Intelligent agent evolution method, equipment and storage medium

Abstract

The invention discloses an agent evolution method and device and a storage medium, and the method comprises the steps: monitoring the group feature information of a service group corresponding to an agent, responding to the condition that a monitoring index determined by the group feature information meets an adjustment triggering condition, and determining an agent adjustment strategy matched with the monitoring index; and based on the intelligent agent adjustment strategy, performing configuration adjustment on the intelligent agent to obtain a target intelligent agent, and using the target intelligent agent to serve the group.

Description

Technical Field

[0001] This application relates to the field of intelligent agent technology, and in particular to an intelligent agent evolution method, device and storage medium. Background Technology

[0002] With the development of agent technology, agent-based virtual employees are gradually becoming an important trend in enterprise intelligent office. These agents participate in group collaboration as "virtual employees," providing support such as information retrieval, task management, and content generation for individuals or the entire team. However, the adaptability of current multi-agent systems in dynamically evolving group environments remains significantly insufficient. Current systems mostly rely on pre-defined initial group tasks, introducing agents with fixed roles and capability configurations through pre-build or pre-selection. This limitation, stemming from predefined persons and fixed toolchains, results in significant constraints for the agents. Summary of the Invention

[0003] In view of this, this application provides a method, apparatus and storage medium for intelligent agent evolution.

[0004] The technical solution of this application embodiment is implemented as follows:

[0005] This application provides a method for intelligent agent evolution, the method including:

[0006] The system monitors the group characteristics of the service group corresponding to the intelligent agent. In response to the monitoring indicators determined by the group characteristics meeting the adjustment trigger conditions, the system determines the intelligent agent adjustment strategy that matches the monitoring indicators.

[0007] Based on the agent adjustment strategy, the agent is configured and adjusted to obtain the target agent, and the target agent is used to serve the group.

[0008] In the above agent reconstruction method, after the target agent continues to serve the group, the method further includes: in response to the monitoring indicators determined by the group feature information meeting the configuration recovery trigger condition, recovering the adjustment configuration information corresponding to the target agent; the adjustment configuration information is a configuration based on the agent adjustment strategy corresponding to the monitoring indicators.

[0009] In the above-mentioned agent reconstruction method, the group feature information includes at least one or more of the following: topic semantic feature information or conversation interaction feature information of the conversation within the group, and background feature information of the members within the group.

[0010] In the above-mentioned agent reconstruction method, monitoring the group feature information of the service group corresponding to the agent includes: obtaining the conversation content of the conversation within the group within a preset range, and performing noise reduction processing on the conversation content to obtain the target conversation content; determining the group feature information based on the target conversation content; wherein, the noise reduction processing includes at least one of the following: filtering non-task dialogues in the conversation content; removing dialogues in the conversation content whose sentence vector offset from the target task is greater than a preset offset; and removing dialogues in the conversation content whose response frequency is lower than a preset frequency.

[0011] In the aforementioned agent reconstruction method, the monitoring indicators determined by group feature information include at least one of the following: determining the topic semantic monitoring indicators of the conversation within the group based on the parameters included in the topic semantic feature information; the topic semantic feature information includes at least one or more of topic model distribution and sentence vector offset; determining the conversation interaction monitoring indicators of the conversation within the group based on the parameters included in the conversation interaction feature information; the conversation interaction feature information includes at least one or more of the average number of turns in each dialogue in the conversation and conversation repetition rate; determining the background monitoring indicators of the members within the group based on the parameters included in the background feature information; the background feature information includes at least one or more of background vector similarity and job difference.

[0012] In the aforementioned agent reconstruction method, the monitoring indicators determined by group feature information satisfy the adjustment triggering conditions, including at least one of the following: The monitoring indicators satisfy the adjustment triggering conditions in response to parameter values ​​in the topic semantic monitoring indicators being greater than a preset topic semantic threshold; the topic semantic monitoring indicators at least include topic model distribution entropy and / or sentence vector offset; the monitoring indicators satisfy the adjustment triggering conditions in response to parameter values ​​in the conversation interaction monitoring indicators being greater than a preset conversation interaction threshold; the conversation interaction monitoring indicators at least include conversation cycle entropy and / or repetitive information ratio; the monitoring indicators satisfy the adjustment triggering conditions in response to parameter values ​​in the background monitoring indicators being greater than a preset background threshold; and the background monitoring indicators at least include background vector clustering divergence.

[0013] In the aforementioned agent reconstruction method, determining an agent adjustment strategy that matches the monitoring indicators includes at least one of the following: In response to the topic semantic monitoring indicator in the monitoring indicators meeting the adjustment trigger condition, determining the agent adjustment strategy as a second agent that focuses on the context of the conversation in the group and / or generates an agent; In response to the conversation interaction monitoring indicator in the monitoring indicators meeting the adjustment trigger condition, determining the agent adjustment strategy as adjusting the agent's role setting and / or adjusting the dialogue mode in the conversation; In response to the background monitoring indicator in the monitoring indicators meeting the adjustment trigger condition, determining the agent adjustment strategy as a third agent that adjusts the agent's role setting and / or generates an agent.

[0014] This application provides an intelligent agent evolution device, including a processor and an intelligent agent capable of running on the processor. The intelligent agent can call at least one processing model deployed in the intelligent agent evolution device to perform the following operations: monitor the group characteristic information of the service group corresponding to the intelligent agent; in response to the monitoring indicators determined by the group characteristic information satisfying the adjustment trigger condition, determine an intelligent agent adjustment strategy matching the monitoring indicators; based on the intelligent agent adjustment strategy, adjust the configuration of the intelligent agent to obtain a target intelligent agent, and use the target intelligent agent to serve the group.

[0015] In the aforementioned agent evolution device, the agent can also invoke at least one processing model deployed in the agent evolution device to perform the following operations: when the agent adjusts its strategy to generate or reclaim a configured agent, generate or reclaim a configured agent; configure the agent as a first agent or a second agent.

[0016] This application provides a computer-readable storage medium that stores one or more computer programs, which can be executed by one or more processors to implement the above-described intelligent agent evolution method.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the technical solutions provided in the embodiments of this application. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein:

[0019] Figure 1 A flowchart illustrating an intelligent agent evolution method provided in an embodiment of this application;

[0020] Figure 2 A flowchart illustrating an intelligent agent evolution method provided in an embodiment of this application;

[0021] Figure 3 A flowchart illustrating the process of determining monitoring indicators is provided for an embodiment of this application;

[0022] Figure 4 This is an exemplary flowchart illustrating whether a triggering condition is met, provided as an embodiment of this application.

[0023] Figure 5 This is an exemplary flowchart illustrating how to determine an agent's adjustment strategy, provided as an embodiment of this application.

[0024] Figure 6 A flowchart illustrating an exemplary agent evolution method provided in this application embodiment;

[0025] Figure 7 This is a schematic diagram of the structure of an intelligent agent evolution device provided in an embodiment of this application;

[0026] Figure 8 This is a schematic diagram of the structure of an intelligent agent evolution device provided in an embodiment of this application. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It is understood that the specific embodiments described herein are merely for explaining the relevant application and not for limiting the application. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application.

[0028] In the following description, references to "some embodiments" refer to a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. The terminology used in the embodiments of this application is for the purpose of describing the embodiments of this application only and is not intended to limit the application. It should also be noted that, for ease of description, only the parts relevant to the application are shown in the accompanying drawings.

[0029] With the development of agent technology, agent-based virtual employees are gradually becoming an important trend in intelligent office work for enterprises. More and more office software systems support the creation of intelligent agents through plugins or native functions. These agents participate in group collaboration as "virtual employees," providing support such as information retrieval, task management, and content generation for individuals or entire teams.

[0030] Despite the significant progress made in the aforementioned research, current multi-agent systems still exhibit considerable limitations in adaptability to dynamically evolving group environments. Specifically, in long-term multi-person conversations, demands are unpredictable, and pre-configured / selected agents within a static framework cannot adapt to the dynamically changing needs of the group. Current agents in these systems are mostly based on initial group task presets, making it difficult to effectively adapt to the specific cognitive state, interaction patterns, and role relationships of a particular group. For example, a group initially consisting of 5 people may grow to 100, leading to an overly complex and unfocused context for individual agents. Alternatively, in a cross-functional team, the persona and narrative style of a technically-oriented agent that was initially suitable may be unsuitable for newly joined business employees.

[0031] This application provides an intelligent agent evolution method, such as... Figure 1 As shown, the process includes the following steps S101 and S102:

[0032] Step S101: Monitor the group characteristic information of the service group corresponding to the intelligent agent, and in response to the monitoring indicators determined by the group characteristic information meeting the adjustment trigger conditions, determine the intelligent agent adjustment strategy that matches the monitoring indicators.

[0033] In the embodiments of this application, the intelligent agent evolution device is an electronic device with intelligent agent evolution function, which may be a tablet computer, a laptop computer, a handheld computer, a personal digital assistant (PDA), a desktop computer, etc. The specific intelligent agent evolution device is not limited here.

[0034] In the embodiments of this application, the above-described agent evolution method can be executed by an agent evolution device.

[0035] In the embodiments of this application, the group can be a group established based on certain task orientations or social relationships, such as a work team, a family group, a course discussion group, an event registration group, etc.

[0036] In the embodiments of this application, the intelligent agent evolution device monitors the group characteristic information of the service group corresponding to the intelligent agent. This information can be collected from information related to members within the group (e.g., response time of interactions between members, dialogue process, background between members), and then the collected data is processed to obtain the group characteristic information.

[0037] In some embodiments, the group feature information includes at least one or more of the following: topic semantic feature information or session interaction feature information of the conversation within the group, and background feature information of the members within the group.

[0038] In the embodiments of this application, the agent evolution device can acquire topic semantic feature information of the conversation within a group. This topic semantic feature information is used to characterize the semantic consistency of the conversation content. Semantic consistency determines whether the conversation content revolves around the same task focus. If the conversation content characterized by the parameters included in the topic semantic feature information does not revolve around the same task focus, it indicates low semantic consistency and a divergent topic. The current agent cannot meet the needs of the current group, i.e., the adjustment trigger condition is met. At this point, corresponding monitoring indicators can be determined based on the group feature information, and then a corresponding agent adjustment strategy can be determined based on the monitoring indicators. Finally, the agent is adjusted based on the agent adjustment strategy. The level of semantic consistency depends on the parameter values ​​in the topic semantic monitoring indicators determined by the topic semantic feature information, and corresponding measurement standards can be pre-set.

[0039] In the embodiments of this application, the intelligent agent evolution device can acquire conversation interaction feature information of a group session. This conversation interaction feature information characterizes the interaction efficiency within the session. Interaction efficiency determines whether communication in the session is efficient and orderly. If the parameters included in the conversation interaction feature information indicate slow task progress and frequent repetition, it indicates low interaction efficiency, meaning the current intelligent agent cannot meet the needs of the current group, thus meeting the adjustment trigger condition. In this case, corresponding monitoring indicators can be determined based on the group feature information, and then a corresponding intelligent agent adjustment strategy can be determined based on these indicators. Finally, the intelligent agent is adjusted based on the intelligent agent adjustment strategy. The level of interaction efficiency depends on the parameter values ​​in the conversation interaction monitoring indicators determined by the conversation interaction feature information, and corresponding measurement standards can be pre-set.

[0040] In the embodiments of this application, the intelligent agent evolution device can acquire background feature information of members within a group. This background feature information characterizes the consistency of backgrounds among group members. Background consistency determines the similarity of backgrounds among group members. If the parameters included in the background feature information indicate significant differences in the backgrounds of members, it indicates low background consistency, meaning the current intelligent agent cannot meet the needs of the current group, i.e., the adjustment trigger condition is met. At this point, corresponding monitoring indicators can be determined based on the group feature information, and then a corresponding intelligent agent adjustment strategy can be determined based on these indicators. Finally, the intelligent agent is adjusted based on the intelligent agent adjustment strategy. The level of background consistency depends on the parameter values ​​in the background monitoring indicators determined by the background feature information, and corresponding measurement standards can be pre-set.

[0041] For example, if the semantic consistency of the monitoring indicators is low, corresponding agent configuration information can be generated for different topics. Each configuration information indicates that the agent can play the corresponding role based on the corresponding configuration information. For example, if the semantic consistency is low and a new topic of discussing the market situation corresponding to the target task is found in the conversation, then some configuration information can be added for the conversation discussing the topic of market situation so that the agent can solve problems related to the market situation. Therefore, adding market situation configuration information to the agent so that the agent can meet the discussion of the corresponding topic is the corresponding agent adjustment strategy.

[0042] For example, if the monitoring indicators have low semantic consistency and a new topic related to the market situation corresponding to the target task is found in the conversation, then a corresponding independent agent can be set up for the new topic of the market situation corresponding to the target task. This allows the new independent agent to solve problems related to the market situation. The new independent agent's ability to satisfy the discussion of the corresponding topic is the corresponding agent adjustment strategy.

[0043] For example, if the monitoring indicators represent low background consistency, corresponding agent configuration information can be generated for members with different backgrounds. Each configuration information indicates that the agent can assume the corresponding role based on the corresponding configuration information. For example, if the background consistency is low and a financial professional is found to have been added to the conversation, then financial professional topic style configuration information can be generated for the newly added financial professional member. This allows the agent to generate a topic style consistent with the member's background. Therefore, adding financial professional topic style configuration information to the agent so that the agent can use the corresponding topic style when replying to the corresponding member is the strategy adjustment for the corresponding agent.

[0044] For example, if the monitoring indicators show low consistency in the background and a financial professional is found to have been added to the group, then an independent agent that can generate financial professional topic styles can be added for the newly added financial professional. This allows the independent agent to use the corresponding topic styles, which is the corresponding agent adjustment strategy.

[0045] For example, if the monitoring indicators find that there are many repeated statements in the conversation, the speaking frequency of the agent can be adjusted, the length of the response can be shortened, and rhythm awareness can be introduced. At this time, no new configuration information is needed; only the existing speaking configuration information needs to be adjusted.

[0046] Step S102: Based on the agent adjustment strategy, the agent configuration is adjusted to obtain the target agent, and the target agent is used to serve the group.

[0047] In the embodiments of this application, if a corresponding agent adjustment strategy is determined, the agent can be configured and adjusted based on the agent adjustment strategy to obtain the target agent.

[0048] In the embodiments of this application, the agent evolution device updates and adjusts the agent in a timely manner when the session requirements in the group change, so that the adjusted target agent adapts to the development and changes of the group state.

[0049] In this way, adjustments can be made to existing agents to change the group state, rather than being based on an unchanging initial group, thus providing greater flexibility.

[0050] In some embodiments, after performing the above step S102 "continuing to serve the group using the target agent", the agent evolution device may also perform the following steps: in response to the monitoring indicators determined by the group feature information satisfying the configuration recovery trigger condition, the adjustment configuration information corresponding to the target agent is recovered; the adjustment configuration information is a configuration based on the agent adjustment strategy corresponding to the monitoring indicators.

[0051] In the embodiments of this application, in response to the monitoring indicators determined by the group feature information satisfying the configuration recycling trigger condition, for example, if the monitoring indicators represent a high degree of consistency in the background, such as if a member in the monitoring session has deleted a financial professional, then the recycling agent can generate financial professional topic style configuration information for the member whose background has been deleted.

[0052] In the embodiments of this application, the conversation content in the group, for example, the entire process of a project, may include different stages and processes such as initiation, planning, execution, monitoring, and closing. Therefore, to adapt to the generation of each stage, the configuration information of the intelligent agent is adjusted accordingly. However, once one stage is completed, the configuration information of the intelligent agent is no longer needed in another stage and can be recycled. For example, in the first stage, group members discuss content related to project initiation. If, upon entering the next stage, the planning stage, it is found that the conversation is entirely about planning, then the configuration information of the intelligent agent previously established for project initiation can be recycled.

[0053] In the embodiments of this application, the recycling of configuration information in the intelligent agent can be set to a preset time period. For example, the operation of the intelligent agent's configuration information related to the project can be paused first, and the relevant intelligent agent's configuration information can be recycled after the preset time period.

[0054] In this way, configuration information can be updated for intelligent agents in a timely manner, and unused configuration information can be recycled, thereby improving the operating efficiency of intelligent agents.

[0055] In some embodiments, the agent evolution device, when performing step S101 above, "monitors the group characteristic information of the service group corresponding to the agent," such as... Figure 2 As shown, the following steps S201 and S202 can also be performed:

[0056] Step S201: Obtain the conversation content of the group conversation within a preset range, and perform noise reduction processing on the conversation content to obtain the target conversation content; wherein, the noise reduction processing includes at least one of the following: filtering non-task dialogues in the conversation content; removing dialogues in the conversation content whose sentence vector offset from the target task is greater than a preset offset; removing dialogues in the conversation content whose response frequency is lower than a preset frequency.

[0057] In embodiments of this application, the session content of conversations within a preset range of groups is obtained. The acquisition of conversations can be based on time thresholds and conversation type segmentation. For example, an inactivity time window (e.g., 30 minutes) can be set; if this window is exceeded, it is considered a new conversation. Alternatively, conversation content from strongly related domains can be grouped into the same context, i.e., treated as the same conversation content, etc.

[0058] In the embodiments of this application, after obtaining the conversation content of the group conversation, the conversation content can be denoising. Exemplary denoising may include: filtering non-task dialogues in the conversation content; removing dialogues in the conversation content whose sentence vector offset from the target task is greater than a preset offset; and removing dialogues in the conversation content whose response frequency is lower than a preset frequency.

[0059] For example, non-task dialogue might manifest as group members chatting, venting, using emoticons, or other non-task-related content. This non-task content can be filtered out using a noise filter (e.g., a multi-class speech classifier). Examples of non-task content include phrases like "Haha, I'm very busy today" or "I ate xxx this morning," etc.

[0060] For example, dialogues in the session whose sentence vector offset from the target task is greater than a preset offset are also removed. For instance, a preset offset can be set, and the sentence vector of the target task is A. Then, if the offset between the sentence vector of a dialogue in the session and A is greater than the preset offset, it will be removed. The preset offset can be set based on actual needs and application scenarios, and this application does not limit it.

[0061] For example, conversations with a response frequency lower than a preset frequency can also be removed. For instance, if someone asks "invoice process" multiple times but other members in the group do not respond, the conversation can be removed. The preset frequency can be 3, 4, or 2 times, etc. The preset frequency can be set based on actual needs and application scenarios, and this application does not limit it.

[0062] For example, the problem manifestations, strategy names, technical implementations, and exemplary references for noise processing are shown in Table 1:

[0063] Table 1

[0064]

[0065] As shown in Table 1, the context semantic alignment verification is performed by checking whether the sentence vectors of the dialogue in the conversation and the topic sentence vectors are within a preset offset. If they are, no removal is needed; otherwise, removal is required. The response frequency awareness gating is performed at the preset frequency. In this way, topics that are casual conversations, not required by the real context, or that receive no response can be removed, or the above-mentioned issues will not trigger adjustments by the agent, thus ensuring the quality of the agent's evolution.

[0066] Step S202: Determine group characteristic information based on the target session content.

[0067] In the embodiments of this application, the intelligent agent evolution device determines the corresponding group feature information based on the target session content after noise removal processing, thus the constructed group feature information will be more accurate.

[0068] In this way, after denoising the conversation content, the group feature information is determined, making the determined group feature information more accurate. Accordingly, the determination of whether to adjust the agent based on the group feature information will also be more accurate.

[0069] In some embodiments, the intelligent agent evolution device performs the "monitoring indicators for determining group characteristic information" in step S101 above, such as... Figure 3 As shown, the following steps S301 can also be performed:

[0070] Step S301: Based on the parameters included in the topic semantic feature information, determine the topic semantic monitoring indicators of the conversation within the group; the topic semantic feature information includes at least one or more of topic model distribution and sentence vector offset.

[0071] In embodiments of this application, topic semantic feature information may include topic model distribution or sentence vector offset. For example, the acquisition of topic model distribution may be achieved by: preprocessing the target session content: cleaning the text (removing stop words, stemming), generating a document-word matrix; topic modeling: training a model using a topic model (Latent Dirichlet Allocation, LDA) or a topic modeling method (BERTopic) to extract the topic distribution.

[0072] For example, sentence vector acquisition can be achieved by using a pre-trained model or a fine-tuned model to generate vectors of dialogue in the group's conversation content.

[0073] In the embodiments of this application, the intelligent agent evolution device can determine the corresponding topic semantic monitoring index based on the topic semantic feature information, including topic model distribution and / or sentence vector offset.

[0074] In some embodiments, the intelligent agent evolution device performs the "monitoring indicators for determining group characteristic information" in step S101 above, such as... Figure 3 As shown, the following steps S302 can also be performed:

[0075] Step S302: Based on the parameters included in the conversation interaction feature information, determine the conversation interaction monitoring indicators for the conversation within the group; the conversation interaction feature information includes at least one or more of the following: the average number of rounds for each dialogue in the conversation and the conversation repetition rate.

[0076] In embodiments of this application, the session interaction feature information includes the average number of rounds and / or session repetition rate for each dialogue in the session.

[0077] For example, the average number of rounds per dialogue can be the average number of rounds required to complete a full dialogue cycle (e.g., from question posing to resolution); the dialogue repetition rate can be the proportion of identical or highly similar information that appears repeatedly in the dialogue; for example, the similarity information in the dialogue can be calculated using a BERT-based sentence embedding model (Sentence-BERT) to calculate the similarity between sentences and mark duplicate content (e.g., similarity > 0.85). Of course, other proportions (similarity) can also be set for the proportion of similar information, such as 0.9, 0.95, etc., which can be set based on actual needs and application scenarios, and this application does not limit this.

[0078] In the embodiments of this application, the intelligent agent evolution device can determine the corresponding conversation interaction monitoring indicators based on conversation interaction feature information, including the average number of rounds of each dialogue in the conversation and / or the conversation repetition rate.

[0079] In some embodiments, the intelligent agent evolution device performs the "monitoring indicators for determining group characteristic information" in step S101 above, such as... Figure 3 As shown, the following steps S303 can also be performed:

[0080] Step S303: Based on the parameters included in the background feature information, determine the background monitoring indicators for members within the group; the background feature information includes at least one or more of background vector similarity and job difference.

[0081] In embodiments of this application, background feature information includes background vector similarity and / or job position difference.

[0082] For example, the calculation of background vector similarity can be as follows: convert text information such as member resumes and historical statements into vectors (such as BERT embedding), and use cosine similarity to measure the distance between member background vectors to ensure the similarity of background vectors. The closer the similarity is to 1, the closer the similarity is.

[0083] For example, the job variability can be calculated by mapping jobs to numerical values ​​(e.g., junior = 1, intermediate = 2, senior = 3), and then using the Gini coefficient or standard deviation to measure the unevenness of job distribution, thus obtaining the job variability.

[0084] In the embodiments of this application, the intelligent agent evolution device can determine the background monitoring indicators of members within a group based on scene feature information, including background vector similarity and / or job difference.

[0085] In embodiments of this application, the group feature information may further include feature matching feature information characterizing the degree of feature matching: tool failure rate and / or negative feedback frequency. The tool failure rate is the proportion of times a tool fails to execute a task out of the total number of calls. For example, in a software development group, if a code formatting tool is called 100 times, and 20 of those calls fail due to syntax errors, the failure rate is 20%. The negative feedback frequency is the number of times a user explicitly expresses dissatisfaction with the tool's response (e.g., clicking the "useless" button, submitting a negative review). For example, receiving 50 "irrelevant answer" negative feedbacks within one week. Thus, feature matching monitoring indicators for the group can be determined based on the tool failure rate and / or the number of negative feedbacks.

[0086] In the embodiments of this application, group feature information includes, but is not limited to, the topic semantic feature information or conversation interaction feature information mentioned above, the background feature information of members in the group, and the function matching feature information.

[0087] In this way, the group status can be detected based on the group's characteristic information to obtain the corresponding monitoring indicators.

[0088] In some embodiments, when the intelligent agent evolution device performs the above step S101, "the monitoring indicators determined by the group characteristic information meet the adjustment trigger conditions," such as Figure 4 As shown, the following steps S401 can also be performed:

[0089] Step S401: In response to the parameter value in the topic semantic monitoring indicator being greater than the preset topic semantic threshold, determine that the monitoring indicator meets the adjustment trigger condition; the topic semantic monitoring indicator includes at least the topic model distribution entropy and / or sentence vector offset.

[0090] In the embodiments of this application, the topic semantic monitoring indicators include topic model distribution entropy and / or sentence vector offset. For example, the corresponding topic model distribution entropy is determined based on the topic model distribution determined in step S301 above. For example, the entropy value of the distribution of all conversation topics or the proportion of dominant topics at the same level is calculated to obtain the corresponding topic distribution entropy.

[0091] For example, based on the sentence vectors determined in step S301 above, for example, the mean of the sentence vectors of all sentences is calculated, and then the standard deviation is determined based on the mean of the sentence vectors to obtain the standard deviation of the sentence vectors (sentence vector offset).

[0092] In the embodiments of this application, corresponding preset topic semantic thresholds (e.g., distribution entropy threshold and sentence vector offset threshold) are set for both topic model distribution entropy and sentence vector offset, or the two are fused to obtain a topic consistency parameter, and then a corresponding preset topic semantic threshold is set for the topic consistency parameter.

[0093] In the embodiments of this application, if either the topic model distribution entropy or the sentence vector offset is greater than the corresponding threshold, it can be determined that the monitoring index meets the adjustment trigger condition.

[0094] In some embodiments, when the intelligent agent evolution device performs the above step S101, "the monitoring indicators determined by the group characteristic information meet the adjustment trigger conditions," such as Figure 4 As shown, the following steps S402 can also be performed:

[0095] Step S402: In response to the fact that the parameter values ​​included in the conversation interaction monitoring indicators are greater than the preset conversation interaction threshold, determine that the monitoring indicators meet the adjustment trigger conditions; the conversation interaction monitoring indicators include at least the conversation cycle entropy and / or the proportion of repeated information.

[0096] In the embodiments of this application, the conversation interaction monitoring indicators include the conversation cycle entropy and / or the proportion of repeated information. For example, the corresponding conversation cycle entropy is determined based on the average number of rounds of each conversation determined in step S302 above. For example, the topic of each round of conversation is identified by LDA or BERTopic, and the topic switching frequency is counted.

[0097] For example, based on the session duplication rate determined in step S302 above, for example, the MinHash algorithm or the SimHash algorithm is used to detect duplicate text blocks and obtain the corresponding duplication information ratio.

[0098] In the embodiments of this application, a corresponding preset session interaction threshold is set for both the dialogue cycle entropy and / or the proportion of repeated information (e.g., preset dialogue cycle entropy threshold and repeated information proportion threshold), or the two are combined to obtain an interaction efficiency parameter, and then a corresponding preset session interaction threshold is set for the interaction efficiency parameter.

[0099] In the embodiments of this application, if either the dialogue cycle entropy or the proportion of repeated information is greater than the corresponding threshold, it can be determined that the monitoring indicator meets the adjustment trigger condition.

[0100] In some embodiments, when the intelligent agent evolution device performs the above step S101, "the monitoring indicators determined by the group characteristic information meet the adjustment trigger conditions," such as Figure 4 As shown, the following step S403 can also be performed:

[0101] Step S403: In response to the background monitoring indicators having parameter values ​​greater than the preset background threshold, determine that the monitoring indicators meet the adjustment trigger conditions; the background monitoring indicators include at least the background vector clustering divergence.

[0102] In embodiments of this application, background monitoring indicators include background vector clustering divergence. For example, based on the background vector similarity determined in step S303, keywords in the text are extracted using algorithms such as Term Frequency-Inverse Document Frequency (TF-IDF) or TextRank Algorithm (TextRank). The proportion of domain-specific words in the total vocabulary is statistically analyzed, and the coverage rate is calculated based on whether high-frequency words cover the core sub-domain. Furthermore, the dispersion of background vector distribution is analyzed using clustering algorithms (such as K-Means and DBSCAN). For example, the Silhouette Score and Calinski-Harabasz index are used as clustering evaluation indicators. The divergence formula is: Divergence = 1 - Silhouette Score; a larger value indicates a more dispersed cluster.

[0103] In the embodiments of this application, if the clustering divergence of the background vector is greater than the corresponding preset background threshold, it can be determined that the monitoring index meets the adjustment triggering condition.

[0104] In the embodiments of this application, the monitoring indicators corresponding to the functional matching feature information may be: invalid response ratio and number of questions, which will not be elaborated here.

[0105] For example, the dimensions, meanings, and characteristic information of group status monitoring, as well as the correspondence between monitoring indicators, are shown in Table 2:

[0106] Table 2

[0107]

[0108] As shown in Table 2, the group state is monitored from three dimensions: semantic consistency (corresponding to topic semantic feature information), interaction efficiency (corresponding to conversation interaction feature information), and background consistency (corresponding to background feature information). If any of these dimensions are problematic, the agent will be adjusted to adapt to the changes in the group state.

[0109] In some embodiments, when the agent evolution device performs the above step S101 of "determining the agent adjustment strategy that matches the monitoring indicators", such as Figure 5 As shown, the following steps S501 can also be performed:

[0110] Step S501: In response to the topic semantic monitoring indicators in the monitoring indicators meeting the adjustment trigger conditions, determine the agent adjustment strategy as context focusing of the conversation in the group and / or generating a second agent.

[0111] In the embodiments of this application, if the topic semantic monitoring index among the monitoring metrics meets the adjustment trigger condition, it indicates that either the topic model distribution entropy or the sentence vector offset is greater than the corresponding threshold, indicating low semantic consistency. The problem can manifest as severe topic divergence, multi-threaded parallelism, and ambiguous agent responses. For example, in cases of severe topic divergence, the direction of adjusting the agent's configuration information can be to focus on the context of the conversation within the group. If multi-threaded parallelism exists, meaning there may be multiple topics, a corresponding second agent can be set for the agent.

[0112] In the embodiments of this application, the second intelligent agent can be a sub-agent of the first intelligent agent or an intelligent agent independent of the first intelligent agent. For example, the second intelligent agent can be a sub-agent of the first intelligent agent: constructing Agent-A-Project1 / Agent-A-Project2, with a more focused context, or if two parallel themes emerge: finance and market, then the generated intelligent agents could be: Agent-Finance and Agent-Marketing. This involves implementing different avatars for different themes, or setting corresponding roles for different topics. In this way, each intelligent agent listens to the corresponding dialogue block in the group. Specifically, this can be implemented by defining context boundaries based on clustering or topic discrimination techniques to enable the corresponding intelligent agent to listen.

[0113] In some embodiments, when the agent evolution device performs the above step S101 of "determining the agent adjustment strategy that matches the monitoring indicators", such as Figure 5 As shown, the following step S502 can also be performed:

[0114] Step S502: In response to the fact that the conversation interaction monitoring indicator in the monitoring indicators meets the adjustment trigger condition, determine the agent adjustment strategy as adjusting the agent's role setting and / or adjusting the dialogue mode in the conversation.

[0115] In the embodiments of this application, if any parameter in the conversation interaction monitoring indicators, namely the dialogue cycle entropy and / or the proportion of repeated information, is greater than the corresponding threshold, it can be determined that the monitoring indicator meets the adjustment trigger condition. If the conversation interaction monitoring indicator meets the adjustment trigger condition, it indicates that the interaction efficiency is low. For example, the problem manifests as: slow task progress, excessive repeated speech, and chaotic agent interruptions. The proposed intelligent agent adjustment strategy is to adjust the intelligent agent's role setting and / or adjust the dialogue method in the conversation.

[0116] For example, if there is excessive repetition or chaotic interruptions from the agent, the agent's dialogue coordination role can be activated. Instead of sending content, the agent can only indicate who is speaking and whether there is repetition. If task progress is slow, a suggested adjustment strategy for the agent is to adjust the dialogue style during the conversation: adjust the agent's speaking frequency, shorten response lengths, and introduce rhythm awareness.

[0117] In some embodiments, when the agent evolution device performs the above step S101 of "determining the agent adjustment strategy that matches the monitoring indicators", such as Figure 5 As shown, the following steps S503 can also be performed:

[0118] Step S503: In response to the background monitoring indicators in the monitoring indicators meeting the adjustment trigger conditions, determine the agent adjustment strategy as adjusting the agent's role setting and / or generating a third agent.

[0119] In the embodiments of this application, if the clustering divergence of the background vector is greater than the corresponding preset background threshold, it can be determined that the monitoring indicator meets the adjustment trigger condition, indicating poor background consistency. For example, the problem manifests as: large differences among members (which can be reflected in intelligence, instructions, language, etc.) and generalized Agent response.

[0120] For example, if members have significant differences, a corresponding third-party agent can be set up for specific groups (such as the finance department or the legal department). This allows for the generation of different strategies for the same topic. For instance, regarding legal issues, if a member's background is legal professional, the corresponding third-party agent can generate a response using legal terminology. If a member's background is finance professional, the agent can translate legal terminology into a response understandable to finance professionals or other non-professionals. This avoids global intervention and allows for binding only to specific individuals. For example, legal professionals in a group can be grouped into user groups and bound to corresponding third-party agents, enabling the agents to provide services to members within those user groups.

[0121] In the embodiments of this application, the third intelligent agent may be a sub-intelligent agent of the first intelligent agent or the second intelligent agent, or it may be an intelligent agent independent of the first intelligent agent and the second intelligent agent.

[0122] For example, the relationship between different dimensions of problems, problem manifestations, suggested adjustment strategies, and examples of adjustment details is shown in Table 3:

[0123] Table 3

[0124]

[0125] As shown in Table 3, different problems manifest for different group dimensions. For different problem manifestations, corresponding Agent reconstruction strategies are set (consistent with the agent adjustment strategies discussed above). Then, the agents are adjusted based on the Agent reconstruction strategies to obtain target agents, so as to use the target agents to provide more accurate services to the members in the group.

[0126] like Figure 6 As shown, an exemplary implementation of an intelligent agent evolution method is provided, including steps S601 to S606:

[0127] Step S601: Group dynamic status monitoring.

[0128] Here, the intelligent agent evolution device can monitor the state of the group. Exemplary monitoring dimensions are shown in Table 2, resulting in the last column of monitoring indicators.

[0129] Step S602: Does it exceed the threshold?

[0130] Here, the intelligent agent evolution device sets corresponding thresholds for the parameter values ​​in the monitoring indicators of each dimension. For example, refer to steps S401, S402, and S403 above. If any parameter exceeds the threshold, step S603 will be triggered. If the threshold is not exceeded, step S604 will be executed.

[0131] Step S603: Trigger the agent reconstruction mechanism.

[0132] Here, if any parameter in the monitoring indicators exceeds the threshold, the intelligent agent evolution device will be triggered to execute step S603 and trigger the Agent reconstruction mechanism, which corresponds to step S101 above: in response to the monitoring indicators determined by the group feature information meeting the adjustment trigger conditions. For an exemplary discussion, please refer to the relevant discussion of step S101 above, which will not be repeated here.

[0133] Step S604: Continue monitoring and maintain the configuration.

[0134] This indicates that no parameter in the monitoring indicators exceeds the threshold. Therefore, the group can continue to be monitored, and the configuration information of the agent can remain unchanged.

[0135] Step S605: Determine the agent reconstruction strategy.

[0136] Here, the agent evolution device can adjust based on which monitoring indicator exceeds the threshold, and then adjust according to the corresponding adjustment strategy for that dimension. For example, refer to the discussion of steps S501, S502, and S503 above, and the correspondence shown in Table 3, to determine the corresponding strategy. It should be noted that for the same monitoring indicator, multiple reconstruction strategies can be determined within the agent reconstruction strategy (consistent with the agent adjustment strategies described above). Exemplary agent reconstruction strategies include: creating an agent clone 641, focusing on a subset of context 642, fine-tuning role settings 643, and fine-tuning speech reception / speech style 644.

[0137] Step S606: Release the new Agent adapter.

[0138] Here, the agent evolution device can adjust existing agents based on agent reconstruction strategies to obtain new agent adapters (corresponding to the target agent discussed above), and then use the released new agents to serve the group.

[0139] In the embodiments of this application, every change to the Agent can be recorded for system auditing and optimization. This information can be provided to operators, target personnel, or the system for optimization. Alternatively, each change can be recorded, and based on the corresponding change, the agent's configuration information can be promptly reclaimed when no longer needed. For example, as shown in Table 4:

[0140] Table 4

[0141]

[0142] As shown in Table 4, if the topic is concurrently generated into concurrent projects, Agent-A-Project1 is generated to serve the first project, and Agent-A-Project2 serves the second project. If the second project ends, Agent-A-Project2 is recycled; if the first project ends, Agent-A-Project1 is recycled. Both Agent-A-Project1 and Agent-A-Project2 are sub-agents of the Agent.

[0143] As shown in Table 4, if multiple new members (legal roles) are added, Agent-F-Legal can be generated to provide services to the new members. If the service is no longer needed, or if the member leaves the group, or if the project ends, Agent-F-Legal can be reclaimed. Agent-F-Legal can also be a sub-agent of Agent.

[0144] As shown in Table 4, if the groups become static and only generate meeting minutes for regular meetings, then the Agent for specific tasks can be reclaimed and deleted, while a general Agent with a small number of parameters is retained to monitor the group.

[0145] The comparative advantages of the embodiments of this application compared with related technologies are shown in Table 5:

[0146] Table 5

[0147]

[0148] As shown in Table 5, the intelligent agent self-evolution method provided in this application has a fast response speed, strong scalability, and can add coverage areas in real time through dynamic binding tools and reconstructed clones.

[0149] This application provides an agent evolution method, which includes: monitoring group characteristic information of a service group corresponding to the agent; determining an agent adjustment strategy matching the monitoring indicators in response to the monitoring indicators determined by the group characteristic information satisfying the adjustment trigger condition; adjusting the agent configuration based on the agent adjustment strategy to obtain a target agent; and using the target agent to serve the group. The agent evolution method provided in this application updates and adjusts the agent in a timely manner when the session requirements in the group change, so that the adjusted target agent adapts to the development and changes in the group state, thus providing greater flexibility.

[0150] This application provides an intelligent agent evolution device 7, such as... Figure 7 As shown, it includes:

[0151] The determination module 71 is used to monitor the group characteristic information of the service group corresponding to the intelligent agent, and in response to the monitoring indicator determined by the group characteristic information satisfying the adjustment trigger condition, determine the intelligent agent adjustment strategy matching the monitoring indicator;

[0152] The adjustment module 72 is used to adjust the configuration of the agent based on the agent adjustment strategy to obtain a target agent, and use the target agent to serve the group.

[0153] In one embodiment of this application, the adjustment module 72 is further configured to, in response to the monitoring indicator determined by the group feature information satisfying the configuration recovery trigger condition, recover the adjustment configuration information corresponding to the target agent; the adjustment configuration information is a configuration based on the agent adjustment strategy corresponding to the monitoring indicator.

[0154] In one embodiment of this application, the group feature information includes at least one or more of the following: topic semantic feature information or session interaction feature information of the conversation within the group, and background feature information of the members within the group.

[0155] In one embodiment of this application, the determining module 71 obtains the conversation content of the group within a preset range, and performs noise reduction processing on the conversation content to obtain target conversation content; based on the target conversation content, it determines the group feature information; wherein, the noise reduction processing includes at least one of the following: filtering non-task dialogues in the conversation content; removing dialogues in the conversation content whose sentence vector offset from the target task is greater than a preset offset; removing dialogues in the conversation content whose response frequency is lower than a preset frequency.

[0156] In one embodiment of this application, the determining module 71 is further configured to determine the topic semantic monitoring index of the conversation within the group based on the parameters included in the topic semantic feature information; the topic semantic feature information includes at least one or more of topic model distribution and sentence vector offset; determine the conversation interaction monitoring index of the conversation within the group based on the parameters included in the conversation interaction feature information; the conversation interaction feature information includes at least one or more of the average number of turns in each dialogue in the conversation and conversation repetition rate; and determine the background monitoring index of the members within the group based on the parameters included in the background feature information; the background feature information includes at least one or more of background vector similarity and job difference.

[0157] In one embodiment of this application, the determining module 71 is further configured to: determine that the monitoring indicator satisfies the adjustment triggering condition in response to a parameter value in the topic semantic monitoring indicator being greater than a preset topic semantic threshold; the topic semantic monitoring indicator includes at least topic model distribution entropy and / or sentence vector offset; determine that the monitoring indicator satisfies the adjustment triggering condition in response to a parameter value in the conversation interaction monitoring indicator being greater than a preset conversation interaction threshold; the conversation interaction monitoring indicator includes at least conversation cycle entropy and / or repetitive information ratio; determine that the monitoring indicator satisfies the adjustment triggering condition in response to a parameter value in the background monitoring indicator being greater than a preset background threshold; the background monitoring indicator includes at least background vector clustering divergence.

[0158] In one embodiment of this application, the determining module 71 is further configured to, in response to the topic semantic monitoring indicator in the monitoring indicators satisfying the adjustment trigger condition, determine that the agent adjustment strategy is context-focusing of the conversation in the group and / or generating a second agent; in response to the conversation interaction monitoring indicator in the monitoring indicators satisfying the adjustment trigger condition, determine that the agent adjustment strategy is adjusting the role setting of the agent and / or adjusting the dialogue mode in the conversation; and in response to the background monitoring indicator in the monitoring indicators satisfying the adjustment trigger condition, determine that the agent adjustment strategy is adjusting the role setting of the agent and / or generating a third agent.

[0159] This application provides an intelligent agent evolution device 8, including a processor 81 and an intelligent agent 82 capable of running on the processor. The intelligent agent 82 can call at least one processing model 83 deployed in the intelligent agent evolution device 8 to perform the following operations: monitor the group characteristic information of the service group corresponding to the intelligent agent; in response to the monitoring indicators determined by the group characteristic information satisfying the adjustment trigger condition, determine an intelligent agent adjustment strategy matching the monitoring indicators; based on the intelligent agent adjustment strategy, adjust the configuration of the intelligent agent to obtain a target intelligent agent, and use the target intelligent agent to serve the group.

[0160] In one embodiment of this application, the agent 82 can also invoke at least one processing model 83 deployed in the agent evolution device to perform the following operations: when the agent adjustment strategy is to generate or reclaim a configured agent, generate or reclaim a configured agent; configure the agent as a first agent or a second agent.

[0161] This application provides an agent evolution device that monitors group characteristic information of the service group corresponding to the agent. In response to monitoring indicators determined by the group characteristic information meeting adjustment trigger conditions, an agent adjustment strategy matching the monitoring indicators is determined. Based on the agent adjustment strategy, the agent's configuration is adjusted to obtain a target agent, which is then used to serve the group. The agent evolution device provided in this application updates and adjusts the agent in a timely manner when session requirements change within the group, enabling the adjusted target agent to adapt to changes in the group's state, thus providing greater flexibility.

[0162] This application provides a computer-readable storage medium storing one or more computer programs that can be executed by one or more processors to implement the aforementioned intelligent agent evolution method. The computer-readable storage medium can be transient or non-transient.

[0163] This application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, it implements some or all of the steps in the above-described method. This computer program product can be implemented specifically through hardware, software, or a combination thereof. In one optional embodiment, the computer program product is specifically embodied as a computer storage medium; in another optional embodiment, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0164] In some embodiments, the storage medium may be a computer-readable storage medium, which may be volatile memory, such as random-access memory (RAM); or non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), ferromagnetic random access memory (FRAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory, magnetic surface memory, optical disk, or compact disk-read-only memory (CD-ROM); or it may be various devices including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0165] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0166] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1A device that provides the functions specified in one or more boxes.

[0167] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0168] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0169] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0170] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file containing other programs or data, for example, in one or more scripts within a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple co-located files (e.g., files storing one or more modules, subroutines, or code sections). As an example, executable instructions may be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0171] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

[0172] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0173] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not performed.

[0174] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent agent evolution, the method comprising: The system monitors the group characteristic information of the service group corresponding to the intelligent agent, and in response to the monitoring indicators determined by the group characteristic information meeting the adjustment trigger conditions, it determines the intelligent agent adjustment strategy that matches the monitoring indicators. Based on the agent adjustment strategy, the agent is configured and adjusted to obtain a target agent, and the target agent is used to serve the group.

2. The agent evolution method according to claim 1, wherein after the target agent continues to serve the group, the method further includes: In response to the monitoring indicators determined by the group feature information satisfying the configuration recovery trigger condition, the adjustment configuration information corresponding to the target intelligent agent is recovered; The adjustment configuration information is a configuration based on the agent adjustment strategy corresponding to the monitoring indicator.

3. The agent evolution method according to claim 1, wherein the group feature information includes at least: The topic semantic features or conversation interaction features of the group conversation, and one or more of the background features of the group members.

4. The agent evolution method according to any one of claims 1 to 3, wherein the group characteristic information of the service group corresponding to the monitoring agent includes: Obtain the conversation content of the group within a preset range, and perform noise reduction processing on the conversation content to obtain the target conversation content; Based on the target session content, the group characteristic information is determined; The noise reduction process includes at least one of the following: filtering non-task dialogues in the session content; removing dialogues in the session content whose sentence vector offset from the target task is greater than a preset offset; and removing dialogues in the session content whose response frequency is lower than a preset frequency.

5. The agent evolution method according to claim 2, wherein the monitoring indicators determined by the group feature information include at least one of the following: Based on the parameters included in the topic semantic feature information, the topic semantic monitoring indicators of the conversation within the group are determined; the topic semantic feature information includes at least one or more of topic model distribution and sentence vector offset; Based on the parameters included in the conversation interaction feature information, the conversation interaction monitoring indicators of the conversation within the group are determined; the conversation interaction feature information includes at least one or more of the following: the average number of rounds for each dialogue in the conversation and the conversation repetition rate. Based on the parameters included in the background feature information, the background monitoring indicators of the members in the group are determined; the background feature information includes at least one or more of background vector similarity and job difference.

6. The agent evolution method according to claim 5, wherein the monitoring indicators determined by the group feature information satisfy the adjustment trigger condition, including at least one of the following: When the parameter value in the topic semantic monitoring indicator is greater than the preset topic semantic threshold, it is determined that the monitoring indicator meets the adjustment trigger condition; the topic semantic monitoring indicator includes at least topic model distribution entropy and / or sentence vector offset. In response to the parameter values ​​included in the conversation interaction monitoring indicators being greater than a preset conversation interaction threshold, it is determined that the monitoring indicators meet the adjustment triggering conditions; the conversation interaction monitoring indicators include at least the conversation cycle entropy and / or the proportion of repeated information. In response to the background monitoring indicator having a parameter value greater than a preset background threshold, it is determined that the monitoring indicator meets the adjustment trigger condition; The background monitoring indicators include at least the background vector clustering divergence.

7. The agent evolution method according to claim 5 or 6, wherein determining the agent adjustment strategy matching the monitoring index includes at least one of the following: In response to the topic semantic monitoring indicator in the monitoring indicators satisfying the adjustment trigger condition, the agent adjustment strategy is determined to be context-focusing of the conversation in the group and / or generating a second agent; If the session interaction monitoring indicator in the monitoring indicators meets the adjustment trigger condition, the agent adjustment strategy is determined to be to adjust the agent's role setting and / or adjust the dialogue method in the session. In response to the background monitoring indicator in the monitoring indicators satisfying the adjustment trigger condition, the agent adjustment strategy is determined to be adjusting the role setting of the agent and / or generating a third agent.

8. An intelligent agent evolution device, comprising a processor and an intelligent agent capable of running on the processor, the intelligent agent being able to invoke at least one processing model deployed in the intelligent agent evolution device to perform the following operations: Monitor the group characteristic information of the service group corresponding to the agent, and in response to the monitoring indicators determined by the group characteristic information meeting the adjustment trigger conditions, determine the agent adjustment strategy that matches the monitoring indicators; Based on the agent adjustment strategy, the agent is configured and adjusted to obtain a target agent, and the target agent is used to serve the group.

9. The intelligent agent evolution device according to claim 8, wherein the intelligent agent is further capable of invoking at least one processing model deployed in the intelligent agent evolution device to perform the following operations: When the agent adjustment strategy is to generate or reclaim a configuration agent, the configuration agent is generated or reclaimed; the configuration agent is either a first agent or a second agent.

10. A computer-readable storage medium storing one or more computer programs, which can be executed by one or more processors to implement the intelligent agent evolution method according to any one of claims 1 to 7.

Images