence set, and determine the operation sequences with a support greater than or equal to a preset minimum support and a sequence length less than or equal to a preset maximum sequence length as target operation sequences; The determined target operation sequences are combined into the target operation sequence set.

3. The method for generating and adaptively evolving intelligent agent skills based on interaction-driven methods according to claim 1, characterized in that, The step of obtaining the target operation sequence set based on the log database includes: The latest stored set of behavior logs is obtained from the log database, and a complete operation sequence is generated in chronological order based on the timestamp of each behavior log in the set. If the sequence length of the complete operation sequence is determined to be less than or equal to a preset length, then the complete operation sequence is determined as the target operation sequence. If it is determined that the sequence length of the complete operation sequence is greater than the preset length, then the complete operation sequence is divided into multiple sub-operation sequences, and each sub-operation sequence is determined as the corresponding target operation sequence. The target operation sequence set is constructed based on the determined target operation sequence.

4. The method for generating and adaptively evolving intelligent agent skills based on interaction-driven methods according to claim 1, characterized in that, The skill base in the human-computer interaction system is updated by merging skills based on the new skills corresponding to each target operation sequence, resulting in an updated skill base, including: For each new skill corresponding to a target operation sequence, a set of candidate skills that are similar to the new skill is obtained from the skill library; For each candidate skill in the candidate skill set, the structural difference degree between the new skill and the candidate skill is calculated to obtain a structural difference degree set; If it is determined that there are structural differences in the set of structural differences that are less than a preset difference threshold, then the structural differences in the set of structural differences that are less than the difference threshold are grouped into a structural difference subset. Based on the new skill and the structural difference subset, the skill library is merged and updated to obtain the updated skill library. If it is determined that there is no structural difference less than the difference threshold in the set of structural differences, then the new skill is added to the skill library to obtain the updated skill library.

5. The method for generating and adaptively evolving intelligent agent skills based on interaction-driven methods according to claim 4, characterized in that, For each candidate skill in the candidate skill set, the structural difference degree between the new skill and the candidate skills is calculated to obtain a structural difference degree set, including: For each candidate skill in the candidate skill set, the module structure of the new skill is aligned with that of the candidate skill to obtain the alignment result; Based on the alignment results, the degree of difference between the new skill and the candidate skill is calculated to obtain the structural difference degree corresponding to the candidate skill; The structural difference degree corresponding to each candidate skill is composed of the structural difference degree set.

6. The method for generating and adaptively evolving intelligent agent skills based on interaction-driven methods according to claim 4, characterized in that, The process of merging and updating the skill library based on the new skill and the structural difference subset to obtain the updated skill library includes: For each structural difference degree in the subset of structural difference degrees, the candidate skill corresponding to the structural difference degree is taken as the target candidate skill, and the conflicting modules in the new skill and the target candidate skill are identified based on the structural difference degree. The conflicting modules in the new skill and the target candidate skill are processed using a preset conflict resolution strategy to obtain a merged skill; The merged skills are used to replace the skills in the skill library that correspond to the target candidate skills, so as to obtain the updated skill library.

7. The method for generating and adaptively evolving intelligent agent skills based on interaction-driven methods according to claim 1, characterized in that, The generation of new knowledge based on the interaction data and the behavior log set includes: Obtain the final task execution result from the interaction data; The new knowledge is generated based at least on the final task execution result and the behavior log set.

8. A device for generating and adaptively evolving intelligent agent skills based on interaction-driven processes, characterized in that, The apparatus is used to execute the interaction-driven agent skill generation and adaptive evolution method as described in any one of claims 1-7, the apparatus comprising: The behavior recording unit is used to respond to task execution instructions, acquire interaction data during the interaction between the user and the human-computer interaction system, generate a behavior log set based on the interaction data, and store it in a preset log database; The behavior mining unit is used to obtain a target operation sequence set based on the log database if it is determined that a preset evolution triggering condition is met. The skill generation unit is used to obtain a new skill corresponding to each target operation sequence in the target operation sequence set based on the target operation sequence set and the pre-trained skill generation model; The skill management unit is used to merge and update the skill library in the human-computer interaction system based on the new skill corresponding to each target operation sequence, so as to obtain the updated skill library. The knowledge update unit is used to generate new knowledge based on the interaction data and the behavior log set, and to update the knowledge base of the human-computer interaction system using the new knowledge to obtain the updated knowledge base.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the interaction-driven intelligent agent skill generation and adaptive evolution method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the interaction-driven intelligent agent skill generation and adaptive evolution method as described in any one of claims 1 to 7.

Images