Self-evolution intelligent dialogue system, dialogue and electronic device

Abstract

The application provides a self-evolution intelligent conversation system, method and electronic equipment, wherein the system comprises an intelligent conversation module, a conversation record and data storage module and a conversation strategy generation and self-evolution optimization module; the application calls a preset large language model through the intelligent conversation module to identify a conversation intention and a target conversation communication channel of a to-be-processed conversation task, then selects a target conversation strategy from a plurality of pre-generated conversation strategy according to the conversation intention recognition result, and then initiates an intelligent conversation with a customer based on the target conversation strategy according to the corresponding target conversation communication channel. In the whole process, the intelligent conversation record between the customer is recorded continuously and a new conversation strategy is generated according to the intelligent conversation record. In this way, the conversation strategy can be automatically adjusted according to the actual conversation without manual intervention, so as to realize the continuous automatic optimization of the conversation efficiency.

Description

Technical Field

[0001] This application relates to the field of intelligent dialogue technology, and in particular to a self-evolving intelligent dialogue system, method and electronic device. Background Technology

[0002] With the development of artificial intelligence technology, engaging in dialogue with customers through intelligent dialogue systems can help improve dialogue efficiency and reduce labor costs. Currently, common intelligent dialogue technologies mainly take the form of: traditional intelligent outbound calling systems based on rule engines and script templates, or dialogue strategy optimization systems based on reinforcement learning, digital dialogue solutions based on machine learning, and so on.

[0003] While existing intelligent dialogue technologies have improved dialogue efficiency and reduced labor costs to some extent, their dialogue strategies are still statically configured or optimized for specific occasions. Once the dialogue system is online, it is difficult to make real-time adaptive updates based on the latest dialogue data and user feedback, lacking the ability for continuous self-evolution at the dialogue level. Summary of the Invention

[0004] In view of this, embodiments of this application provide a self-evolving intelligent dialogue system, method, and electronic device to automatically adjust the dialogue strategy according to the actual dialogue, thereby achieving continuous automatic optimization of dialogue efficiency.

[0005] In a first aspect, embodiments of this application provide a self-evolving intelligent dialogue system, wherein the system includes: The module comprises an intelligent dialogue module, a dialogue recording and data storage module, and a dialogue strategy generation and self-evolution optimization module, among which: The intelligent dialogue module is used to respond to a dialogue task to be processed, call a pre-set large language model to identify the dialogue intent and target dialogue communication channel of the dialogue task to be processed, select a target dialogue script strategy from the dialogue script strategy generated by the dialogue strategy generation and the dialogue script strategy pre-generated by the self-evolution optimization module according to the dialogue intent identification result, and initiate an intelligent dialogue with the customer corresponding to the dialogue task to be processed according to the target dialogue communication channel based on the target dialogue script strategy. The dialogue record and data storage module is used to record the intelligent dialogue records between the intelligent dialogue module and the customer; The dialogue strategy generation and self-evolution optimization module is used to generate new dialogue script strategies based on the intelligent dialogue records between the intelligent dialogue module and the customer, and store the new script strategies in the dialogue record and data warehouse module.

[0006] Secondly, embodiments of this application provide a self-evolving intelligent dialogue method, wherein the method is applied to the self-evolving intelligent dialogue system described in the first aspect, and the method includes: In response to a pending dialogue task, a pre-set large language model is invoked to identify the dialogue intent and target dialogue communication channel of the pending dialogue task. Based on the dialogue intent identification result, a target dialogue script strategy is selected from the pre-generated dialogue script strategy. Based on the target dialogue script strategy, an intelligent dialogue is initiated with the customer corresponding to the pending dialogue task according to the target dialogue communication channel. The system records the intelligent dialogue between the intelligent dialogue module and the customer, and generates and stores new dialogue strategies based on these records.

[0007] Thirdly, embodiments of this application provide an electronic device, wherein the electronic device includes: a processor; and a memory storing a program; wherein the program includes instructions, which, when executed by the processor, cause the processor to perform the self-evolving intelligent dialogue method described in the second aspect.

[0008] Fourthly, embodiments of this application provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute the self-evolving intelligent dialogue method described in the second aspect.

[0009] The beneficial effects of this application are: This application provides a self-evolving intelligent dialogue system, method, and electronic device. The system includes an intelligent dialogue module, a dialogue recording and data storage module, and a dialogue strategy generation and self-evolving optimization module. The intelligent dialogue module uses a pre-built large language model to identify the dialogue intent and target communication channel of the dialogue task to be processed. Then, based on the dialogue intent identification result, it selects a target dialogue script strategy from a large number of pre-generated dialogue script strategies. Finally, based on the target dialogue script strategy, it initiates an intelligent dialogue with the customer according to the corresponding target communication channel. Throughout the process, the intelligent dialogue records with the customer are continuously recorded, and new dialogue script strategies are generated based on these records. In this way, the dialogue script strategy can be automatically adjusted according to the actual dialogue without human intervention, thereby achieving continuous automatic optimization of dialogue efficiency. Attached Figure Description

[0010] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This paper presents a schematic diagram of a system architecture for the self-evolving intelligent dialogue system provided in this application. Figure 2 This paper presents a schematic diagram of another system architecture for the self-evolving intelligent dialogue system provided in this application. Figure 3This paper presents a flowchart illustrating a self-evolving intelligent dialogue method provided in this application. Figure 4 A structural block diagram of an exemplary electronic device that can be used to implement embodiments of this application is shown. Detailed Implementation

[0011] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0012] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0013] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the following description. It should be noted that the concepts of "first", "second", etc., mentioned in this application are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0014] It should be noted that the terms "a" and "a plurality of" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0015] Currently, common intelligent dialogue technologies mainly present the following typical solutions: The first category is traditional intelligent outbound calling systems based on rule engines and script templates. These systems typically have pre-set fixed dialogue flows and script templates, identifying user intent through keyword matching and advancing the conversation according to predefined branching logic. For example, some companies' intelligent dialogue solutions provide intelligent outbound calling, intelligent interaction, and operation guidance functions through technologies such as natural language processing, deep learning, knowledge graphs, and multi-turn dialogue. Other companies' intelligent dialogue solutions achieve personalized dialogue scripts for each user through intelligent interaction and human-computer collaboration.

[0016] The second category is dialogue strategy optimization systems based on reinforcement learning. For example, in consumer finance scenarios, some intelligent telephone dialogue strategies based on reinforcement learning are used to organize data such as debtor information, account information, and call collection records. A Long Short-Term Memory (LSTM) network model is trained to predict the repayment rate under different dialogue strategies, and a Markov decision process is used to find the optimal dialogue time and target. This type of solution mainly focuses on optimizing dialogue timing and calling strategies, but does not involve the automatic evolution of the dialogue content itself.

[0017] The third category is intelligent dialogue systems based on intelligent quality inspection. Platforms like Megaview provide intelligent quality inspection services based on ASR (Automatic Speech Recognition) speech recognition and NLP semantic analysis. They support functions such as compliance inspection, SOP (Standard Operating Procedure) process inspection, and correlation inspection, performing sensitive word detection, sentiment analysis, and process compliance checks on dialogues. These systems primarily serve as post-event monitoring tools; the inspection results require manual interpretation and strategy adjustments, failing to form an automated closed loop.

[0018] The fourth category is a machine learning-based digital dialogue solution built on the TrueAccord platform, which uses the HeartBeat decision engine to automatically optimize grouping, scripts, and sending rhythm based on user behavior feedback.

[0019] In summary, existing intelligent dialogue technologies have improved dialogue efficiency to some extent, but they still have the following shortcomings: It is a static configuration or only optimized for timing, lacking the ability for continuous self-evolution at the level of content; there is a disconnect between quality inspection and strategy updates, relying on manual intervention.

[0020] In view of this, this application provides a self-evolving intelligent dialogue system, method, and electronic device to automatically adjust the dialogue strategy according to the actual dialogue without human intervention, thereby achieving continuous automatic optimization of dialogue efficiency.

[0021] In one aspect, this application provides a self-evolving intelligent dialogue system. This self-evolving intelligent dialogue system is a software system with a communication interface through which it can communicate with external clients. It is understood that the intelligent dialogue system in this application is an AI dialogue system; as the name suggests, the self-evolving intelligent dialogue system provided in this application is a dialogue system capable of self-optimization.

[0022] In some possible embodiments, the self-evolving intelligent dialogue system can be as follows: Figure 1As shown, it includes: an intelligent dialogue module, a dialogue recording and data storage module, and a dialogue strategy generation and self-evolution optimization module. Among them: The intelligent dialogue module is used to respond to the dialogue task to be processed, call the pre-set large language model to identify the dialogue intent and target dialogue communication channel of the dialogue task to be processed, select the target dialogue script strategy from the dialogue script strategies pre-generated by the dialogue strategy generation and self-evolution optimization module based on the dialogue intent recognition result, and initiate intelligent dialogue according to the target dialogue communication channel and the customer corresponding to the dialogue task to be processed based on the target dialogue script strategy. The dialogue recording and data storage module is used to record the intelligent dialogue records between the intelligent dialogue module and the customer; the dialogue strategy generation and self-evolution optimization module is used to generate new dialogue script strategies based on the intelligent dialogue records between the intelligent dialogue module and the customer, and store the new script strategies in the dialogue recording and data storage module.

[0023] The self-evolving intelligent dialogue system provided in this application will be described in detail below with specific examples: The self-evolving intelligent dialogue system provided in this application can be applied to any application scenario requiring intelligent dialogue. The corresponding tasks to be processed depend on the specific application scenario of the self-evolving intelligent dialogue system. It can be understood that the tasks to be processed are tasks requiring communication with customers, categorized according to business type. As one implementation method, the tasks to be processed are initiated by other systems, and the self-evolving intelligent dialogue system provided in this application is invoked through data interfaces between systems to communicate with customers and complete the dialogue tasks required by other systems. For example: In the context of consumer finance, the task to be processed can be a collection task. Through the self-evolving intelligent dialogue system provided in this application, the intelligent dialogue module can conduct intelligent dialogue with the borrower based on the collection task to remind the customer to repay on time.

[0024] Alternatively, in a supply chain management scenario, the task to be processed is a production expediting task. Through the self-evolving intelligent dialogue system provided in this application, the intelligent dialogue module can conduct intelligent dialogue with the supplier based on the production expediting task to remind the supplier to deliver on time.

[0025] Alternatively, in auxiliary medical application scenarios, the task to be processed is a reminder for medical treatment. Through the self-evolving intelligent dialogue system provided in this application, the intelligent dialogue module can conduct intelligent dialogue with the patient based on the reminder for medical treatment task, so as to remind the patient to go to the medical institution for treatment on schedule.

[0026] The following section will use the self-evolving intelligent dialogue system in the consumer finance scenario for intelligent debt collection as an example. Other application scenarios can be extended as needed, and this application will not elaborate on them.

[0027] In some possible embodiments, it may be as follows Figure 2 As shown, the self-evolving intelligent dialogue system provided in this application also includes: a data access and case management module, used for: The system uses a pre-defined data interface to obtain dialogue case data from a pre-defined business platform, generates pending dialogue tasks based on the dialogue case data, and assigns priorities to the pending dialogue tasks; wherein, the dialogue case data records the communication method of the target customer and the basic information of the target customer; After the intelligent dialogue module ends its intelligent dialogue with the customer, it sends the dialogue result data back to the business platform through the preset data interface so that the business platform can update the dialogue case data.

[0028] The dialogue case data refers to the background data that initiates this intelligent dialogue, including the communication method of the target customer and their basic information. For example, taking the aforementioned intelligent debt collection scenario, this dialogue case data refers to the background data for initiating this intelligent debt collection, including loan background information. This loan background information includes the target customer's communication method (such as contact information and address) and their basic information, which may include detailed data such as debtor basic information, debt status, historical dialogue records, user profile tags, and other multi-dimensional data.

[0029] As described above, the self-evolving intelligent dialogue system provided in this application is a software system that works in conjunction with other systems to communicate with customers. Taking a lending scenario as an example, this self-evolving intelligent dialogue system assists the financial lending business platform in communicating with customers. The financial lending business platform and the self-evolving intelligent dialogue system automatically retrieve online collection dialogue case data through data interface calls between systems, and perform standardized processing and lifecycle management. When the financial lending business platform needs to use the self-evolving intelligent dialogue system for automatic collection, the financial lending business platform grants the self-evolving intelligent dialogue system data access, allowing the system to automatically retrieve loan background data and then initiate intelligent dialogue with customers based on the corresponding pending tasks, thereby achieving automatic collection.

[0030] In this embodiment, the data access and case management module supports different interface methods to adapt to the technical architecture of different business systems. It standardizes and transforms the raw data retrieved from the business platform, unifying field naming, data format, and encoding specifications to ensure consistency in subsequent module processing. If the dialogue case data retrieved from the business platform has missing fields, the data access and case management module can infer and fill in the missing fields based on existing information or mark them as data to be confirmed, avoiding the impact of incomplete data on the communication accuracy of the intelligent dialogue module.

[0031] The data access and case management module can analyze the acquired case data to determine whether corresponding pending tasks need to be generated. For example, taking this self-evolving intelligent dialogue system as a debt collection system, the data access and case management module can analyze data transmitted from the financial business platform to determine if the customer is overdue. If overdue, a pending task is generated and sent to the intelligent dialogue module.

[0032] During this process, priority can be assigned to pending dialogue tasks based on the dialogue case data. For example, priority can be assigned based on the amount of overdue payment; dialogue cases with larger overdue amounts correspond to higher priority tasks. This helps the intelligent dialogue module prioritize the execution of high-priority pending dialogue tasks. After the intelligent dialogue module completes its intelligent dialogue with the customer, the data access and case management module can transmit the intelligent dialogue results back to the financial business platform through a data interface. This facilitates subsequent updates of the dialogue case data within the financial business platform. For instance, if a customer reports a low credit limit during the intelligent dialogue, this feedback can be transmitted back to the financial business platform through this data interface, allowing the platform to increase the credit limit based on the customer's feedback.

[0033] As one implementation method, the self-evolving intelligent dialogue system provided in this application may include several intelligent dialogue modules, each of which is used to initiate an intelligent dialogue. Therefore, the data access and case management module can allocate tasks to be processed according to the attributes of the dialogue case data and the occupancy status of the intelligent dialogue modules, and can use an intelligent allocation algorithm to assign different cases to the dialogue queues of different intelligent dialogue modules.

[0034] As the name suggests, the data access and case management module is used to manage pending tasks. In addition to prioritizing pending dialogue tasks, the data access and case management module can also add a human assistance flag to pending tasks that require human intervention, and assign pending dialogue tasks with the human assistance flag to the human agent queue, so that human agents can initiate human dialogue with customers based on the pending dialogue tasks in the human agent queue.

[0035] Furthermore, the data access and case management module can select the distribution time for pending dialogue tasks based on the customer's historical conversation connection records and the customer's geographical location. When the distribution time arrives, the pending dialogue task is distributed to the intelligent dialogue module, allowing the intelligent dialogue module to conduct intelligent dialogue with the customer at that time, which helps improve the success rate of intelligent dialogue. After the intelligent dialogue is completed, the data access and case management module will send the communication result data of the intelligent dialogue (such as the promised repayment amount, repayment plan, and reasons for failure in intelligent collection scenarios) back to the upstream business system, supporting both real-time interface push and batch file export to ensure the integrity of the business loop.

[0036] It is understood that the intelligent dialogue module in the self-evolving intelligent dialogue system provided in this application is the core functional module, responsible for engaging in dialogue with external customers. Compared with the traditional solution of engaging in dialogue with external customers based on static dialogue strategies, the intelligent dialogue module provided in this application understands the dialogue intent of this intelligent dialogue by parsing the dialogue task to be processed after receiving it.

[0037] As one implementation method, this intelligent dialogue module is an AI intelligent assistant, such as an AI agent, or AI intelligent agent in the AI ​​field. This intelligent dialogue module is an agent intelligent agent with intelligent dialogue capabilities. This agent intelligent agent can parse the received dialogue task by calling a large language model, and then understand the dialogue intent based on the natural language description of the dialogue task. Specifically, this intelligent dialogue module is an intelligent agent pre-configured based on role constraint rules, which are behavioral guidelines manually stipulated by business personnel according to the roles they play in the intelligent dialogue. These role constraint rules can be configured by business personnel using the Prompt configuration method in the AI ​​field to constrain the agent.

[0038] For example, in the intelligent debt collection scenario, the intelligent dialogue module plays the role of a consumer finance debt collector. When collecting debts from customers, these debt collectors adhere to certain behavioral guidelines, such as honesty, friendliness, and legal compliance. Business personnel pre-write behavioral norms based on these guidelines and configure the AI ​​Agent as a role constraint. This allows the AI ​​Agent to understand the behavioral norms that need to be followed during conversations with customers, thus avoiding any negative impact on customers or any violations or illegal activities.

[0039] The role constraint rules can also be adaptively optimized based on customer feedback during the continuous dialogue process of the self-evolving intelligent dialogue system. Then, the intelligent dialogue module is optimized based on the optimized role constraint rules to improve the intelligent dialogue capability of the intelligent dialogue model.

[0040] In this embodiment, the core architecture of the intelligent dialogue module (hereinafter referred to as Agent) is a combination of a state machine and a memory mechanism. The state machine maintains the current stage of the dialogue to ensure the integrity and standardization of the dialogue process. For example, in an intelligent debt collection scenario, the state machine ensures the integrity and standardization of the process at each stage, such as identity verification, debt explanation, repayment plan negotiation, objection handling, and termination confirmation, to avoid violating relevant laws and regulations. The memory mechanism stores key information during the dialogue process, providing contextual support for subsequent script generation. Again, using the aforementioned intelligent debt collection application scenario as an example, the memory mechanism can store key information such as user feedback, promises, objections, and emotional changes during the dialogue process, providing contextual support for subsequent debt collection script generation. At the beginning of each round of dialogue, the intelligent dialogue module first reads the basic case information and historical dialogue memory, and then, combined with the current dialogue state, generates targeted script content.

[0041] In this application, the decision-making process of the intelligent dialogue module is divided into three stages: intent understanding, strategy selection, and dialogue generation. Among them: In the intent understanding phase, the intelligent dialogue module (referred to as the Agent) utilizes the natural language understanding capabilities of a large language model to analyze customer responses based on historical dialogue records from case data in the pending tasks, identifying their true intent and implied emotional states. For example, in an intelligent debt collection scenario, it can analyze historical debt collection dialogue records from case data in the pending tasks to identify the customer's repayment intent for overdue bills. This includes confirmation of repayment, objections, requests for extensions, difficulty expressing oneself, refusal to communicate, and implied emotional states such as anxiety, resistance, cooperation, and hesitation.

[0042] Similarly, during the intelligent dialogue module's interaction with the customer, the intelligent dialogue module can also be used for: The system uses a large language model to analyze customer responses and identifies their intent based on those responses. In the early stages of intelligent dialogue, this module can analyze historical customer responses using the large language model to identify their intent. However, as the intelligent dialogue develops, it can continuously use the large language model to identify the customer's current intent based on their real-time responses, thus providing a more tailored dialogue strategy based on the customer's current attitude.

[0043] In some possible embodiments, it may be as follows Figure 2As shown, the self-evolving intelligent dialogue system provided in this application may further include a task skill package invocation module. This module integrates a large number of task skill packages for intelligent dialogue. Therefore, in response to an invocation request from the intelligent dialogue module, the task skill package invocation module determines the task skill package matching the various types of intent representations of the client and executes the execution logic encapsulated within the task skill package.

[0044] The intelligent dialogue module is also used to obtain dialogue script strategies that match the customer's intent from the dialogue records and data warehouse module; and to generate new dialogue scripts and output them to the customer according to the execution results of the task skill package and the dialogue script strategies.

[0045] It is understandable that during the strategy selection phase, the intelligent dialogue module can select an appropriate dialogue strategy branch and decide on the sequence of task skill packages to be invoked based on the intent recognition results, case attributes, and dialogue history. In this embodiment, the task skill package is a task software code package pre-encapsulated based on the dialogue interaction processing flow. This task software code package can be simply referred to as a Skill, corresponding to a skill package in the field of artificial intelligence. Different task software code packages are used to perform different business tasks. For example, in an intelligent debt collection scenario, a corresponding response skill package can be designed separately for different customer responses. For instance, if a customer expresses a low credit limit, a separate credit limit increase task software code package can be designed. By executing the credit limit calculation and verification logic built into this task software code package, a more suitable credit limit can be provided to the customer.

[0046] Based on this, different Skiils offer different operational capabilities. The intelligent dialogue module can invoke the appropriate Skill to address the customer's needs based on their expressed intent. For example, if a customer expresses difficulty in making repayments, the intelligent dialogue module (referred to as the Agent) will invoke the repayment ability assessment Skill and the installment plan recommendation Skill. If the customer raises an objection, the intelligent dialogue module (Referred to as the Agent) will invoke the objection handling Skill to provide a targeted response.

[0047] During the dialogue generation phase, the intelligent dialogue module (referred to as Agent) generates the final dialogue dialogue strategy based on the selected strategy and the information returned by Skill, combined with the dialogue style and compliance constraints in the role constraint rules (referred to as Prompt) configuration.

[0048] The role constraint rules (Prompt) configuration of the intelligent dialogue module (Agent) is the core carrier of strategy optimization. This role constraint rule (Prompt) includes modules such as system role definition, dialogue style guidelines, dialogue process specifications, compliance script constraints, and key information templates. In intelligent debt collection application scenarios, the system role of this intelligent dialogue module is that of an official bank representative, enhancing the authority and credibility of the dialogue. The dialogue style guidelines stipulate that the Agent should maintain a professional, patient, and firm communication attitude, avoiding extreme styles that are either too forceful or too weak. The dialogue process specifications define the notifications and inquiries that must be completed at each stage, ensuring the completeness of information in the dialogue. The compliance script constraints list prohibited expressions and mandatory risk warnings to prevent compliance risks. These configuration items in the role constraint rule (Prompt) can be dynamically adjusted through a self-evolution mechanism, enabling continuous optimization of the script strategy.

[0049] In this embodiment, the intelligent dialogue module (referred to as Agent) supports multi-channel interaction capabilities, allowing it to select appropriate communication channels based on case configuration and user preferences. For example, in telephone conversations, the intelligent dialogue module can convert text-based speech into speech output via text-to-speech (TTS) and convert user voice responses into text input via audio-visual recognition (ASR), achieving full-duplex voice interaction. In text-based conversations (such as app messages or SMS links), the Agent interacts directly with the customer in text format. The interaction logic across different channels shares the same Agent core, with adaptation only performed at the output layer, ensuring consistency in the dialogue strategy.

[0050] Agents' ability to differentiate their strategies is key to improving dialogue effectiveness. The self-evolving intelligent dialogue system provided in this application can match differentiated dialogue strategies to different customers based on their basic information. For example, in intelligent debt collection scenarios, this intelligent dialogue module can match differentiated intelligent debt collection dialogue strategies to different types of customers based on dimensions such as the customer's different delinquency stages (early reminders, mid-term dialogues, late-term pressure), risk level (low risk, medium risk, high risk), historical behavior (first-time delinquency, multiple delinquencies, tendency to be a defaulter), and communication preferences (time preference, channel preference, dialogue style preference).

[0051] For example, for low-risk users with early delinquencies, the intelligent dialogue module can adopt a gentle reminder style, emphasizing the convenience of repayment and the impact on credit; for high-risk users with multiple delinquencies, the intelligent dialogue module can adopt a more assertive style, emphasizing the legal consequences and the cost of defaulting. This personalized strategy significantly improves the targeting and effectiveness of the dialogue.

[0052] As described above, a task skill pack (or simply Skill) encapsulates various specific functions in the dialogue process into independently callable and flexibly orchestratable capability units. It provides standardized tool interfaces for the intelligent dialogue module (or simply Agent), enabling modular management and dynamic expansion of dialogue capabilities. The task skill packs are designed according to the single responsibility principle, with each task skill pack dedicated to completing a specific dialogue task. For intelligent debt collection applications, the self-evolving intelligent dialogue system provided in this application pre-configures the following core task skill packs: The identity verification task skill pack is used to verify the identity of customers, such as asking for the last four digits of their ID number, date of birth, home address, and other information to confirm the person being contacted.

[0053] The Debt Explanation Task Skills Package is used to clearly inform customers of key information such as the amount owed, the number of overdue days, and the method of calculating late payment penalties.

[0054] The repayment ability assessment task skill package is used to inquire about a customer's current income status, expenditure structure, asset situation, etc., to assess the customer's actual repayment ability.

[0055] The installment plan recommendation task skill package is used to intelligently match parameters such as the number of installments, the amount per installment, and the first installment date based on the customer's repayment ability assessment results, and generate a personalized installment plan.

[0056] The objection handling task skills package provides standardized response strategies for various objections raised by customers (such as questions about the amount, objections to fees, claims that payments have been made, etc.).

[0057] The Compliance Tips Task Skills Kit is used to insert legal risk warnings and explanations of the impact of credit information at critical moments to ensure the compliance of the dialogue process.

[0058] The Emotional Soothing Skills Kit is used to soothe customers who are emotionally agitated or resistant by employing empathetic language to reduce communication friction.

[0059] The repayment commitment confirmation task skill package is used to clearly confirm the repayment amount, repayment time, and repayment method when the customer expresses their willingness to repay, thus forming a clear repayment agreement.

[0060] The standard interface definition of the task skill package provided in this application determines whether the task skill package calling module can be normally called by the intelligent dialogue module. Each task skill package exposes a unified calling interface, comprising three parts: input parameters, output results, and status feedback. Input parameters include case context information (customer information, outstanding debt details, historical records), dialogue context information (current dialogue round, user's most recent reply, dialogue status), and task skill package-specific parameters (such as the required installment period range and interest rate parameters for the installment plan recommendation task skill package). Output results include execution status (success, failure, manual intervention required), returned data (such as the repayment ability assessment task skill package returning the assessment level and suggested plan), and script suggestions (script snippets or prompts generated by the task skill package). Status feedback records the execution log and performance indicators of the task skill package for subsequent quality control analysis and strategy optimization.

[0061] The decision to invoke task skill packs is made by the intelligent dialogue module, but this module offers flexible configuration capabilities to support different invocation strategies. Specifically, the intelligent dialogue module can schedule task skill packs based on rules, pre-defining the sequence of task skill packs to be invoked and the triggering conditions for different dialogue states. Alternatively, it can schedule task skill packs based on models, dynamically deciding on invocation based on the dialogue context. Regardless of the scheduling strategy used, this self-evolving intelligent dialogue system records complete information for each task skill pack invocation (invocation timing, input parameters, output results, and user response), providing data support for subsequent strategy optimization.

[0062] In this embodiment, the task skill pack also possesses scalability. Specifically, the development and integration of new task skill packs follow a unified interface specification. Simply implementing the standard interface and registering it with the task skill pack manager allows it to be dynamically invoked by the intelligent dialogue module. The task skill pack invocation module also supports hot updates of task skill packs, meaning that new task skill packs can be dynamically loaded or the implementation logic of existing task skill packs can be updated without restarting the system. This enables the self-evolving intelligent dialogue system to quickly respond to changes in business needs and flexibly expand its dialogue capabilities.

[0063] The task skill pack calling module is also used to manage the versions of task skill packs. It can manage multiple versions of each task skill pack, and record the version number, effective time, changes and effect indicators.

[0064] When the self-evolution mechanism triggers adjustments to task skill pack parameters, the self-evolutionary intelligent dialogue system generates a new version of the task skill pack configuration. After gray-scale verification, the old version is gradually replaced. Version management ensures the traceability and rollbackability of task skill pack evolution, reducing the risk of strategy updates. Gray-scale verification refers to first launching a portion of the new version of the task skill pack, and then determining whether to launch a large number of new versions based on the application effects of the launched new task skill packs.

[0065] In this application, the dialogue record and data warehouse module is the data hub of the entire self-evolving intelligent dialogue system. It is responsible for the structured storage of all dialogues and related metadata, providing unified data support for quality inspection analysis, strategy optimization, and business statistics.

[0066] Among these, the collection of dialogue data is the primary task of the module. From the beginning to the end of each dialogue, the dialogue recording and data warehouse module will collect and store the following information in real time: original dialogue content (audio calls are stored as audio files, and text dialogues are stored as message records), dialogue round information (user speech, agent response, and timestamp for each round), dialogue metadata (case ID, customer ID, dialogue agent version, Prompt configuration version used, skill sequence and parameters called), dialogue tags (intent tags, emotion tags, compliance tags, and effect tags), and dialogue results (call duration, reason for ending, promised repayment status, user feedback, etc.).

[0067] For voice calls, the dialogue recording and data storage module can integrate an ASR engine to convert recordings into text and perform timestamp alignment, supporting subsequent text analysis and voice feature analysis.

[0068] For the data storage portion of the dialogue history and data storage module, a data storage architecture is implemented. This architecture employs a layered design to help balance query efficiency and storage costs. The layered design is based on the time elapsed since the dialogue, specifically including: The hot data layer is used to store recent (e.g., the last 30 days) conversation records.

[0069] The warm data layer is used to store conversation records for the medium term (such as the past year).

[0070] The cold data layer is used to store historical (more than 1 year) conversation records, which are then loaded into the hot data layer when needed.

[0071] This layered architecture ensures that the system's query performance and cost are controllable when processing massive amounts of dialogue data.

[0072] The dialogue record and data storage module can create multi-dimensional indexes for intelligent dialogue content, such as by case ID, customer ID, time range, dialogue result, compliance tag, and sentiment tag. This facilitates quick retrieval based on index information such as case ID, customer ID, time range, dialogue result, compliance tag, and sentiment tag.

[0073] In addition, the dialogue record and data storage module also supports real-time or offline full-text search capabilities. Operations personnel can use keyword searches to locate dialogue records with specific content, which helps them quickly and comprehensively understand the specific content of a dialogue and facilitates quality inspection analysis and strategy optimization of intelligent dialogue content in the later stages.

[0074] The dialogue recording and data warehouse module is also used for de-identification and security management of sensitive data in intelligent dialogue content. For example, intelligent dialogue data may contain sensitive personal information (such as ID card numbers, bank card numbers, and home addresses). The dialogue recording and data warehouse module replaces sensitive information with masks when storing data to achieve automatic de-identification. The de-identification rules are configurable, supporting different user roles to access data with different de-identification levels. For data security management, access control, operation auditing, and data encryption mechanisms are used to ensure the security and compliance of dialogue data.

[0075] In addition, the conversation log and data warehouse module needs to perform data verification on the incoming data, including integrity checks (required field checks), consistency checks (related data checks), and reasonableness checks (numerical range checks), to ensure data accuracy. During the verification process, if any data fails to meet integrity, consistency, or reasonableness checks, it can be determined that the data does not meet quality requirements. Data that does not meet quality requirements is marked and isolated to prevent it from affecting subsequent analysis and optimization. The conversation log and data warehouse module can periodically generate data quality reports to help operations personnel understand the data quality status and address issues promptly.

[0076] like Figure 2 As shown, the self-evolving intelligent dialogue system provided in this application further includes: a quality inspection and evaluation module, wherein the quality inspection and evaluation module is used for: The system performs real-time, multi-dimensional quality inspection and analysis on the intelligent dialogue records between the intelligent dialogue module and the customer, and generates a structured quality inspection and evaluation report. The multi-dimensional quality inspection and analysis includes: compliance analysis, execution of dialogue script strategies, customer experience analysis, and logical analysis.

[0077] Specifically, the quality inspection and evaluation module is the core of the self-evolving intelligent dialogue system's intelligent analysis. It is used to conduct multi-dimensional quality assessments of intelligent dialogues, identify dialogue issues and optimization opportunities, and provide a basis for decision-making regarding the subsequent self-evolution of dialogue strategies. The quality inspection and evaluation employs a multi-dimensional indicator system that comprehensively covers all quality elements of the dialogue.

[0078] Compliance analysis includes detecting whether the dialogue contains any illegal content, which is the primary task of quality inspection. This specifically includes the following detection items: sensitive word detection (insults, threats, intimidation, discriminatory language, violent expressions), identification of illegal promises (promises of reduction or exemption beyond authorized limits, misleading statements, false advertising), detection of prohibited dialogue behaviors (impersonating law enforcement officials, harassing third parties, improper pressure, insults and defamation), completeness of legal risk warnings (whether legal consequences are communicated at the prescribed time), privacy information protection (whether third-party information is leaked), and compliance of dialogue time (whether the dialogue takes place within legally permitted time periods). The self-evolving intelligent dialogue system provided in this application has a built-in dialogue compliance rule library covering industry standards such as the "Guidelines for Post-Loan Dialogue Risk Control in Internet Finance Personal Online Consumer Credit," which strictly intercepts illegal dialogue.

[0079] The execution status of dialogue script strategies includes: evaluating whether the intelligent dialogue module executes according to the established dialogue script strategy, including the completeness of task skill package invocation (e.g., whether the Skill required by the strategy is invoked), process node coverage (whether all necessary process nodes are completed), omission of key questions (e.g., whether necessary identity verification or information confirmation is omitted), and compliance with role constraint rules (whether the generated script conforms to the style and constraints in the Prompt).

[0080] Customer experience analysis includes analyzing customers' communication feelings, including politeness (whether the language used is appropriate), emotional calming effect (whether it is effective in calming agitated customers), responsiveness (whether the response is timely), and clarity of information (whether the content is easy to understand).

[0081] Logical analysis includes analyzing the business effectiveness of the dialogue. Business effectiveness can be measured by the actual output of the intelligent dialogue. For intelligent debt collection applications, business effectiveness can include changes in repayment willingness (whether the user's repayment willingness has increased before and after the dialogue), commitment conversion rate (the proportion of committed repayments), commitment fulfillment rate (the actual repayment ratio after commitment), call efficiency (dialogue output per unit of time), etc.

[0082] The logical dimension assesses the internal logic of the dialogue, including dialogue coherence (whether the wording is consistent), information consistency (whether the information in different rounds is contradictory), and the completeness of key information (whether the necessary information is fully disclosed).

[0083] In this embodiment, the quality inspection and evaluation module incorporates a built-in quality inspection engine, which employs a hybrid architecture combining a rule engine and a machine learning model. The rule engine is used to accurately match explicit compliance requirements, ensuring a high detection rate. For example, for explicit compliance requirements (such as sensitive words or mandatory content), the rule engine performs precise matching to ensure a 100% detection rate. The machine learning model is used to evaluate instructions at the semantic level, employing pre-trained language models and custom-trained classification models for analysis, providing quantitative scores. This quality inspection engine supports both real-time and offline inspection modes. Real-time inspection is executed synchronously during the dialogue, providing real-time alerts or even interrupting the dialogue for serious violations (such as threatening language). Offline inspection is executed in batches after the dialogue ends, allowing for more comprehensive and in-depth analysis.

[0084] For each intelligent dialogue, the quality inspection and evaluation module generates a structured quality inspection report containing scores for each dimension, a list of identified issues, a severity level of the issues, specific dialogue segments corresponding to the issues, and optimization suggestions. It also statistically summarizes all dialogues within a certain period or for a specific type of case, generating issues distribution reports, trend change reports, and strategy effectiveness comparison reports to provide clear direction and basis for subsequent strategy optimization.

[0085] This quality inspection and evaluation module is also used to identify and efficiently analyze problematic dialogue patterns. Specifically, by mining large amounts of dialogue data, the module can automatically identify common problematic dialogue patterns, such as repeatedly interrupting users, failing to respond to key user questions, applying excessive pressure that leads to user resistance, and using overly mechanical and inflexible dialogue. These problematic patterns are then abstracted into detectable pattern rules and applied to the quality inspection of subsequent dialogues, enabling continuous accumulation and improvement of problem identification capabilities.

[0086] Among these, efficient dialogue analysis can focus on identifying high-performing dialogue patterns. Specifically, by comparing and analyzing the dialogue characteristics of conversations with high commitment and high fulfillment rates, the quality inspection and evaluation module can extract common features of efficient dialogues, such as opening structure, objection response strategies, and pressure pacing control. These efficient patterns can serve as reference material for strategy optimization, helping the self-evolving intelligent dialogue system learn better dialogue techniques.

[0087] It is understood that the dialogue strategy generation and self-evolutionary optimization module in this application is responsible for generating the dialogue scripts and strategies required for intelligent dialogue with customers. In some possible embodiments, this dialogue strategy generation and self-evolutionary optimization module is used to: Based on the structured quality inspection and evaluation report, the role constraint rules of the intelligent dialogue module are updated by combining the intelligent dialogue records between the intelligent dialogue module and the customer, and the intelligent dialogue module is updated based on the new role constraint rules.

[0088] In this embodiment, the dialogue strategy generation and self-evolution optimization module is the intelligent hub of the self-evolutionary intelligent dialogue system provided in this application. It is used to transform quality inspection and evaluation results into specific strategy adjustment schemes, achieving continuous automatic optimization of the dialogue strategy. The core of this dialogue strategy generation and self-evolution optimization module is to map the problems discovered by the quality inspection and evaluation module to the parameter adjustments of the corresponding dialogue script strategies. For example, if the quality inspection report output by the quality inspection and evaluation module identifies problem types including: compliance issues (such as the use of sensitive words, illegal promises), process issues (such as missing skill calls, missing key questions), script issues (such as insufficient persuasiveness, inappropriate emotional handling), and effectiveness issues (such as low commitment rate, strong user resistance), the dialogue strategy generation and self-evolution optimization module can adopt different optimization strategies for different types of problems. Specifically: For compliance issues, the dialogue strategy generation and self-evolution optimization module can directly locate the corresponding role constraint rule fragment (i.e., locate the corresponding field in the corresponding Prompt) or the configuration of the task skill pack, and further generate revision schemes (such as replacing sensitive words or adding compliance constraints).

[0089] For process issues, the dialogue strategy generation and self-evolution module can analyze the differences in the Skill call sequence and generate adjustment schemes for the call logic (such as adding mandatory Skills or adjusting the call order).

[0090] Regarding dialogue issues, the dialogue strategy generation and self-evolution module can generate suggestions for adjusting the dialogue style based on the characteristics of efficient dialogue patterns (such as adjusting the opening structure and optimizing objection response strategies).

[0091] It's understandable that the core function of the dialogue strategy generation and self-evolution module is to automatically optimize the Prompt. The role constraint rules (Prompt) of the intelligent dialogue module consist of multiple paragraphs, including system role definitions, dialogue style guidelines, dialogue process specifications, compliance script constraints, and key information templates. The dialogue strategy generation and self-evolution module can automatically adjust each paragraph of the Prompt with fine-grained precision. For example, when quality control finds that the customer hang-up rate in the opening phase is below a certain threshold, the dialogue strategy generation and self-evolution module can analyze the characteristics of effective openings and automatically generate an optimized opening template (such as adjusting the opening greeting, optimizing the self-introduction, and adjusting the expression of the dialogue purpose). When the quality control evaluation module finds that the promised conversion rate is low in the objection handling phase, the dialogue strategy generation and self-evolution module can analyze the objection response strategies of successful cases and automatically generate optimized objection handling script segments. These adjustments, while maintaining the overall stability of the Prompt structure, refine the local content, achieving gradual improvement of the strategy.

[0092] Skill invocation logic optimization is another core capability of the dialogue strategy generation and self-evolution module. The skill invocation logic comprises three levels: invocation conditions, invocation order, and invocation parameters. Invocation conditions define which skill should be invoked under what circumstances. The dialogue strategy generation and self-evolution module can automatically adjust the invocation condition rules based on skill invocation omissions or errors detected during quality checks. Invocation order defines the sequential relationship between multiple skills. The dialogue strategy generation and self-evolution module can explore better combinations of invocation orders based on dialogue performance data. Invocation parameters define the specific parameter settings when a skill is executed. For example, in the field of intelligent debt collection, this includes the period range and interest rate parameters for the recommended installment plan skill. The dialogue strategy generation and self-evolution module can automatically optimize parameter configurations based on response data from different types of debtors.

[0093] In this embodiment, the dialogue strategy generation and self-evolutionary optimization module includes a pre-built strategy optimization algorithm, which combines rule engine and large language model decision-making. For adjustments with clear causal relationships (such as replacing sensitive words or adding essential content), the rule engine directly generates adjustment schemes to ensure the determinism and interpretability of the adjustments. For adjustments with complex causal relationships (such as optimizing speech style or exploring skill call order), strategy generation based on a large language model is used. The selection of this strategy optimization algorithm can be automatically matched according to the characteristics of the optimization task, or it can be manually specified by the operations personnel.

[0094] The dialogue strategy generation and self-evolution optimization module also provides interpretable explanations for the generated dialogue script strategy optimization schemes. Specifically, for each dialogue script strategy adjustment, this module generates detailed adjustment instructions, including the reason for the adjustment (which quality control issue it corresponds to), the content of the adjustment (which configurations were modified), the expected effect (what goals are hoped to be achieved), and a risk assessment (potential negative impacts). These instructions help operations personnel understand the system's self-evolution logic and enable manual review and intervention when necessary.

[0095] The dialogue strategy generation and self-evolution optimization module provided in this application also includes pre-set compliance boundary constraints. These constraints define insurmountable optimization boundaries during the intelligent dialogue process, ensuring that the self-evolution process does not generate any illegal content. Specifically, the following content is set as core rules that cannot be automatically optimized: legally required content (such as the obligation to disclose debts and explanations of credit impact stipulated in the Civil Code), prohibited expressions required by regulations (such as threatening language, impersonation of public security organs, procuratorates, and courts), a compliance sensitive word library (no optimization scheme may contain words from the sensitive word library), and dialogue time restrictions (cannot be adjusted to legally prohibited time periods). When generating adjustment schemes, the dialogue strategy generation and self-evolution optimization module first performs pre-verification through a compliance rule engine. Any adjustment scheme that violates the core rules will be automatically rejected. In addition, the optimized Prompt or dialogue template will undergo a second verification by an independent compliance review module before going live to ensure that it does not violate relevant laws and regulations. In this application, only schemes that pass both the pre-verification and the second verification by the compliance review module can enter the gray-scale release process.

[0096] As one implementation method, it can be as follows: Figure 2 As shown, the self-evolving intelligent dialogue system provided in this application further includes: a dialogue script strategy management and release control module, wherein the dialogue script strategy management and release control module is used for: Update the new dialogue script strategy with a preset ratio to the data access and case management module, so that the data access and case management module can add the new dialogue script strategy to the dialogue case data with the corresponding preset ratio.

[0097] It is understood that the dialogue strategy management and release control module provided in this application is used to manage the version lifecycle of dialogue strategies, control the release rhythm of strategy updates, and ensure the security and controllability of the self-evolution process. Specifically, dialogue strategy management allows for version management of dialogue script strategies. The self-evolving intelligent dialogue system can maintain a complete configuration snapshot of each strategy version, including the role constraint rules (Prompt) configuration of the intelligent dialogue module Agent, the calling logic configuration of the task skill package, the script template library, and the quality inspection rule configuration. Each version records metadata such as version number, generation time, effective time, expiration time, changes, reasons for changes, and performance metrics. Versions are linked through parent-child relationships to form a complete strategy evolution tree. In this way, operations personnel can view the complete configuration of any version strategy at any time and support comparative analysis of configurations.

[0098] In this embodiment, the updated dialogue script strategy is not rolled out all at once, but rather through a canary release mechanism. Specifically, a certain percentage of the new dialogue script strategy is first updated to the data access and case management module. The data access and case management module then updates the corresponding percentage of dialogue case data based on the new dialogue script strategy. This allows the intelligent dialogue module to communicate with customers based on the newly updated dialogue case data. This canary release mechanism is crucial for ensuring the security of strategy updates. When the self-evolution module generates a new strategy version, it is not immediately rolled out to all cases, but rather gradually increased through canary releases. For example, canary releases can be conducted proportionally (the new strategy is first tested on 5% of cases, gradually increasing to 100%) or by feature (the new strategy is first tested on specific types of cases, such as early overdue cases, and expanded to other types after verifying the effect). During the canary release process, the module continuously monitors key indicators (connection rate, commitment rate, complaint rate, compliance rate). If any abnormality is detected, the canary release is automatically paused and rolled back to the previous version.

[0099] In this embodiment, the dialogue script strategy management and release control module employs an A / B testing framework to support comparative verification of different strategy versions. Specifically, multiple dialogue script strategy versions can be run simultaneously, and similar cases are randomly assigned to different versions to ensure fairness in the comparison. After the testing period ends, the dialogue script strategy management and release control module performs statistical significance testing on the effect data of each group to determine whether the new strategy is significantly better than the old strategy. Only optimization schemes that pass the statistical test will be adopted and enter the full release process.

[0100] In this embodiment, the dialogue script strategy management and release control module also includes a pre-built strategy rollback mechanism to protect against strategy risks. Specifically, when a released dialogue script strategy version encounters problems during operation, the module can roll back to any historical version with a single click. The rollback operation records detailed reasons for the rollback and the responsible party, and the rolled-back version is automatically marked as requiring optimization to prevent the same problem from recurring. The dialogue script strategy management and release control module can also perform emergency freezing of dialogue script strategies, immediately halting all strategy changes and maintaining system stability in the event of systemic risks.

[0101] In the process of managing dialogue script strategies, if changes are needed, a change approval process can be initiated. For minor, low-risk adjustments (such as replacing a few sensitive words or fine-tuning the wording), the dialogue script strategy management and release control module can automatically approve and release the changes. For major, high-risk adjustments (such as modifying core dialogue flows, adjusting pressure intensity, or changing compliance prompts), the module can submit the adjustment plan to the approval queue, where it will be released after dual review and confirmation by operations and compliance personnel. Furthermore, the self-evolving intelligent dialogue system has mandatory human intervention conditions: when the optimization plan involves adjusting the pressure intensity of the dialogue, when quality inspection discovers new violation patterns, when multiple consecutive optimizations fail to produce good results, or when the customer complaint rate exceeds a preset complaint rate threshold, the self-evolving intelligent dialogue system will pause automatic release and forcibly transfer to a human review process. The complaint rate threshold for the approval rules is configurable, allowing operations personnel to adjust the level of automation and the triggering conditions for human intervention based on actual business conditions and regulatory requirements.

[0102] In some possible embodiments, it may be as follows Figure 2 As shown, this self-evolving intelligent dialogue system also includes a monitoring and visualization module, used for: Statistical analysis is performed on the dialogue results data based on intelligent dialogue to obtain the dialogue effect of the target dialogue strategy. The dialogue strategy generation and self-evolution optimization module is also used for: Based on the dialogue effect of the target dialogue strategy, generate a new dialogue strategy.

[0103] The monitoring and visualization module provides operations and management personnel with real-time business monitoring and historical data analysis capabilities, helping them understand the system's operational status, self-evolution effects, and business output. This module includes a real-time monitoring dashboard that displays the system's core operational metrics. These core metrics depend on the application scenario of the self-evolving intelligent dialogue system. Taking intelligent debt collection as an example, these metrics include case processing volume (daily, weekly, monthly), case processing progress (pending, in progress, completed), dialogue efficiency (connection rate, effective communication rate, commitment rate, collection rate), and case distribution (distribution by overdue stage, amount range, and risk level). Quality metrics include quality inspection pass rate, detection rate of inappropriate language, complaint rate, and user satisfaction. Self-evolution metrics include the number of strategy versions, version iteration frequency, gray-scale release progress, and A / B testing results. System metrics include the response time, processing success rate, and resource utilization rate of each module. The real-time monitoring dashboard supports multi-dimensional drill-down, allowing operations personnel to drill down from overall metrics to specific cases or dialogue records.

[0104] Based on the self-evolving intelligent dialogue system provided in the first aspect, in a second aspect, this application provides a self-evolving intelligent dialogue method, which is applicable to any electronic device with self-evolving intelligent dialogue functionality, including but not limited to personal mobile terminals, computers, or servers. Figure 3 As shown, the method includes the following steps: S31. In response to the dialogue task to be processed, a pre-set large language model is invoked to identify the dialogue intent and target dialogue communication channel of the dialogue task to be processed. S32. Select the target dialogue script strategy from the pre-generated dialogue script strategies based on the dialogue intent recognition results; S33. Based on the target dialogue script strategy, initiate intelligent dialogue with the customer corresponding to the target dialogue communication channel and the dialogue task to be processed. S34. Record the intelligent dialogue between the intelligent dialogue module and the customer. Generate and store new dialogue strategies based on the intelligent dialogue between the intelligent dialogue module and the customer.

[0105] It can be understood that the self-evolving intelligent dialogue method provided in this application can be understood in the following stages: Phase 1: Data Collection. After the intelligent dialogue module (Agent) completes a dialogue, the dialogue recording and data storage module automatically collects and stores the complete dialogue data, including the original dialogue content, dialogue turn information, Agent decision-making process, Skill call records, and dialogue results. After cleaning and standardization, the data forms an analyzable structured dataset.

[0106] Phase Two: Quality Inspection and Evaluation. The quality inspection and evaluation module performs multi-dimensional analysis on newly added dialogue data, generating a quality inspection and evaluation report. The report identifies problems in the dialogues (such as compliance risks, missing processes, insufficient wording, and poor results), as well as high-performing wording patterns. The quality inspection results are stored in a structured format for subsequent strategy optimization analysis.

[0107] Phase Three: Strategy Analysis. The dialogue script strategy generation and self-evolution optimization module performs dialogue script strategy analysis periodically (e.g., daily) or triggered (e.g., when a certain number of issues accumulate). This module aggregates quality inspection results over a period of time, statistically analyzes the distribution and trends of issues, and identifies strategy parameters that need optimization. Using a rule engine and large-scale models, this module generates strategies and outputs targeted and risk-controlled dialogue script strategy adjustment plans.

[0108] Phase Four: Strategy Update. After interpretability analysis and risk assessment, the adjusted plan generates a new strategy version. The dialogue script strategy management and release control module implements a phased rollout control for the new version, first verifying its effectiveness on a small number of cases, and then deciding whether to expand the release scope or roll back based on monitoring indicators. The validated version gradually replaces the old version, completing the online strategy update.

[0109] Phase 5: Feedback on Effectiveness. After the new dialogue strategies are implemented, the monitoring and visualization module continuously tracks their performance and collects feedback data. This data serves as input for the next round of self-evolution, forming a continuous optimization cycle. Through multiple iterations, the system learns, adapts, and evolves, making the dialogue strategies increasingly precise and efficient.

[0110] The entire self-evolutionary process is highly automated, with human intervention limited to approving high-risk strategies and handling exceptional situations. This automated closed-loop mechanism enables the system to continuously learn from business data, constantly improving the effectiveness of dialogue while maintaining compliance and security.

[0111] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in this application comply with relevant laws and regulations and do not violate public order and good morals.

[0112] The names of the messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0113] Thirdly, exemplary embodiments of this application also provide an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to cause the electronic device to perform a method according to an embodiment of this application.

[0114] An exemplary embodiment of this application also provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a computer's processor, is used to cause the computer to perform a method according to an embodiment of this application.

[0115] An exemplary embodiment of this application also provides a computer program product, including a computer program, wherein, when executed by a computer's processor, the computer program is used to cause the computer to perform a method according to an embodiment of this application.

[0116] refer to Figure 4The present invention describes a structural block diagram of an electronic device 400 that can serve as a server or client of this application, which is an example of a hardware device that can be applied to various aspects of this application. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the application described and / or claimed herein.

[0117] like Figure 4 As shown, the electronic device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM 402) or a computer program loaded from a storage unit 408 into a random access memory (RAM 403). The RAM 403 may also store various programs and data required for the operation of the electronic device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output interface (I / O interface 405) is also connected to the bus 404.

[0118] Multiple components in electronic device 400 are connected to I / O interface 405, including: input unit 406, output unit 407, storage unit 408, and communication unit 409. Input unit 406 can be any type of device capable of inputting information to electronic device 400. Input unit 406 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of electronic device. Output unit 407 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 408 may include, but is not limited to, disks and optical discs. Communication unit 409 allows electronic device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, wireless communication transceivers, and / or chipsets, such as Bluetooth™ devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0119] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above. For example, in some embodiments, the aforementioned self-evolving intelligent dialogue method can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 400 via ROM 402 and / or communication unit 409. In some embodiments, the computing unit 401 can be configured to perform the aforementioned self-evolving intelligent dialogue method by any other suitable means (e.g., by means of firmware).

[0120] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0121] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0122] As used in this application, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or apparatus (e.g., disk, optical disk, memory, programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including machine-readable media that receive machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0123] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0124] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0125] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other.

Claims

1. A self-evolving intelligent dialogue system, characterized in that, The system includes: an intelligent dialogue module, a dialogue record and data storage module, and a dialogue strategy generation and self-evolution optimization module, wherein: The intelligent dialogue module is used to respond to a dialogue task to be processed, call a pre-set large language model to identify the dialogue intent and target dialogue communication channel of the dialogue task to be processed, select a target dialogue script strategy from the dialogue script strategy generated by the dialogue strategy generation and the dialogue script strategy pre-generated by the self-evolution optimization module according to the dialogue intent identification result, and initiate an intelligent dialogue with the customer corresponding to the dialogue task to be processed according to the target dialogue communication channel based on the target dialogue script strategy. The dialogue record and data storage module is used to record the intelligent dialogue records between the intelligent dialogue module and the customer; The dialogue strategy generation and self-evolution optimization module is used to generate new dialogue script strategies based on the intelligent dialogue records between the intelligent dialogue module and the customer, and store the new script strategies in the dialogue record and data warehouse module.

2. The system according to claim 1, characterized in that, The system also includes a quality inspection and evaluation module, used for: The system performs real-time multi-dimensional quality inspection analysis on the intelligent dialogue records between the intelligent dialogue module and the customer, and generates a structured quality inspection evaluation report. The multi-dimensional quality inspection analysis includes: compliance analysis, execution of the dialogue script strategy, customer experience analysis, and logical analysis.

3. The system according to claim 1, characterized in that, The system further includes: a task skill package invocation module; during the intelligent dialogue process with the customer, the intelligent dialogue module is specifically used for: The large language model is invoked to parse the customer's response content, and the customer's intent is identified based on the response content. The task skill package invocation module is used to respond to the invocation request of the intelligent dialogue module, determine the task skill package that matches the intent representation for each type of intent representation of the customer, and execute the execution logic encapsulated in the task skill package; The intelligent dialogue module is also used to obtain dialogue script strategies that match the customer's intent expression from the dialogue record and data storage module; generate new dialogue scripts according to the execution result of the task skill package and the dialogue script strategies, and output them to the customer.

4. The system according to claim 3, characterized in that, The intelligent dialogue module is an intelligent agent pre-configured based on role constraint rules, and the task skill package is a task software code package pre-encapsulated based on the dialogue interaction processing flow.

5. The system according to claim 2, characterized in that, The dialogue strategy generation and self-evolution optimization module is also used for: Based on the structured quality inspection and evaluation report, the role constraint rules of the intelligent dialogue module are updated in conjunction with the intelligent dialogue records between the intelligent dialogue module and the customer, and the intelligent dialogue module is updated based on the new role constraint rules.

6. The system according to claim 1, characterized in that, The system also includes a data access and case management module, used for: The system uses a pre-defined data interface to obtain dialogue case data from a pre-defined business platform, generates pending dialogue tasks based on the dialogue case data, and assigns priorities to the pending dialogue tasks; wherein, the dialogue case data records the communication method of the target customer and the basic information of the target customer; After the intelligent dialogue module ends its intelligent dialogue with the customer, it sends the dialogue result data back to the business platform through the preset data interface so that the business platform can update the dialogue case data.

7. The system according to claim 6, characterized in that, The system also includes: a dialogue script strategy management and release control module, used for: The new dialogue script strategy is updated to the data access and case management module at a preset ratio, so that the data access and case management module adds the new dialogue script strategy to the dialogue case data at the corresponding preset ratio.

8. The system according to claim 7, characterized in that, The system also includes a monitoring and visualization module, used for: Statistical analysis is performed on the dialogue result data of the intelligent dialogue to obtain the dialogue effect of the target dialogue script strategy. The dialogue strategy generation and self-evolution optimization module is also used for: Based on the dialogue effect of the target dialogue strategy, a new dialogue strategy is generated.

9. A self-evolving intelligent dialogue method, characterized in that, The method is applied to the system as described in any one of claims 1 to 8, and the method includes: In response to a pending dialogue task, a pre-set large language model is invoked to identify the dialogue intent and target dialogue communication channel of the pending dialogue task. Based on the dialogue intent identification result, a target dialogue script strategy is selected from the pre-generated dialogue script strategy. Based on the target dialogue script strategy, an intelligent dialogue is initiated with the customer corresponding to the pending dialogue task according to the target dialogue communication channel. The system records the intelligent dialogue between the intelligent dialogue module and the customer, and generates and stores new dialogue strategies based on these records.

10. An electronic device, characterized in that, The electronic device includes: a processor and a memory storing a program; wherein the program includes instructions that, when executed by the processor, cause the processor to perform the method according to claim 9.

Images