Response method and system

Abstract

The invention provides a response method and system, the response system comprises a route planning task module and a response task module, the route planning task module comprises a route skill unit, the route skill unit is used for routing a response task to a target skill unit in a plurality of response skill units, the response task is obtained based on a query text, and the target skill unit is used for receiving the response task. The response skill capability of the target response skill unit is matched with the response skill capability required by the response task; the response task module comprises a plurality of response skill units, the plurality of response skill units respectively execute response tasks of different response skill requirements by the user, and the target response skill unit executes the response task to generate a response result corresponding to the query text; wherein each of the routing skill unit and the response skill unit comprises one or more intelligent agents. According to the response system provided by the invention, skills are taken as organization cooperation units, the complexity of agent tasks is simplified, and the function expandability of the system is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence (AI) technology, and in particular to a response method and system. Background Technology

[0002] The rapid development of large language models (LLMs) has brought new directions to the recommendation field, such as large model-enhanced recommendation systems, large recommendation models, and large model-driven recommendation agents (LLM-powered agents). Among them, LLM-driven agents have become a new paradigm for interactive recommendation, as seen in Amazon Rufus and Shopify AI. LLM-driven recommendation agents can not only accurately capture users' real-time needs but also provide personalized recommendations and responses based on users' past preferences and real-time needs, enhancing the user's personalized experience. Therefore, interactive recommendation agents are an important direction for achieving personalized and intelligent services and improving product competitiveness.

[0003] Building and improving the ability of large-scale model-driven recommendation agents presents challenges in the collaborative construction of multi-agent systems. Summary of the Invention

[0004] The embodiments of this application provide a response method and system that uses skills as the organizational and collaborative unit, which simplifies the complexity of intelligent agent tasks and improves the functional scalability of the system.

[0005] In a first aspect, this application provides a response system including a route planning task module and a response task module. The route planning task module includes a route skill unit, which routes the response task to a target skill unit among multiple response skill units. The response task is obtained based on a query text, and the response skill capabilities of the target response skill unit match the response skill capabilities required by the response task. The response task module includes multiple response skill units, which respectively execute response tasks with different response skill requirements. The target response skill unit executes the response task and generates a response result corresponding to the query text. Both the route skill unit and the response skill unit include one or more intelligent agents.

[0006] The response system provided in this application uses skills as the unit of organization and collaboration, which simplifies the complexity of agent tasks and improves the system's functional scalability.

[0007] In one possible implementation, the multiple response skill units include one or more of the following: a chatty response skill unit, a knowledge acquisition response skill unit, a recommendation response skill unit, and a knowledge and recommendation response skill unit; wherein, the chatty response skill unit is used to handle the response task with chatty needs, the knowledge acquisition response skill unit is used to handle the response task with knowledge acquisition needs, the recommendation response skill unit is used to handle the response task with recommendation needs, and the knowledge and recommendation response skill unit is used to handle the response task with both knowledge acquisition and recommendation needs.

[0008] By setting up multiple response skill units in the response task module, different response skill units within these units handle response tasks that meet different user needs, thus addressing diverse user requirements. For example, in practical applications, users may have diverse needs such as casual conversation, knowledge acquisition, recommendation requests, and a combination of knowledge and recommendations, thereby improving the user experience.

[0009] In another possible implementation, the chatter response skill unit includes a general chatter agent and a domain chatter agent. The general chatter agent handles response tasks with general chatter needs, and the domain chatter agent handles response tasks with domain chatter needs. And / or, the knowledge acquisition response skill unit includes a search planning agent and an answer summarizing agent. The search planning agent is configured with a web search tool and uses the web search tool to search for knowledge information based on the query text. The answer summarizing agent summarizes the retrieved knowledge information and generates an answer corresponding to the query text. And / or, the recommendation response skill unit includes... It includes a state-tracking agent, a recommendation agent, and an interaction agent. The recommendation agent is equipped with search / recommendation tools. The state-tracking agent tracks the current user's information needs. When the user's information needs are clear, it calls the recommendation agent to generate a recommendation list. When the current user's information needs are unclear, it generates clarification questions to ask the user to obtain clear information needs. The recommendation agent calls the search / recommendation tools to generate a recommendation list based on the user's information needs. The interaction agent generates explanations and reasons for the recommendations based on the query text and the recommendation list. And / or, the knowledge and recommendation response skill unit includes a search planning agent and an answer summary agent.

[0010] This scheme improves the quality of recommendation tasks by having a state-tracking agent, a recommendation agent, and an interaction agent work together to complete the recommendation task. When the user intent expressed in the query text is not very clear, the state-tracking agent can generate clarification questions to ask follow-up questions in order to clarify the user intent and ensure the accuracy of the recommendation results.

[0011] In another possible implementation, the state-tracking agent uses a slot-filling approach to track the state of user needs. For example, based on the query text and user preferences, corresponding keywords are filled into the appropriate slots in the slot template. Optionally, a single round of interaction can have multiple slot templates to capture various user needs. Using a slot-filling approach to track the state of user needs makes it easier to organize the parameters called by the recommendation tool, improving the accuracy of the model's tool calls, and thus improving the accuracy of the recommendation list generated by the recommendation agent.

[0012] In another possible implementation, the routing skill unit includes a routing agent that performs one or more of the following operations on the query text: domain recognition, user intent recognition, sentiment recognition, and task decomposition.

[0013] By identifying the user intent expressed in the query text through a routing agent, the requirements of the response task are matched with multiple response skill units in the response task module. The target response skill unit that matches the user intent is obtained, and the response task is processed by the target response skill unit to meet the user's needs and improve the user experience.

[0014] In another possible implementation, the response system provided in this application further includes a preprocessing task module for preprocessing the query text. The preprocessing module includes one or more of the following: a user profiling skill unit, a homepage query text recommendation skill unit, a memory extraction skill unit, a memory retrieval skill unit, a sensitive word detection skill unit, and a typo correction skill unit; wherein the user profiling skill unit, the homepage query text recommendation skill unit, the memory extraction skill unit, the memory retrieval skill unit, the sensitive word detection skill unit, and the typo correction skill unit each include one or more intelligent agents.

[0015] The preprocessing task module in this solution includes multiple system skill units to meet diverse system functional requirements, such as user profiling, homepage query text recommendation, memory extraction, memory retrieval, sensitive word detection, and typo correction.

[0016] In another possible implementation, the user profiling skill unit includes a user profiling agent, which generates a user preference profile based on the user's interaction history; and / or, the homepage query text recommendation skill unit includes a homepage query text recommendation agent, which generates homepage query text recommendations based on the user preference profile; and / or, the memory extraction skill unit includes a memory extraction agent, which is configured with a memory extraction tool and uses the memory extraction tool to extract memory knowledge data from the generated dialogue, storing the memory knowledge data in the user interaction database; the memory retrieval skill unit... The unit includes a memory retrieval agent, which is equipped with a memory search tool and is used to retrieve memory knowledge data related to the query text from the user interaction database; and / or, a sensitive word detection skill unit includes a sensitive word detection agent, which is equipped with a sensitive word search tool and is used to call the sensitive word detection tool to detect sensitive words in the query text, and if a sensitive word is detected, a security recovery is generated; and / or, a typo correction skill unit includes an error correction agent, which is equipped with an entity search tool and is used to call the entity search tool to correct errors in the query text.

[0017] In another possible implementation, the preprocessing task module is also used to send a response request to the routing planning task module, the response request carrying the query text and the memory data retrieved by the memory retrieval skill unit; the routing planning task module is also used to perform intent recognition on the query text, and when the recognized user intent is unclear, rewrite the query text based on the memory data.

[0018] In another possible implementation, the preprocessing task module is also used to send a response request to the routing planning task module, the response request carrying the query text and the memory data retrieved by the memory retrieval skill unit; the routing planning task module is also used to perform intent recognition on the query text and route the response request to the target response skill unit based on the intent recognition; the response task module is also used to call the target response skill unit to generate the response result corresponding to the query text based on the query text and the memory data.

[0019] In another possible implementation, the response system provided in this application further includes a display module and a memory update module, wherein the display module is used to display the retrieved personalized memory data of the user; and the memory update module is used to update the personalized memory data according to the user's operation on the displayed personalized memory data.

[0020] This application enables online / offline extraction and updating of memories. Memories include the activation, extraction, storage, updating, retrieval, and use of user-personalized memories, enabling personalized query rewriting and personalized response generation, thereby enhancing the user's personalized experience.

[0021] In another possible implementation, the route planning task module, response task module, and preprocessing task module all support the expansion of skill units based on requirements, thereby improving the system's functional scalability.

[0022] Secondly, this application also provides a response method, including inputting query text into a routing planning task module, so that the routing skill unit in the routing planning task module routes the response task to a target skill unit among multiple response skill units, the response task is obtained based on the query text, and the response skill capabilities of the target response skill unit match the response skill capabilities required by the response task; calling the target skill unit in the response task module to process the response task and generate a response result corresponding to the query text, wherein the routing skill unit and the response skill unit each include one or more intelligent agents.

[0023] In one possible implementation, the multiple response skill units include one or more of the following: a chatty response skill unit, a knowledge acquisition response skill unit, a recommendation response skill unit, and a knowledge and recommendation response skill unit; wherein, the chatty response skill unit is used to handle the response task with chatty needs, the knowledge acquisition response skill unit is used to handle the response task with knowledge acquisition needs, the recommendation response skill unit is used to handle the response task with recommendation needs, and the knowledge and recommendation response skill unit is used to handle the response task with both knowledge acquisition and recommendation needs.

[0024] In another possible implementation, the chatter response skill unit includes a general chatter agent and a domain chatter agent. The general chatter agent handles response tasks with general chatter needs, and the domain chatter agent handles response tasks with domain chatter needs. And / or, the knowledge acquisition response skill unit includes a search planning agent and an answer summarizing agent. The search planning agent is configured with a web search tool and uses the web search tool to search for knowledge information based on the query text. The answer summarizing agent summarizes the retrieved knowledge information and generates an answer corresponding to the query text. And / or, the recommendation response skill unit includes... It includes a state-tracking agent, a recommendation agent, and an interaction agent. The recommendation agent is equipped with search / recommendation tools. The state-tracking agent tracks the current user's information needs. When the user's information needs are clear, it calls the recommendation agent to generate a recommendation list. When the current user's information needs are unclear, it generates clarification questions to ask the user to obtain clear information needs. The recommendation agent calls the search / recommendation tools to generate a recommendation list based on the user's information needs. The interaction agent generates explanations and reasons for the recommendations based on the query text and the recommendation list. And / or, the knowledge and recommendation response skill unit includes a search planning agent and an answer summary agent.

[0025] In another possible implementation, the state-tracking agent uses a slot-filling method to track the state of user demand information.

[0026] In another possible implementation, the routing skill unit includes a routing agent that performs one or more of the following operations on the query text: domain recognition, user intent recognition, sentiment recognition, and task decomposition.

[0027] In another possible implementation, the response method provided in this application further includes preprocessing the query text before routing the response task to the target skill unit among multiple response skill units. The preprocessing module includes one or more of the following: user profiling skill unit, homepage query text recommendation skill unit, memory extraction skill unit, memory retrieval skill unit, sensitive word detection skill unit, and misspelling correction skill unit; wherein the user profiling skill unit, homepage query text recommendation skill unit, memory extraction skill unit, memory retrieval skill unit, sensitive word detection skill unit, and misspelling correction skill unit each include one or more intelligent agents.

[0028] In another possible implementation, the user profiling skill unit includes a user profiling agent, which generates a user preference profile based on the user's interaction history; and / or, the homepage query text recommendation skill unit includes a homepage query text recommendation agent, which generates homepage query text recommendations based on the user preference profile; and / or, the memory extraction skill unit includes a memory extraction agent, which is configured with a memory extraction tool and uses the memory extraction tool to extract memory knowledge data from the generated dialogue, storing the memory knowledge data in the user interaction database; the memory retrieval skill unit... The unit includes a memory retrieval agent, which is equipped with a memory search tool and is used to retrieve memory knowledge data related to the query text from the user interaction database; and / or, a sensitive word detection skill unit includes a sensitive word detection agent, which is equipped with a sensitive word search tool and is used to call the sensitive word detection tool to detect sensitive words in the query text, and if a sensitive word is detected, a security recovery is generated; and / or, a typo correction skill unit includes an error correction agent, which is equipped with an entity search tool and is used to call the entity search tool to correct errors in the query text.

[0029] In another possible implementation, the response method provided in this application, after preprocessing the query text, further includes sending a response request to the route planning task module. The response request carries the query text and memory data retrieved by the memory retrieval skill unit. The route planning task module performs intent recognition on the query text, and when the recognized user intent is unclear, it rewrites the query text based on the memory data.

[0030] In another possible implementation, the response method provided in this application, after preprocessing the query text, further includes sending a response request to the routing planning task module, the response request carrying the query text and memory data retrieved by the memory retrieval skill unit; the routing planning task module performs intent recognition on the query text and routes the response request to the target response skill unit based on the intent recognition; the response task module calls the target response skill unit to generate the response result corresponding to the query text based on the query text and memory data.

[0031] In another possible implementation, the response method provided in this application further includes displaying the retrieved personalized memory data of the user; and updating the personalized memory data based on the user's actions on the displayed personalized memory data.

[0032] In another possible implementation, the route planning task module, response task module, and preprocessing task module all support the expansion of skill units based on requirements, thereby improving the system's functional scalability.

[0033] Thirdly, embodiments of this application provide a computing device, including a memory and a processor, wherein the memory stores instructions that, when executed by the processor, cause the method described in the first aspect or any possible implementation of the first aspect to be implemented.

[0034] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the method described in the first aspect or any possible implementation thereof to be implemented.

[0035] Fifthly, embodiments of this application also provide a computer program or computer program product, which includes instructions that, when executed, cause a computer to perform the method described in the first aspect or any possible implementation thereof.

[0036] In a sixth aspect, embodiments of this application also provide a chip including at least one processor and a communication interface, the processor being configured to perform the method described in the first aspect or any possible implementation thereof.

[0037] It is understandable that the beneficial effects of the second to sixth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here for the sake of brevity. Attached Figure Description

[0038] Figure 1 A schematic diagram of a system architecture is shown;

[0039] Figure 2 This paper illustrates an overall framework diagram of an interactive recommendation agent system provided in an embodiment of this application.

[0040] Figure 3 A schematic diagram illustrating the generation of slot filling and clarification issues within state tracking is provided.

[0041] Figure 4 A flowchart illustrating a response method provided in an embodiment of this application;

[0042] Figure 5 This shows an example of the retrieved personalized memory data of a user being displayed to the user;

[0043] Figure 6 A schematic diagram illustrating the signaling interaction between a user and a recommendation agent system in a music recommendation scenario is shown.

[0044] Figure 7 This diagram illustrates the dialogue effect of an interactive recommendation system agent system applying the response method provided in the embodiments of this application.

[0045] Figure 8 This illustration shows the effect of personalized recovery with enhanced memory in the interactive recommendation system intelligent agent system provided in the embodiments of this application;

[0046] Figure 9 A schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0047] The term "and / or" used in this article describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The symbol " / " in this article indicates that the related objects are in an "or" relationship; for example, A / B means A or B.

[0048] The terms "first" and "second," etc., used in the specification and claims herein are used to distinguish different objects, not to describe a specific order of objects. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same properties in the description of embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a list of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such processes, methods, systems, products, or apparatus.

[0049] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0050] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0051] Building and improving the capabilities of large-model-driven recommendation agents presents challenges in multi-agent collaborative construction. On the one hand, user needs are diverse and varied, requiring systems to adapt to these changing demands and provide accurate responses. On the other hand, there is a lack of guiding methods for building multi-agent collaborative systems, and single-agent and simple multi-agent systems, such as InteRecAgent, MACRec, MACRS, and RAH, cannot address the diverse and varied user needs. Specifically, single-agent systems require high-capacity large models, perform complex tasks, are difficult to extend functionality, and lack personalization capabilities. Simple multi-agent systems, while offering diverse collaborative methods, are inefficient, computationally expensive, and lack the ability to respond to diverse user needs. Overall, current large-model-driven recommendation agents suffer from the following problems: high requirements for large model capabilities, poor functional scalability, lack of support for diverse user needs, and lack of utilization of personalized preference memory.

[0052] For example, one relevant technical solution is the single-agent solution driven by a large language model. This solution autonomously solves interactive recommendation problems based on a single large model. The agent can decompose the task into multiple sub-tasks, and each sub-task is completed by this single agent. Since the single-agent system contains only one large language model agent, all tasks, including interacting with the user, understanding user needs, identifying user tasks, task decomposition, task execution (tool invocation), reflecting on task execution results, and improvement, must be undertaken by this single large language model agent. A representative single-agent recommendation solution is InteRecAgent. Specifically, InteRecAgent implements a single-agent solution; it adopts a "First plan, then Execute" planning approach, planning the invocation of different tools one step at a time based on user requests, and then executing the tool's plan one step at a time. This approach saves significantly more time than the plan-step-plan-step strategy and meets the latency requirements of recommendation scenarios.

[0053] However, this approach has the following drawbacks: the single-agent approach has high requirements for the capabilities of large models and the execution of tasks is complex: when planning the steps of tool invocation, it is necessary to consider not only the invocation of multiple tools, but also the current dialogue history, user preferences, etc.

[0054] Difficulty in diversifying functionalities: It is hard to expand to a wide range of practical scenarios, such as casual conversation, knowledge acquisition, recommendations, user interest status tracking, secure replies, typo correction, and initial conversation query recommendations.

[0055] Therefore, it is evident that single-agent solutions use a single large language model agent to handle complex tasks in recommendation scenarios (which often involve many stages). The large model has a heavy task burden, and the overall difficulty of the task is also relatively high. At the same time, single-agent solutions are difficult to extend to rich practical scenario functions.

[0056] The second related technical solution is a multi-agent scheme driven by a large language model. Due to the limitations of the single-agent framework of large language models, more and more researchers are turning their attention to multi-agent schemes. Multi-agent systems driven by large language models perform recommendations by assigning different agents to handle the sub-tasks of each stage of the recommendation process, working together to complete the recommendation task. The advantages of this are that it minimizes the complexity of the sub-tasks handled by a single agent, reducing the demands on the large model's capabilities; furthermore, since sub-tasks can be completed by different agents, the number of sub-tasks or functions can be increased by increasing the number of agents. Currently, representative multi-agent recommendation schemes include the MACRec scheme, the MACRS scheme, and the RAH scheme. The MACRec scheme is specifically implemented by using a multi-agent approach, where each agent focuses on a specific recommendation task. Different combinations of agents were used for different recommendation tasks: (1) Conversational recommendation: Agents such as Manager (task distribution + reflection), Interpreter (query rewriting), and Searcher (search / recommendation tool invocation) were used; (2) User product rating prediction used agents such as Manager, User Analyst, and Item Analyst.

[0057] However, this solution has the following drawbacks: lack of user interest state tracking: the conversational recommendation solution does not consider the situation where the user's requested information is insufficient, and has no way to obtain accurate information from the user.

[0058] The collaboration method is too simplistic: it relies too heavily on the capabilities of the Manager as a single intelligent agent, with both input and output of information generated by the Manager.

[0059] The MACRS scheme is implemented as follows: a multi-agent approach is adopted, in which three different responding agents generate three responses in parallel based on the user's request, and a planning agent integrates the results of the three different responses to generate the final response: Asking Responder Agent: generates clarifying questions; Recommending Responder Agent: generates recommendation tool calls; Chit-Chatting Responder Agent: engages in small talk; Planner Agent: plans and selects the most appropriate response from the responses of the three tasks.

[0060] However, this scheme has the following drawbacks: High computational cost: The planning agent ultimately only uses the response from one of the three agents, wasting the calls to the other two models. Monotonous multi-agent cooperation: The Planner only selects the response from one agent, failing to leverage the advantages of multiple agents, resulting in limited cooperation. Poor scalability: The PlannerAgent's input is chaotic, requiring consideration of three different behavioral responses; as the scenario expands, the PlannerAgent's input becomes even more complex.

[0061] The RAH scheme is implemented as follows: a multi-agent approach is adopted, with the agent acting as an assistant between the recommendation system and the user: filtering products, generating fictitious recommendation data labels, and strengthening the recommendation model; the multi-agent approach determines the quality of products in the recommendation candidate set: Perceive Agent: identifies user reviews as labels; Learn Agent: analyzes why users like / dislike recommended products; Act Agent: filters results based on the traditional recommendation model; Reflect Agent: updates user preferences.

[0062] However, this solution has the following drawbacks: it lacks the ability to respond to diverse user needs. It is also inefficient, as each product requires excessively large model inference before being filtered, resulting in low recommendation efficiency.

[0063] In summary, both single-agent and multi-agent solutions have numerous limitations in the relevant technical approaches:

[0064] Large-scale models have high capability requirements: The relevant technical solutions involve Manager Agent or Planner Agent, which needs to plan tasks and integrate content for responses. The tasks are highly complex and require high capabilities from the large-scale models.

[0065] Poor functional scalability: The relevant technical solution modules are fixed, making it difficult to flexibly expand the solution and add functional modules to meet the needs of different scenarios.

[0066] Lack of support for diverse user needs: User needs are complex and ever-changing, and related technical solutions cannot guarantee that they will meet the rich and varied needs of users.

[0067] Lack of utilization of personalized preference memory: Related technical solutions lack consideration for personalized preference memory information, such as memory retrieval, management, searching, and use.

[0068] In view of this, embodiments of this application provide a response method and system that uses skills as the organizational and collaborative unit, simplifies the complexity of intelligent agent tasks, and improves the functional scalability of the system.

[0069] The response method and system provided in this application can be applied to scenarios in the field of intelligent response technology, such as recommendation scenarios, especially interactive recommendation scenarios. The following describes the response method and system provided in this application in detail, taking the application of the response method and system in an interactive recommendation scenario as an example. It is understood that when the response method and system provided in this application are applied to an interactive recommendation scenario, the response method provided in this application can be called a recommendation method or an interactive recommendation method; the response system provided in this application can be called a recommendation system, or more specifically, an interactive recommendation system, or an interactive recommendation intelligent agent system.

[0070] To facilitate understanding of the response method and apparatus provided in the embodiments of this application, some technical terms involved in the embodiments of this application will be briefly explained below.

[0071] Large language models (LLMs) are artificial intelligence systems trained on massive amounts of text data, capable of understanding and generating natural language. They learn language structure, semantics, and context through deep learning algorithms. Examples include OpenAI's GPT series and Meta's Llama series, and these models are widely used in dialogue systems, text generation, and translation.

[0072] Large language model-powered agents (LLM-powered agents) are applications that leverage large language models to perform complex tasks. These agents execute tasks by combining large language models with key modules such as planning and memory. In building such agents, the LLM acts as the control center or "brain," managing a series of operations required to complete tasks or respond to user requests. The construction of such agents relies on key modules such as planning, memory, and tool usage.

[0073] Large language model-powered recommender agent (LLM-powered recommender agent): This is an application that utilizes large language models for recommendation tasks.

[0074] The specific implementation of the training data acquisition method provided in this application embodiment is described in detail below with reference to the accompanying drawings and embodiments.

[0075] Figure 1A schematic diagram of a system architecture is shown. The system includes a terminal 110, a server 130, and a network 120 that communicatively connects the terminal 110 and the server 130. The server 130 may include one or more servers (…). Figure 1 (This example uses only one server) Server 130 can provide interactive recommendation tasks for one or more terminals 110 to meet the diverse needs of users.

[0076] In some embodiments, server 130 may also provide other services or software applications, including both non-virtual and virtual environments. In some embodiments, these services may be provided as web-based services or cloud services, such as to end users under a Software as a Service (SaaS) model.

[0077] exist Figure 1 In the system shown, server 130 may include one or more components that implement the functions performed by server 130. These components may include software components, hardware components, or combinations thereof that can be executed by one or more processors. A user of operating terminal 110 may sequentially interact with server 130 and utilize the services provided by these components using one or more client applications. It should be understood that... Figure 1 This is merely one example of a system for implementing the various methods described in the embodiments of this application; other different configurations are possible, and the embodiments of this application do not limit this.

[0078] The terminal 110 may have an interactive recommendation-related application (such as a mobile smart recommendation assistant) installed on it, or a recommendation-related webpage opened. The aforementioned application and webpage can provide an interactive interface. Optionally, the interactive interface includes a text box. The terminal 110 can receive a query entered by the user in the text box and send the query to the service 130. The server 130 can perform a response task based on the received query using the response method provided in the embodiments of this application.

[0079] Terminal 110 may include various types of computer devices, such as portable handheld devices (e.g., smartphones), general-purpose computers (e.g., personal computers or laptops), workstation computers, wearable devices, etc. These computer devices may run various types and versions of software applications and operating systems, such as Microsoft Windows, Apple iOS, UNIX-like operating systems, Linux or Linux-like operating systems (e.g., Google Chrome OS), Harmony OS; or various mobile operating systems, such as Microsoft Windows Mobile OS, Windowsphone, Android, iPhone OS, HarmonyOS NEXT, etc. Portable handheld devices may include cellular phones, smartphones, tablets, personal digital assistants (PDAs), etc. Wearable devices may include head-mounted displays and smartwatches, etc.

[0080] Network 120 can be any type of network well known to those skilled in the art, and can use any of a variety of available protocols, including but not limited to TCP / IP, SNA, IPX, 3G, 4G, 5G, etc., to support data communication. For example, one or more networks 120 can be a local area network (LAN), an Ethernet-based network, a token ring network, a wide area network (WAN), the Internet, a virtual network, a virtual private network (VPN), an intranet, an extranet, the public switched telephone network (PSTN), an infrared network, a wireless network (e.g., Bluetooth, Wi-Fi), and / or any combination of these and / or other networks.

[0081] Server 130 may include one or more general-purpose computers, special-purpose server computers (e.g., personal computer servers, UNIX servers, terminal servers), blade servers, mainframe computers, server clusters, or any other suitable arrangement or combination. Server 330 may include one or more virtual machines running a virtual operating system, or other computing architectures involving virtualization, such as one or more flexible pools of logical storage devices that can be virtualized to maintain virtual storage devices for the server. In various embodiments, server 130 may run one or more services or software applications that provide the functionality described below.

[0082] In some implementations, server 130 can be a server for a distributed system or a server integrated with blockchain. Server 130 can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. A cloud server is a host product in the cloud computing service system, designed to address the shortcomings of traditional physical hosts and virtual private servers (VPS) services, such as high management difficulty and weak business scalability.

[0083] It should be noted that in some other examples, the terminal 110 may also execute the response method provided in the embodiments of this application on its own, providing users with response skills that meet the diverse needs of users, without the need for the cooperation of the server 130. Alternatively, the server 130 may also execute the response method provided in the embodiments of this application on its own, providing users with response skills that meet the diverse needs of users, without the need for the cooperation of the terminal 110. This application does not specifically limit this.

[0084] Figure 2 This diagram illustrates the overall framework of an interactive recommendation agent system provided in an embodiment of this application. Figure 2 As shown in the embodiments of this application, the interactive recommendation intelligent agent system mainly consists of three parts: preprocessing, route planning, and action options. The preprocessing part involves the pre-requirement system skills for task execution, including user profiling, homepage query recommendation, memory extraction, memory retrieval, sensitive word detection, and typo correction. The route planning part, as a key component of the system, is mainly responsible for domain identification, user intent recognition, sentiment recognition, and task decomposition. The action options part (also known as the response task part) mainly involves the basic skills and combinations that meet user information needs, such as chit chat, knowledge acquisition, recommendation, and knowledge & recommendation.

[0085] It should be noted that the three parts of preprocessing, route planning, and behavior options can also be called the preprocessing task module, the route planning task module, and the response task module, or the preprocessing task subsystem, the route planning task subsystem, and the response task subsystem.

[0086] Depend on Figure 2As can be seen, the embodiments of this application design an interactive recommendation agent system with hierarchical skill collaboration, namely, a preprocessing task module, a routing planning task module, and a response task module. This hierarchical skill collaboration essentially decomposes a complex system into simple task modules. Through hierarchical collaboration, the execution efficiency and accuracy of tasks can be greatly improved, while also reducing the capability requirements of large language models to a certain extent.

[0087] See also Figure 2 This application's embodiments design a series of skills and combinations, and perform layered collaboration according to a certain execution order. For example... Figure 2 As shown, the preprocessing task module includes a user profiling skill unit, a homepage query recommendation skill unit, a memory extraction skill unit, a memory retrieval skill unit, a sensitive word detection skill unit, and a typo correction skill unit. The user profiling skill unit includes a User Profile Agent that generates a user profile from user interaction logs. For example, if the user is "Zhang San," the user profiling skill unit calls the User Profile Agent to generate a user profile based on "Zhang San's" interaction logs. For instance, the generated user profile for "Zhang San" might be: white-collar worker, 34 years old, male, high-income earner, enjoys sports, fitness, music, and high-tech products, etc., so that query recommendations matching Zhang San's preferences can be generated.

[0088] The Homepage Query Recommendation Skill Unit includes a Homepage Query Recommendation Agent that generates homepage query recommendations based on user profiles. In other words, the Homepage Query Recommendation Skill Unit generates query recommendations based on user profiles. For example, when user "Zhang San" opens the interactive recommendation client, the Homepage Query Recommendation Skill Unit calls the Homepage Query Recommendation Agent to generate homepage query recommendations based on "Zhang San's" user profile. For instance, the interactive recommendation dialog box might suggest several queries that "Zhang San" might ask, allowing the user to choose. If a suggested query matches Zhang San's question, Zhang San only needs to click to select and input the query, improving the user experience. For example, the Homepage Query Recommendation Skill Unit might generate homepage query recommendations based on Zhang San's user profile such as: "When will the Huawei Mate 70 be released?", "Please recommend new energy vehicles with good intelligent driving capabilities", and "Please recommend good music by Taylor Swift", etc.

[0089] The memory extraction skill unit includes a Memory Extractor Agent. Optionally, the Memory Extractor Agent is configured with a Memory Extractor Tool. The Memory Extractor Agent uses the Memory Extractor Tool to extract memories and store them in the user interaction database. For example, if the current conversation includes information that the user likes Taylor Swift, the Memory Extractor Agent calls the Memory Extractor Tool to extract the memory data of the user liking Taylor Swift from the current conversation and stores this memory data in the user interaction database. Optionally, the user interaction database may include a user memory data sub-database. The user memory data extracted by the Memory Extractor Agent can be stored in the user memory data sub-database, thus increasing the efficiency of memory retrieval.

[0090] The memory retrieval skill unit includes a Memory Retriever Agent. Optionally, the Memory Retriever Agent is equipped with a Memory Search Tool. The Memory Retriever Agent can use the Memory Search Tool to retrieve memories related to the current query from the user interaction database, enabling personalized query rewriting and personalized response generation based on these memories, thus enhancing the user's personalized experience. For example, when the user's intent is unclear, user memories can be used to rewrite the query.

[0091] The sensitive word detection skill unit includes a Sensitive Word Detection Agent, which is equipped with a Sensitive Keywords Search Tool. This agent can use the tool to detect sensitive words in queries and provide a safe response if any are found. For example, if the agent detects a sensitive word in a user's query, it generates a security message, "I cannot answer your question," ensuring that the generated response does not involve sensitive topics and increasing security.

[0092] The typo correction skill unit includes a correction agent, which is equipped with an entity search tool. The correction agent can use the entity search tool to obtain the correct entities related to the query and correct them. This supports the correction of typos in user input queries, improving the accuracy of user intent recognition in subsequent routing planning modules and the accuracy of responses generated by response modules.

[0093] The route planning task module includes a route agent, which performs domain recognition, user intent recognition, sentiment recognition, and task decomposition on the query text. In other words, for the route planning task module, this embodiment designs a route agent to perform the operations of domain recognition, user intent recognition, sentiment recognition, and task decomposition.

[0094] Of course, in some other examples, the routing agent may perform fewer or more operations. For example, the routing agent may only perform user intent recognition and task decomposition operations. The specific operations implemented by the routing agent can be set according to actual needs, and this application does not make specific limitations on this.

[0095] The response task module includes a casual conversation response skill unit, a knowledge acquisition response skill unit, a recommendation response skill unit, and a knowledge and recommendation response skill unit. The casual conversation response skill unit is used to engage in casual conversation with users and meet their casual conversation needs. This unit includes a General Chit Chat Agent and a Domain Chit Chat Agent, whose persona is defined by setting different prompts, showcasing response skills and conversational techniques specific to different domains.

[0096] The knowledge acquisition and response skill unit includes a search planning agent and an answer composer agent. The search planning agent is equipped with a web search tool (such as PetalSearch) and can use it to acquire knowledge information related to the query. The answer composer agent summarizes the acquired knowledge information to answer the knowledge acquisition questions involved in the query. In this way, the knowledge acquisition task is completed through the collaboration of the search planning agent and the answer composer agent.

[0097] The recommendation skill (Rec.) comprises a State Tracking Agent, a Recommender Agent, and an Interactor Agent. The State Tracking Agent tracks the current state of the user's information needs. The Recommender Agent, equipped with a Search Tool / Recommendation Tool, generates a Recommendation List. The Interactor Agent provides explanations and justifications for the recommendations based on the query and the generated list. These three agents work together to fulfill the user's recommendation needs. For example, the State Tracking Agent processes the current query while tracking the user's information needs. When the user's information needs are clear (Sufficient Information), the Recommender Agent and its Search Tool provide a list of recommended items through a slot-filling mechanism. Finally, the Interactor Agent provides explanations and justifications for the recommendations based on the user's query and the generated list, thus satisfying the user's information needs. If the user's information needs are unclear, the clarification questions generated by the state tracking agent can be selected by the interactive agent and presented to the user as appropriate clarification questions, so that the user can provide more specific information.

[0098] Optionally, the user information demand status is divided into demand definition (i.e., sufficient information) and demand definition (i.e., insufficient information). The definition method is whether the user input information can fully fill the information slots in the slot template, such as song name, artist, genre, language, filter fields, etc. The large language model can make its own judgment by setting an information sufficiency threshold.

[0099] Figure 3 A schematic diagram illustrating the generation of slot filling and clarification issues within state tracking is provided. Figure 3 As shown, taking music recommendation as an example, based on the query text and user preferences, corresponding keywords can be filled into the corresponding slots in the slot template, for example... Figure 3In the query, Query 1 is "What are some good English songs?". The user preference / profile information obtained through the preprocessing task module is: 1. Likes Taylor Swift's songs. 2. Frequently listens to country music. 3. Only listens to English songs. 4. Dislikes jazz music. The slot template is: Song Name: ""; Artist: ""; Genre: ""; Language: ""; Filter Field: Genre: "". Based on the query text Query 1 and user preferences, the state tracking agent can perform slot filling on the template: Song Name: ""; Artist: "Taylor Swift"; Genre: ""; Language: "English"; Filter Field: Genre: "". At this point, some slots in the template are missing, indicating that the user's intent is not very clear. The state tracking agent generates several follow-up questions to address the missing information, such as "1. Do you want to listen to Taylor Swift's songs? 2. Which English country music songs do you want to listen to?", and presents these questions to the user to provide more specific information. If the slot filling task information is sufficient, or if all slots in the slot template are filled, then the slot information is extracted and sent to the recommendation agent. The recommendation agent determines the recommendation parameters based on the slot information and generates a recommendation list.

[0100] The embodiments of this application use a slot-filling method to track the status of user demand information, which makes it easier to organize the parameters called by the recommendation tool more clearly, improves the accuracy of the model calling the tool, and thus improves the accuracy of the recommendation list generated by the recommendation agent.

[0101] Optionally, a single-round interaction can also have multiple slot templates to capture various user needs, such as multiple song attributes.

[0102] In another example, the state-tracking agent generates clarifying questions to supplement missing information when the user has a very specific preference for the search results but the intent is not very clear; for example, if the user has a very specific preference for the search results, the agent ensures that the more specific slots are filled before calling the recommendation tool.

[0103] Knowledge and recommendation skills, as a combined skill set, address scenarios where user needs are relatively clear but require external knowledge acquisition, such as "Please recommend some songs suitable for listening to while running." In this scenario, a search planning agent and an answer summarizing agent can be used to obtain relevant knowledge information for the current query, such as "What songs are suitable for listening to while running?". Once the relevant knowledge information is obtained, a recommendation agent can be used to provide recommended items mentioned in the answer summarizing agent's answer, thus forming a list of knowledge answers and item recommendations.

[0104] Understandable, Figure 2The overall framework of the interactive recommendation agent system shown is merely an example of an implementable framework and does not constitute a limitation on the embodiments of this application. For example, in some other examples, the interactive recommendation agent system may not include a preprocessing task module, but only a route planning task module and a response task module. That is, the query input by the user is not preprocessed and is directly routed through the route planning task module to the response skill unit in the response task module that matches the user's needs for response generation.

[0105] Optionally, in the interactive recommendation intelligent agent system provided in this application embodiment, each task module supports adding corresponding skill units according to user needs, thereby expanding the system's functions. For example, the preprocessing task module, planning task module, and response task module all support expanding skill units based on needs, thus improving the system's functional scalability.

[0106] In summary, this application provides an interactive recommendation agent system based on skill-based hierarchical collaboration. By using skills as the organizational unit for collaboration, it simplifies the complexity of agent tasks and improves the system's functional scalability. This application is user-demand oriented, matching user needs with system skills to enhance the system's task success rate and personalized user experience.

[0107] Figure 4 This is a flowchart illustrating a response method provided in an embodiment of this application. This method can be executed by any device, apparatus, platform, or cluster of devices with computing capabilities. This application does not specifically limit the specific computing device executing the method; a suitable computing device can be selected as needed. For example, it can be executed by… Figure 1 The implementation is executed on server 130, providing intelligent response services to users. Alternatively, the implementation can be executed collaboratively by terminal 11 and server 130. The response method provided in this application embodiment is described below using a collaborative implementation by a terminal device and server as an example. Figure 4 As shown, the response method provided in this application embodiment includes at least steps S401 to S404.

[0108] In step S401, the query text is input into the routing planning task module so that the routing skill unit in the routing planning task module routes the response task to the target skill unit among multiple response skill units.

[0109] For example, a user opens a smart response client (e.g., a mobile phone assistant) on a terminal device (e.g., a smartphone) and enters a query on the client's interactive interface. For instance, the interface provides an input text box, where the user enters the query. Alternatively, opening the smart response client triggers an interactive recommendation system. The terminal device sends user information (e.g., user ID) to the server. The server receives the user information, generates several queries based on the user's historical interaction information, and sends these queries back to the terminal device for display on the interactive interface. The user can select the query that matches their intent from the displayed queries, increasing user-friendliness. The terminal generates a response request carrying the user-input query and sends it to the server. The server receives the response request and executes the response task. The server then sends the response request to the routing planning task module, which invokes a routing agent to perform one or more operations on the query, including domain recognition, user intent recognition, sentiment recognition, and task decomposition. Based on the recognition results, the response task is then routed to the target skill unit in the response task module that matches the user's needs.

[0110] For example, the routing agent can perform only one of the following operations on the query: domain recognition, user intent recognition, sentiment recognition, and task decomposition. For instance, the routing agent can perform only user intent recognition on the query. The query is input into the routing agent, which then performs intent recognition. If the query indicates the user intent is casual conversation, the response task is routed to the casual conversation response skill unit; if it indicates knowledge acquisition, the response task is routed to the knowledge acquisition response skill unit; if it indicates a recommendation need, the response task is routed to the recommendation response skill unit; and if it indicates both recommendation and knowledge acquisition needs, the response task is routed to both recommendation and knowledge acquisition response skill units.

[0111] For example, a routing agent can perform multiple operations on a query, including domain identification, user intent recognition, sentiment recognition, and task decomposition. For instance, when a query is input into the routing agent, it identifies the domain involved, the user intent expressed, the user's sentiment, and decomposes the task into several simpler subtasks. These subtasks are then routed to matching response skill units. During the routing process, the identified domain and sentiment information is carried with the target response skill unit, enabling it to generate higher-quality responses. For example, if the domain is identified as music, the relevant agent in the target response skill unit can access a music-related database to provide a more professional response. Similarly, if the identified sentiment is happiness, the relevant agent in the target response skill unit can consider generating a happy sentiment when generating a response. For instance, if the target response skill unit is a recommendation skill unit, the recommendation agent might recommend cheerful music to the user.

[0112] Optionally, the route planning task module can also rewrite the query input by the user. For example, the user input query can be input into the route agent, and the route agent can rewrite the input query to make the intent of the rewritten query clearer and improve the accuracy of the subsequent response generation.

[0113] In step S402, the target skill unit in the response task module is invoked to process the response task and generate the response result corresponding to the query text.

[0114] In this embodiment, after the target response skill unit in the response task module receives the routed response task, it calls the relevant intelligent agents to process the response task and generate the response result corresponding to the query. For example, if the target response skill unit is a recommendation response skill unit, then the target response skill unit calls three intelligent agents—a state tracking agent, a recommendation agent, and an interaction agent—to collaboratively complete the response task, generating a recommendation list along with recommendation explanations and reasons.

[0115] For details on the specific structure of each response skill unit and how each response skill unit calls the relevant intelligent agent to execute the routed response task, please refer to the description above. For the sake of brevity, it will not be repeated here.

[0116] In some other examples, before step S401 is executed, a preprocessing task module is invoked to preprocess the query data. For example, the preprocessing module invokes a sensitive word detection unit and a typo correction unit to perform sensitive word detection and typo correction on the user-input query. For example, if the sensitive word detection unit detects a sensitive word in the user-input query, it directly generates a safe response, such as "I cannot answer your question." If the typo correction unit detects a typo in the user-input query, it corrects the typo to ensure the accuracy of the query and facilitate the accuracy of subsequent work.

[0117] In other examples, the preprocessing task module further includes a user profiling skill unit, a memory extraction skill unit, and a memory retrieval skill unit. The user profiling skill unit generates a user profile based on the user's historical interaction records. The memory extraction skill unit extracts specific information generated in the current dialogue (e.g., information related to user preferences) to form memory knowledge data and stores this memory knowledge data in a memory database. The memory retrieval skill unit retrieves and recalls memory knowledge data related to the current query from the memory database. The preprocessing task module's preprocessing of the query data also includes calling the memory retrieval skill unit to retrieve and recall memory knowledge data related to the current query from the memory database based on the current query. The response method provided in this application embodiment further includes, after preprocessing the query text, the preprocessing task module sending a response request to the routing planning task module. The response request carries the current query and the memory data retrieved by the memory retrieval skill unit. The routing planning task module performs intent recognition on the query text; when the identified user intent is unclear, it rewrites the query text based on the memory data.

[0118] In other examples, the response method provided in this application, after preprocessing the Query, further includes: a preprocessing task module sending a response request to a routing planning task module, the response request carrying the query text and memory data retrieved by the memory retrieval skill unit; the routing planning task module performing intent recognition on the query text and routing the response request to the target response skill unit based on the intent recognition; and the response task module calling the target response skill unit to generate the response result corresponding to the Query based on the current Query and the memory data. For example, the target response skill unit could be a recommendation response skill unit, which can generate recommendations that better match user preferences and generate more reasonable explanations and reasons for the recommendations based on the memory data. For example, if the Query is to recommend Cantonese songs, and the preprocessing memory retrieval finds that the user likes songs by Eason Chan, then Cantonese songs by Eason Chan would be recommended to the user, improving the accuracy of the recommendations and enhancing the user experience.

[0119] Optionally, the response method provided in this application further includes, before preprocessing the Query, a preprocessing task module calling the homepage Query recommendation skill unit to generate homepage Query recommendations based on the user profile. For example, when a user opens the smart response client, the interactive recommendation system is triggered. The terminal device sends user information (e.g., user ID) to the server. The server receives the user information, the preprocessing task module calls the homepage Query recommendation skill unit to generate homepage Query recommendations based on the user profile, and returns the generated homepage Query recommendations to the terminal device. The terminal device displays the generated homepage Query recommendations on the interactive interface, and the user can input a Query that matches their intent from the several homepage Query recommendations displayed on the interactive interface, thereby increasing the user-friendliness of the interaction.

[0120] The response method provided in this application can realize online / offline memory retrieval and updating. Memory includes the activation, retrieval, storage, updating, retrieval, and use of user-personalized memories, enabling personalized query rewriting and personalized response generation, thus enhancing the user's personalized experience. When the user's intent is unclear, user memories can be used for personalized query rewriting; furthermore, user-personalized information can be strengthened through state tracking and intent clarification.

[0121] In one example, the response method provided in this application further includes displaying the retrieved personalized memory data of the user; and updating the personalized memory data based on the user's actions on the displayed personalized memory data.

[0122] Figure 5 This illustrates an example of how retrieved personalized memory data of a user is displayed to the user. For example... Figure 5 As shown, the retrieved personalized memory data for the user includes phrases like "I like his song 'Love Story'" and "I quite like Taylor Swift's 'You Belong With Me'". Each memory data entry has an action button; for example, the memory data "I quite like Taylor Swift's 'You Belong With Me'" has a "Delete0" button. If the user clicks the "Delete0" button, the terminal device generates a deletion request, which carries the memory data to be deleted and sends the deletion request to the server. The server responds to the deletion request, deletes the memory data "I quite like Taylor Swift's 'You Belong With Me'", and updates the memory database.

[0123] Figure 6 This diagram illustrates the signaling interaction between a user and a recommendation agent system in a music recommendation scenario. Figure 6As shown, the user opens the music assistant APP of the interactive recommendation intelligent agent system provided in this application embodiment, and enters a query. After receiving the user's input query, the music assistant APP calls the preprocessing task module to preprocess the user's input query, including query error correction, sensitive word detection and filtering, etc. Then, the preprocessed query is sent to the routing intelligent agent. The routing intelligent agent performs intent recognition task decomposition and query rewriting on the received query, and then the routing intelligent agent routes the response task to the target response skill unit that matches the user's needs. Figure 6 The diagram illustrates the specific implementation of how the chatty response skill unit, knowledge acquisition response skill unit, recommendation response skill unit, and knowledge and recommendation response skill unit in the response task module call the relevant intelligent agents to process the response tasks after receiving them.

[0124] Figure 7 A schematic diagram illustrating the dialogue effect of an interactive recommendation system agent system applying the response method provided in the embodiments of this application is shown. Figure 7 As shown, this interactive recommendation system's intelligent agent system possesses response skills such as casual conversation and knowledge retrieval, and can automatically match users' casual conversation and knowledge retrieval needs to meet users' diverse needs.

[0125] Figure 8 This illustration shows the effect of personalized recovery with enhanced memory in the interactive recommendation system intelligent agent system provided in an embodiment of this application. Figure 8 As shown, this interactive recommendation system's intelligent agent system can extract personalized preference memories, enabling personalized response generation and improving user experience.

[0126] This application embodiment also provides a computing device, including at least one processor, a memory, and a communication interface, wherein the processor is used to execute... Figure 4-5 The method described.

[0127] Figure 9 A schematic diagram of the structure of a computing device provided in an embodiment of this application.

[0128] like Figure 9As shown, the computing device 900 includes at least one processor 901, a memory 902, and a communication interface 903. The processor 901, memory 902, and communication interface 903 are communicatively connected, which can be achieved via a wired (e.g., bus) or wireless connection. The communication interface 903 is used to send and / or receive data from other devices. The memory 902 stores computer instructions, which the processor 901 executes to perform the methods described in the foregoing method embodiments, thereby fulfilling the response task that meets the diverse needs of the user and providing the user with a high-quality response.

[0129] It should be understood that in the embodiments of this application, the processor 901 may be a central processing unit (CPU), and the processor 601 may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0130] The memory 902 may include read-only memory and random access memory, and provides instructions and data to the processor 901. The memory 902 may also include non-volatile random access memory. Optionally, the random access memory may be, for example, high bandwidth memory (HBM).

[0131] The memory 902 can be volatile memory or non-volatile memory, or it can include both. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0132] It should be understood that the computing device 900 according to the embodiments of this application can perform the implementation of the embodiments of this application. Figure 4-5 The method shown is described in detail above, and will not be repeated here for the sake of brevity.

[0133] Embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein when the computer instructions are executed by a processor, the aforementioned method is implemented.

[0134] An embodiment of this application provides a chip including at least one processor and an interface, wherein the at least one processor determines program instructions or data through the interface; the at least one processor is used to execute the program instructions to implement the method mentioned above.

[0135] Embodiments of this application provide a computer program or computer program product that includes instructions that, when executed, cause a computer to perform the methods mentioned above.

[0136] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0137] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented using hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0138] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A response system, characterized in that, include: The routing planning task module includes a routing skill unit, which is used to route a response task to a target skill unit among multiple response skill units. The response task is obtained based on query text, and the response skill capabilities of the target response skill unit match the response skill capabilities required by the response task. The response task module includes multiple response skill units, each of which performs response tasks with different response skill requirements. The target response skill unit performs the response task and generates a response result corresponding to the query text. The routing skill unit and the response skill unit each include one or more intelligent agents.

2. The response system according to claim 1, characterized in that, The plurality of response skill units includes one or more of the following: chat response skill unit, knowledge acquisition response skill unit, recommendation response skill unit, and knowledge and recommendation response skill unit; The casual conversation response skill unit is used to process response tasks with casual conversation needs, the knowledge acquisition response skill unit is used to process response tasks with knowledge acquisition needs, the recommendation response skill unit is used to process response tasks with recommendation needs, and the knowledge and recommendation response skill unit is used to process response tasks with both knowledge acquisition and recommendation needs.

3. The response system according to claim 2, characterized in that, The chatter response skill unit includes a general chatter agent and a domain chatter agent. The general chatter agent is used to handle the response task with general chatter needs, and the domain chatter agent is used to handle the response task with domain chatter needs. And / or, The knowledge acquisition and response skill unit includes a search planning agent and an answer summarization agent. The search planning agent is equipped with a web search tool and is used to call the web search tool to search for knowledge information based on the query text. The answer summarization agent summarizes the retrieved knowledge information and generates the answer corresponding to the query text. And / or, The recommendation response skill unit includes a state tracking agent, a recommendation agent, and an interaction agent. The recommendation agent is configured with a search / recommendation tool. The state tracking agent tracks the current user information need status. When the user information need is clear, it calls the recommendation agent to generate a recommendation list. When the current user information need is unclear, it generates a clarifying question to ask the user to obtain a clear understanding of the user information need. The recommendation agent calls the search / recommendation tool to generate a recommendation list based on the user information need. The interaction agent generates explanations and reasons for the recommendations based on the query text and the recommendation list. And / or, The knowledge and recommendation response skill unit includes the search planning agent and the answer summarization agent.

4. The response system according to claim 3, characterized in that, The state tracking agent tracks the state of the user's needs information using a slot-filling method.

5. The response system according to any one of claims 1-4, characterized in that, The routing skill unit includes a routing agent, which performs one or more of the following operations on the query text: domain recognition, user intent recognition, sentiment recognition, and task decomposition.

6. The response system according to any one of claims 1-5, characterized in that, Also includes: The preprocessing task module is used to preprocess the query text. The preprocessing module includes one or more of the following: user profiling skill unit, homepage query text recommendation skill unit, memory extraction skill unit, memory retrieval skill unit, sensitive word detection skill unit, and misspelling correction skill unit. The user profiling skill unit, the homepage query text recommendation skill unit, the memory extraction skill unit, the memory retrieval skill unit, the sensitive word detection skill unit, and the typo correction skill unit each include one or more intelligent agents.

7. The response system according to claim 6, characterized in that, The user profiling skill unit includes a user profiling agent, which is used to generate user preference profiles based on user interaction history records. And / or, The homepage query text recommendation skill unit includes a homepage query text recommendation intelligent agent, which is used to generate homepage query text recommendations based on the user preference profile; and / or The memory extraction skill unit includes a memory extraction agent, which is equipped with a memory extraction tool. The memory extraction agent is used to call the memory extraction tool to extract memory knowledge data from the generated dialogue and store the memory knowledge data in the user interaction database. The memory retrieval skill unit includes a memory retrieval agent, which is configured with a memory search tool. The memory retrieval agent is used to invoke the memory search tool to retrieve and recall memory knowledge data related to the query text from the user interaction database; and / or, The sensitive word detection skill unit includes a sensitive word detection agent, which is equipped with a sensitive word search tool. The sensitive word detection agent is used to call the sensitive word detection tool to perform sensitive word detection on the query text. If a sensitive word is detected, a security recovery is generated. And / or, The misspelling correction skill unit includes a correction agent, which is equipped with an entity search tool. The correction agent is used to call the entity search tool to correct the query text.

8. The response system according to claim 6 or 7, characterized in that, The preprocessing task module is also used to send a response request to the routing planning task module, the response request carrying the query text and the memory data retrieved by the memory retrieval skill unit; The routing planning task module is also used to perform intent recognition on the query text. When the recognized user intent is unclear, the query text is rewritten based on the memory data.

9. The response system according to any one of claims 6-8, characterized in that, The preprocessing task module is also used to send a response request to the routing planning task module, the response request carrying the query text and the memory data retrieved by the memory retrieval skill unit; The routing planning task module is also used to perform intent recognition on the query text and route the response request to the target response skill unit based on the intent recognition; The response task module is also used to call the target response skill unit to generate a response result corresponding to the query text based on the query text and the memory data.

10. The response system according to any one of claims 6-9, characterized in that, The routing planning task module, the response task module, and the preprocessing task module all support the expansion of skill units based on requirements.

11. The response system according to any one of claims 6-10, characterized in that, Also includes: The display module is used to display the retrieved personalized memory data of the user; The memory update module is used to update the personalized memory data based on the user's actions on the displayed personalized memory data.

12. A response method, characterized in that, include: The query text is input into the routing planning task module, so that the routing skill unit in the routing planning task module routes the response task to the target skill unit among multiple response skill units. The response task is obtained based on the query text, and the response skill capabilities of the target response skill unit match the response skill capabilities required by the response task. The target skill unit in the response task module is invoked to process the response task and generate the response result corresponding to the query text. Both the routing skill unit and the response skill unit include one or more intelligent agents.

13. The method according to claim 12, wherein the plurality of response skill units includes one or more of a casual conversation response skill unit, a knowledge acquisition response skill unit, a recommendation response skill unit, and a knowledge and recommendation response skill unit; in, The casual conversation response skill unit is used to process the response task with a casual conversation requirement, the knowledge acquisition response skill unit is used to process the response task with a knowledge acquisition requirement, the recommendation response skill unit is used to process the response task with a recommendation requirement, and the knowledge and recommendation response skill unit is used to process the response task with both knowledge acquisition and recommendation requirements.

14. The method according to claim 13, characterized in that, The chatter response skill unit includes a general chatter agent and a domain chatter agent. The general chatter agent is used to handle the response task with general chatter needs, and the domain chatter agent is used to handle the response task with domain chatter needs. And / or, The knowledge acquisition and response skill unit includes a search planning agent and an answer summarization agent. The search planning agent is equipped with a web search tool and is used to call the web search tool to search for knowledge information based on the query text. The answer summarization agent summarizes the retrieved knowledge information and generates the answer corresponding to the query text. And / or, The recommendation response skill unit includes a state tracking agent, a recommendation agent, and an interaction agent. The recommendation agent is configured with a search / recommendation tool. The state tracking agent tracks the current user information need status. When the user information need is clear, it calls the recommendation agent to generate a recommendation list. When the current user information need is unclear, it generates a clarifying question to ask the user to obtain a clear understanding of the user information need. The recommendation agent calls the search / recommendation tool to generate a recommendation list based on the user information need. The interaction agent generates explanations and reasons for the recommendations based on the query text and the recommendation list. And / or, The knowledge and recommendation response skill unit includes the search planning agent and the answer summarization agent.

15. The method according to any one of claims 12-14, characterized in that, The module for inputting query text into the routing planning task previously included: The query text is input into the preprocessing task module for preprocessing. The preprocessing module includes one or more of the following: user profiling skill unit, homepage query text recommendation skill unit, memory extraction skill unit, memory retrieval skill unit, sensitive word detection skill unit, and misspelling correction skill unit. The user profiling skill unit, the homepage query text recommendation skill unit, the memory extraction skill unit, the memory retrieval skill unit, the sensitive word detection skill unit, and the typo correction skill unit each include one or more intelligent agents.

16. The method according to claim 15, characterized in that, The routing planning task module, the response task module, and the preprocessing task module all support the expansion of skill units based on requirements.

17. A computing device, comprising a memory and a processor, characterized in that, The memory stores instructions that, when executed by a processor, cause the method described in any one of claims 12-16 to be implemented.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it causes the method described in any one of claims 12-16 to be implemented.

Images