Content generation method, system, apparatus, and electronic device

Abstract

This application discloses a content generation method, system, apparatus, and electronic device. The method includes: acquiring input information; invoking the intent recognition module of each task processing service to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; selecting a target service from the task processing services capable of processing the input information according to the first processing capability information; invoking the solution module of the target service to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; and determining the response information corresponding to the input information based on the preliminary response information. The solution provided by this application can respond more accurately to complex and diverse user inputs, improving the user experience.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a content generation method, system, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] With the rapid development of computer technology, intelligent customer service and other intelligent question-answering systems are increasingly favored by various industries. Intelligent question-answering systems in this field typically train an intelligent question-answering model. This model can generate answers based on user input and a pre-set knowledge base, and then return them to the user. Alternatively, it can respond to user input based on a pre-set knowledge base and content generation rules. However, due to the limited capabilities of a single model, it is difficult to cope with complex and diverse user needs. When the user's input question exceeds the model's training range or the knowledge base content, the system may fail to provide accurate and effective responses, resulting in a poor user experience. Summary of the Invention

[0003] This application provides a content generation method, system, apparatus, electronic device, and computer-readable storage medium, which can more accurately respond to complex and diverse user input, thereby improving user experience. The specific solution is as follows: Firstly, this application provides a content generation method applied to a server, the method comprising: Obtain input information; The intent recognition module of each task processing service is invoked to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; Based on the first processing capability information, a target service is selected from each of the task processing services that have the capability to process the input information; The solution module of the target service is invoked to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; The response information corresponding to the input information is determined based on the preliminary response information.

[0004] Secondly, this application provides a content generation system, including: an input / output interface system, an intelligent gateway, and a task processing server cluster; The input / output interface system is used to allow users to input information and send the user's input information to the smart gateway; The smart gateway is used to call the intent recognition module of each task processing service in the task processing server cluster, and send the input information to the intent recognition module; The intent recognition module of each task processing service in the task processing server cluster is used to determine the first processing capability information of the task processing service for the input information, and send the first processing capability information to the smart gateway. The smart gateway is also configured to select a target service from the task processing services that have the ability to process the input information based on the first processing capability information, and call the solution module of the target service; The solution module of the target service is used to generate preliminary response information corresponding to the input information and send the preliminary response information to the smart gateway; The smart gateway is also used to determine the response information corresponding to the input information based on the preliminary response information, and send the response information to the input / output interface system; The input / output interface system is also used to display the response information.

[0005] Thirdly, this application also provides a content generation apparatus applied to a server, the apparatus comprising: The acquisition unit is used to acquire input information; The calling unit is used to call the intent recognition module of each task processing service to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; The selection unit is configured to select a target service from each of the task processing services capable of processing the input information, based on the first processing capability information. The calling unit is also used to call the solution module of the target service to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; The determining unit is configured to determine the response information corresponding to the input information based on the preliminary response information. Fourthly, this application also provides an electronic device, including: a processor, a memory, and computer program instructions stored in the memory and executable on the processor; the processor, when executing the computer program instructions, implements the method as described in the first aspect.

[0006] Fifthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect.

[0007] In a sixth aspect, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0008] Compared with the prior art, this application has the following advantages: The content generation method provided in this application embodiment involves the server acquiring input information and then invoking the intent recognition module of each task processing service. This intent recognition module determines the first processing capability information of the task processing service for the input information and sends this first processing capability information to the server. The server can then select a target service from among the task processing services capable of processing the input information based on the first processing capability information, and then invoke the solution module of the target service. This solution module generates preliminary response information corresponding to the input information and sends this preliminary response information to the server. The server then determines the corresponding response information based on the preliminary response information.

[0009] As can be seen, the content generation method provided in this application integrates multiple task processing services with different professional capabilities through a server, forming a processing system in which the server schedules multiple external services to work collaboratively. When faced with user input information, the server can determine the corresponding first processing capability information by evaluating the capabilities of each task processing service through its intent recognition module. This allows the server to accurately locate the target service that has the capability to process the input information and is most suitable for processing the current input information. This multi-service collaboration and optimal selection mechanism effectively compensates for the limitations of a single model in terms of capability range and knowledge coverage. By calling the solution module of the target service to generate preliminary response information, the solution module of the task processing service most suitable for processing the complex and diverse input information can be selected for targeted processing, thereby generating preliminary response information that is more in line with user needs and more accurate and effective. On this basis, the server determines the final response information based on the preliminary response information, further ensuring the quality and reliability of the response. Compared with traditional single-model intelligent question answering systems, this application embodiment significantly improves the processing capability and response accuracy for complex and diverse user input through multi-service collaboration and dynamic selection mechanisms, thereby effectively improving the user experience.

[0010] Furthermore, in this application, each task processing service, through modular processing of the intent recognition module and the solution module, can quickly identify whether the input information belongs to its own processing capabilities and quickly generate the corresponding response. This modular design not only improves the response efficiency of each task processing service but also facilitates system expansion and maintenance. Moreover, this application allows for the flexible integration of new task processing services or the removal of unnecessary services based on actual business needs, enabling the system to quickly adapt to content generation requirements in different domains and scenarios, demonstrating strong scalability and adaptability. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the application scenario of the solution provided in this application, and also a schematic diagram of the structure of the content generation system provided in this application.

[0012] Figure 2 This is a flowchart illustrating an example of the content generation method provided in the embodiments of this application.

[0013] Figure 3 This is a flowchart illustrating another example of the content generation method provided in the embodiments of this application.

[0014] Figure 4 This is a flowchart illustrating yet another example of the content generation method provided in the embodiments of this application.

[0015] Figure 5 This is a schematic diagram of the structure of the content generation device provided in the embodiments of this application.

[0016] Figure 6 This is a structural block diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions of this application, the application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. However, this application can be implemented in many other ways different from those described below. Therefore, based on the embodiments provided in this application, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this application.

[0018] It should be noted that the terms "first," "second," "third," etc., in the claims, specification, and drawings of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. Such data are interchangeable where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown or described in this application. Furthermore, the terms "comprising," "having," and their variations are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0019] It should be understood that in the embodiments of this application, "at least one" means one or more, and "more than one" means two or more. "And / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship. "Contains A, B and / or C" means containing any one, two, or three of A, B, and C.

[0020] It should be understood that in the embodiments of this application, "B corresponding to A", "B corresponding to A", "A corresponds to B" or "B corresponds to A" means that B is associated with A, and B can be determined based on A. Determining B based on A does not mean that B is determined solely based on A; B can also be determined based on A and / or other information.

[0021] To facilitate understanding of the various embodiments of this application, the application background of the embodiments will be explained.

[0022] With the rapid development of computer technology, intelligent customer service and other intelligent question-answering systems are increasingly favored by various industries. Intelligent question-answering systems in this field typically train an intelligent question-answering model. This model can generate answers based on user input and a pre-set knowledge base, and then return them to the user. Alternatively, it can respond to user input based on a pre-set knowledge base and content generation rules. However, due to the limited capabilities of a single model, it is difficult to cope with complex and diverse user needs. When the user's input question exceeds the model's training range or the knowledge base content, the system may fail to provide accurate and effective responses, resulting in a poor user experience.

[0023] To address the above issues, embodiments of this application provide a content generation method, system, apparatus, electronic device, computer-readable storage medium, and computer program product. The aim is to provide more accurate responses to complex and diverse user input, thereby improving user experience.

[0024] The content generation method provided in this application can be applied to content generation in various fields, specifically to scenarios such as intelligent customer service, intelligent Q&A, and intelligent writing. For example, in the intelligent customer service scenario of an e-commerce platform, users may input diverse information such as product inquiries, order inquiries, and after-sales issues. The method provided in this application can accurately identify the intent of the user's input information by integrating multiple external task processing services such as product inquiry processing, order processing, and after-sales processing, and select the corresponding target service to generate a response, thereby efficiently and accurately responding to user needs. This application does not limit the specific application scope of content generation.

[0025] To facilitate understanding of the method embodiments of this application, their application scenarios are described. Please refer to... Figure 1 , Figure 1 This is a schematic diagram illustrating an application scenario of the solution provided in the embodiments of this application. This application scenario is an illustrative example and is not intended to limit the specific application scenario.

[0026] like Figure 1 As shown, this application scenario includes a smart gateway 100, a client 200, and a task processing server cluster 300. The smart gateway 100 integrates a main control server 400. The task processing server cluster 300 includes multiple task processing services, each corresponding to a different business area or processing capability, such as product consultation service, order inquiry service, after-sales problem handling service, and technical support service. One task processing service can correspond to one server or a server cluster formed by multiple servers. The client 200 serves as the entry point for user interaction with the system, providing an input / output interface. Users input information through the client 200, which is transmitted to the smart gateway 100 via the network. After receiving the input information, the smart gateway 100 forwards it to the main control server 400. The main control server 400 coordinates and schedules the various task processing services in the task processing server cluster 300 to collaboratively process and obtain response information. The server then sends the response information to the client 200 for display through the smart gateway 100. The task processing service includes an intent recognition module 301 and a solution module 302. The intent recognition module 301 is used to evaluate the task processing service’s ability to process input information, and the solution module 302 is used to generate preliminary response information after determining that the service is the target service.

[0027] The client 200 can be an electronic device with data processing capabilities, such as a mobile phone, tablet, smartwatch, desktop computer, smart TV, VR device, in-vehicle device, wearable device, or laptop. The client 200 provides an input interface for the user to receive various input information such as text, voice, and images, and sends the input information to the smart gateway 100. At the same time, it receives the response information returned by the smart gateway 100 and displays it to the user through output components such as a display screen and speakers.

[0028] The intelligent gateway 100 acts as a communication bridge between the client 200 and the main control server 400. It is responsible for receiving input information sent by the client 200 and information associated with the input information (such as the context information of the input information) and forwarding it to the main control server 400. It also receives the response information generated by the main control server 400, assembles it into a preset format, and forwards it to the corresponding client 200, thus ensuring the stability and security of data transmission.

[0029] The master control server 400 can include a single server, a cluster of multiple distributed servers, or a cloud server. The master control server 400 can possess high computing power. Upon receiving input information forwarded by the smart gateway 100, the master control server 400 selects the task processing service that meets the current input information processing requirements from various task processing services as the target service. The master control server 400 then sends a call command to the solution module 302 of the target service, triggering its specific response content generation process based on the input information. The solution module 302 generates preliminary response information and sends it to the active server. After receiving the preliminary response information, the master control server 400 further verifies, optimizes, or integrates it, and finally feeds back the response information to the client 200 through the smart gateway 100, which then presents it to the user.

[0030] Data interaction can be performed between the client 200 and the smart gateway 100, between the smart gateway 100 and the main control server 400, and between the main control server 400 and the various task processing services in the task processing server cluster 300 through wired or wireless communication networks.

[0031] In other application scenarios, the smart gateway 100 may not be required, and the client 200 may communicate directly with the main control server 400. Alternatively, other application scenarios with different structures may be used. The specific communication architecture can be adjusted according to actual deployment requirements and network environment, and this application does not limit it.

[0032] Example 1 The first embodiment of this application provides a content generation method. The method is applied to a server, which can be a master server. Specifically, the server may include a single server or a server cluster.

[0033] like Figure 2 As shown, the content generation method provided in the first embodiment of this application includes the following steps S110 to S150.

[0034] Step S110: Obtain input information.

[0035] The input information mentioned above can be one or more types of information entered by the user on the client, such as text, voice, images, and video. It can also be command information triggered by the user through other interactive operations on the client. For example, the client's user interface may display multiple query selection controls, each used to trigger a query command for a corresponding category. The command information triggered by the user clicking on a query selection control can also serve as input information. The server can establish a communication connection with the client to receive the input information sent by the client.

[0036] The types of input information typically differ depending on the use case of the content generation method. For example, in an intelligent customer service scenario, the input information could be order query information entered by the user through the client's interactive interface; in an intelligent question-and-answer scenario, the input information could be content inquiry information entered by the user via voice; and in image recognition-related content generation scenarios, the input information could be images uploaded by the user and instructions for recognizing content from the images (such as identifying the names of plants in the image). These all fall under the category of input information entered by the user on the client side.

[0037] Step S120: Invoke the intent recognition module of each task processing service to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server.

[0038] After receiving the input information, the server can send a call request to each connected task processing service. This call request carries the input information and related information, such as context information, user interaction history, and current session context. Each task processing service can be externally connected to the server or integrated within it. When the task processing services are externally connected to the server, the internal computational load is reduced, thereby improving the overall system efficiency. It also facilitates the independent development, testing, and iteration of each task processing service. Therefore, the following description primarily uses the example of task processing services being externally connected to a server to illustrate this application.

[0039] After receiving a call request, each task processing service's intent recognition module will assess the extent to which the task processing service can process the input information based on the input information and, in conjunction with the related information of the input information, obtain the first processing capability information and send the first processing capability information to the aforementioned main control server.

[0040] The aforementioned task processing services can be external services connected to the main control server. Each task processing service can handle tasks in different domains or of different types. For example, in an e-commerce scenario, these services might include product information consultation services, order status query services, logistics tracking services, and after-sales service processing services. Each task processing service can be configured on its own task processing server, and these servers communicate with the main control server via a network. When determining the first processing capability information, the intent recognition module of each task processing service will make judgments based on the task processing scope and knowledge domain of the service. For example, the intent recognition module of the product information consultation service will analyze whether the input information involves product parameters, prices, performance, or other related content, and generate a quantitative score between 0 and 1 based on the matching degree as the first processing capability information. The higher the score, the stronger the service's ability to process the current input information and the higher the matching degree.

[0041] The internal implementation logic of each task processing service can respond to various input information. The main control server only needs to interact with each task processing service to accurately respond to various input information, thereby reducing the coupling of the system. This allows each task processing service to be developed, tested, deployed, and upgraded independently without affecting the scheduling logic of the main control server, enhancing the flexibility and maintainability of the entire content generation system. When business requirements in a certain domain change, or when it is necessary to optimize the processing capacity of a specific task, developers can adjust and update only the corresponding task processing service without modifying the main control server or other unrelated task processing services, effectively reducing the cost and risk of system iteration. At the same time, new task processing services can be connected to the main control server through access interfaces to expand the overall processing capacity of the system to adapt to constantly changing business needs and user scenarios.

[0042] Each task processing service includes an intent recognition module and a solution module. The intent recognition module is used to determine the first processing capability information of the corresponding task processing service for the input information. The solution module is used to generate preliminary response information for the input information based on the input information and the associated information of the input information (such as context information, user history data, etc.).

[0043] Upon receiving a request, each task processing service's intent recognition module determines its initial processing capability information for the input information. This initial processing capability information indicates the service's ability to process the input information; a stronger capability indicates a higher likelihood that the service will accurately and efficiently generate a preliminary response. The initial processing capability information may include, but is not limited to, at least one of the following: a rating value for processing the input information, an evaluation level, a probability value, a judgment of whether it can process the information, a confidence level of its processing capability, or a descriptive text describing its processing capability. This initial processing capability information can be qualitative or quantitative.

[0044] The intent recognition module of the task processing service can determine the first processing capability information by means of keyword matching, semantic similarity calculation, and pre-trained model classification, and send the first processing capability information to the main control server.

[0045] Specifically, the intent recognition module can parse the input information to obtain various query elements such as the query intent, the relevant field, and the type of problem to be solved. These query elements are then matched with the service configuration information of the task processing service to obtain the first processing capability information. The service configuration information of the task processing service may include at least one of the following: the field to which the task processing service belongs, the processing scope, the processing experience, and historical successful cases.

[0046] In one specific embodiment, the intent recognition module can perform intent recognition through a generative artificial intelligence system. This system can perform semantic understanding of the input information, extract semantic features from the input information using natural language processing techniques, and then compare and analyze these semantic features with service configuration information such as the domain knowledge graph and historical processing samples of the task processing service. This generates a quantitative score or qualitative evaluation result reflecting the task processing service's ability to process the input information, serving as the primary processing capability information. The generative artificial intelligence system possesses powerful contextual understanding and reasoning capabilities, enabling it to more accurately capture the intent of the input information. This makes the evaluation of the primary processing capability information more accurate and comprehensive, helping the main control server to more reliably select suitable target services.

[0047] Generative artificial intelligence systems can include pre-trained intent recognition models, which have powerful text understanding, generation, and reasoning capabilities and can identify the matching degree between input information and task processing services.

[0048] In step S120, the first processing capability information of the task processing service for the input information can be determined through the following steps S121~S123.

[0049] Step S121: Provide the following information to the intent recognition module: at least the input information and the associated information of the input information.

[0050] Specifically, the main control server can provide the intent recognition module with input information, related information of the input information, and other information.

[0051] In step S121, the intent recognition module may only provide the aforementioned input information. The associated information for the input information is described above. By providing the intent recognition module with associated information, the module can more comprehensively and accurately understand the precise intent of the user's input. For example, when a user inputs "When will this order arrive?", if the associated information includes the user's currently logged-in account information and the corresponding incomplete order number, the intent recognition module can combine this information to more accurately determine that the input information belongs to the intent of "order query" and clearly indicates a logistics timeliness query for a specific order. This makes the assessed first processing capability information more aligned with actual processing needs. If only the input information "When will this order arrive?" is provided, without associated information such as the user account and order number, the intent recognition module may only be able to vaguely determine that it is related to "order query," but its assessment of its ability to process this fuzzy query may be inaccurate because it lacks crucial contextual information to locate the specific order. Therefore, providing associated information helps improve the accuracy and reliability of the intent recognition module's assessment of the input information processing capability, providing a more valuable reference for the subsequent selection of target services by the main control server.

[0052] Step S122: The intent recognition module determines the relationship between the intent of the input information and the processing capability of the task processing service based on the information provided to it and the pre-set task setting statement.

[0053] Task setting statements can be predefined based on the business domain and processing scope of the task processing service. For example, task setting statements for a product consultation service could include "identify product price, specifications, and functions," "determine if the input information involves product recommendations," and "recommend complementary products." After receiving the input information and related information, the intent recognition module uses a generative artificial intelligence system to identify the intent of the input information. It compares the intent of the input information (such as "query the product warranty period" obtained through semantic parsing) with the content of the task setting statement to determine the degree of matching, the closeness of association, and whether the task processing service has the resources (such as the corresponding knowledge base, interface permissions, etc.) required to satisfy the intent of the input information. This determines whether the processing capability of the task processing service can meet the intent of the input information.

[0054] For example, if the intent of the input information is "to query the battery capacity of a certain mobile phone model," and the task setting statement of the product information consultation service includes "providing product specification parameter queries," and the service's knowledge base contains detailed parameters for that mobile phone model, then the intent recognition module will determine that the two have a high degree of matching and that the service has the processing resources, thus providing a higher initial processing capability information. Conversely, if the intent of the input information is "to query the driving trajectory of a car," but the task setting statement of the product information consultation service does not involve the field of historical vehicle driving trajectories, and its knowledge base does not contain relevant data, then the intent recognition module will determine that the matching degree is low, and the initial processing capability information will also be correspondingly low, indicating that the product information consultation service cannot effectively process this type of input information.

[0055] Step S123: The intent recognition module obtains at least one of the following based on the discrimination result: confidence level of the task processing service's ability to process the input information and judgment result of whether it can process the input information.

[0056] For example, when the judgment result indicates that the matching degree between the input information intent and the task setting statement is higher than a preset high matching degree threshold, the intent recognition module can generate a high capability confidence score, such as 0.95, and simultaneously output the judgment result "can be processed"; if the matching degree is medium, the capability confidence score can be a medium execution degree, such as 0.6, and the judgment result is still "can be processed"; if the matching degree is lower than a preset low matching degree threshold, the capability confidence score can be below 0.1, and the judgment result is "cannot be processed". This initial processing capability information will serve as the basis for the master control server to filter target services, helping the master control server to find the most suitable service for the current input information from among many task processing services.

[0057] Specifically, the intent recognition module can use the aforementioned generative artificial intelligence system to determine the confidence level of the task processing service's ability to process the input information and the judgment result of whether it can process the input information based on the discrimination result.

[0058] In this embodiment, the intent recognition module, through a generative artificial intelligence system, can determine the relationship between the intent of the input information and the processing capability of the task processing service. This determines the confidence level of the task processing service's ability to process the input information and the judgment result of whether it can process the input information, making the evaluation of the first processing capability information more intelligent and accurate. The generative artificial intelligence system has powerful semantic understanding and reasoning capabilities, enabling it to deeply understand the intent of the input information while comprehensively considering the service configuration information of the task processing service, thereby accurately analyzing the degree of matching between the two. Compared with traditional methods such as keyword matching, this generative artificial intelligence-based discrimination method can better handle complex and ambiguous input information and reduce capability evaluation errors caused by information misunderstanding biases.

[0059] In one embodiment, prior to step S120, the above method may further include the following steps S120a to S120b.

[0060] Step S120a: Detect whether the input information meets the preset conditions for switching to manual intervention.

[0061] The preset conditions for transferring to a human agent include at least one of the following: the input information indicates that a human customer service representative will respond; the input information contains sensitive information; or the input information indicates that a negative emotion is present.

[0062] Step S120b: If the input information meets the preset conditions for transferring to human assistance, the input information is sent to the human customer service client so that the human customer service client can obtain the reply information corresponding to the input information and send the reply information to the server.

[0063] If the input information is detected to meet the preset conditions for transferring to human customer service, such as the user explicitly entering "transfer to human customer service", or the input information containing sensitive information, or the user expressing negative emotions (such as anger or a strong desire to complain) identified through sentiment analysis, the main control server can directly assign the input information to the human customer service system for processing without performing the subsequent steps of calling the task processing service, so as to ensure that the user's needs are handled more properly.

[0064] If the input information does not meet the preset conditions for transferring to human intervention, then continue to execute step S120, which calls the intent recognition module of each external task processing service to determine the first processing capability information of the task processing service for the input information.

[0065] Step S130: Based on the first processing capability information, select a target service from each of the task processing services that have the capability to process the input information.

[0066] The task processing service capable of processing the input information can be a task processing service that is determined to be capable of processing the input information in the first processing capability information, or a task processing service whose confidence in processing the input information is greater than the first confidence threshold in the first processing capability information, or a task processing service whose evaluation level in the first processing capability information reaches a preset level, etc. The specific filtering conditions can be flexibly set according to the processing capability requirements in the actual application scenario.

[0067] When there is one task processing service capable of processing the input information, that task processing service is identified as the target service. When there are multiple task processing services capable of processing the input information, the task processing service with the highest processing capability indicated by the first processing capability information can be selected as the target service.

[0068] In one specific embodiment, when the first processing capability information includes the judgment result of whether the task processing service can process the input information and the confidence level of the ability of the task processing service to process the input information, step S130 can be implemented according to the following steps S131 to S132.

[0069] Step S131: Select a candidate service from each of the task processing services whose judgment result for whether it can process the input information is yes and whose capability confidence is greater than a preset confidence threshold.

[0070] The preset reliability threshold can be set based on the system's comprehensive requirements for processing accuracy and efficiency. For example, in areas where high accuracy is required, a relatively large threshold such as 0.85 or 0.9 can be set to ensure that only services with strong processing capabilities become candidates. In scenarios with high real-time requirements, such as casual conversations in intelligent customer service, the threshold can be appropriately lowered to 0.6 to improve response speed and expand service coverage. Furthermore, the preset reliability threshold can be dynamically optimized based on historical user feedback data. For instance, if the actual processing accuracy of a candidate service at a certain threshold is lower than expected, the threshold can be automatically raised, and vice versa, thus making the selection criteria more aligned with actual needs.

[0071] Step S132: Select the target service from the candidate services according to the predetermined selection method.

[0072] Specifically, when there is only one candidate service, it can be directly selected as the target service. When there are multiple candidate services, the predetermined selection method is to choose the candidate service with the highest capability confidence as the target service. This ensures that the selected service has a strong ability to process the input information, thereby generating more accurate preliminary response information. Alternatively, for specific types of input information, a target service matching the type of input information can be selected from the candidate services. For example, for user after-sales service requests, even if the capability confidence of the corresponding after-sales service processing service is slightly lower than other candidate services, it can still be selected as the target service to improve user satisfaction.

[0073] In this embodiment, task processing services with a capability confidence level greater than a preset confidence threshold and a judgment result of "able to handle" are selected as candidate services. This effectively avoids assigning input information to services that are clearly incapable of handling the task, thereby ensuring the quality of the initial response information from the source and further improving the quality of content generation.

[0074] In one specific embodiment, step S132 can be performed by selecting the target service according to the following steps S132a to S132b.

[0075] Step S132a: When there are multiple candidate services, and the confidence difference between the highest and second-highest capability confidence in each candidate service is greater than a preset difference, the candidate service with the highest capability confidence is selected as the target service.

[0076] The preset difference can be set according to the system's requirements for the stability of service selection. For example, if the maximum confidence value of capability is 1, the preset difference can be set to any difference between 0.15 and 0.25. The larger the preset difference, the stronger the processing capability of the selected target service. The smaller the preset difference, the simpler the selection process of the target service, and the faster the target service can be determined.

[0077] For example, if the confidence level of candidate service A's capability is 0.92 and the confidence level of candidate service B's capability is 0.75, with a preset difference of 0.15, then the difference between the highest capability confidence level (0.92) and the second highest capability confidence level (0.75) is 0.17, which is greater than the preset difference of 0.15. Therefore, candidate service A can be directly selected as the target service. This method can quickly identify the target service with the strongest capability when there is a service with a clear advantage, reducing the computational overhead of subsequent complex decisions.

[0078] Step S132b: When there are multiple candidate services, and the confidence difference between the highest and second-highest capability confidence in each candidate service is less than or equal to a preset difference, select the service that has processed similar information from each candidate service as the target service, wherein the similarity between the similar information and the input information is greater than a preset similarity threshold.

[0079] If the confidence difference between the highest and second-highest capability confidence scores is less than or equal to a preset difference, it indicates that the processing capabilities of the candidate services are relatively similar. In this case, relying solely on capability confidence scores is insufficient to accurately determine which service can better process the current input information. Therefore, experience in processing similar information can be used as a selection criterion.

[0080] Specifically, the master server can extract semantic features of the input information (such as key entities, intent types, question structures, etc.) and calculate similarity with information in the historical processing sample databases of each candidate service. For example, if the input information is "query the after-sales service outlets of a certain brand of laptop", the capability confidence of the candidate service C with the highest capability confidence and the second highest candidate service D with the second highest capability confidence are 0.88 and 0.86 respectively, with a confidence difference of 0.02, which is less than the preset difference of 0.1. In this case, if the historical processing samples of candidate service C contain similar historical processing samples of "query the after-sales service outlets of electronic products" and meet the processing accuracy requirements, and the semantic similarity between the similar historical processing samples and the above input information is higher than the preset similarity threshold, while the historical processing samples of candidate service D have fewer such similar information or lower similarity, then candidate service C is selected as the target service. This approach fully considers the actual processing experience of services. For services with similar capabilities, it prioritizes the service that performs better on similar tasks, thereby further improving the accuracy and reliability of content generation.

[0081] In one specific embodiment, step S132 may select the target service according to the following step S132c.

[0082] Step S132c: When there are multiple candidate services, and the confidence difference between the highest and second-highest capability confidence in each candidate service is less than or equal to a preset difference, further, if none of the candidate services have processed the similar information, then the service description information of each candidate service is semantically matched with the semantics of the input information, and the service whose semantic matching degree meets the preset matching condition is selected as the target service; if none of the candidate services has a service whose service description information and the semantic matching degree meet the preset matching condition, then the service that meets the preset rule is selected from the candidate services as the target service.

[0083] Service description information is used to describe the functions, processing scope, and supported scenarios of the task processing service through text. Matching the service description information semantically with the input information allows us to determine whether the service matches the input information from a service positioning perspective. For example, if the input information is "how to apply for a return or exchange of a certain product," and the confidence scores of candidate services E and F are 0.87 and 0.85 respectively, with a confidence difference of 0.02, and neither has any highly similar historical processing samples, then if the description information of service E includes "processing return and exchange applications," while the description information of service F mainly includes "consultation on product usage methods," then the semantic matching degree of service E's description information is higher, meeting the preset matching conditions (such as a matching score higher than 0.8), and it can be selected as the target service.

[0084] Specifically, a pre-trained semantic matching model can be used to determine the semantic similarity between service description information and input information. When the score is higher than a preset matching threshold (e.g., 0.75), it is determined that the preset matching condition is met. If multiple candidate services meet the preset matching condition after semantic matching, the service with the highest semantic matching degree can be identified as the target service.

[0085] If the semantic matching degree between the description information of each candidate service and the input information does not meet the preset matching conditions, the selection can be made according to the preset rules, such as giving priority to services with faster average response speed, services with higher historical user satisfaction ratings, or selecting according to the priority order of service registration, so as to determine the target service.

[0086] In this embodiment, when candidate services have similar processing capability confidence levels and lack experience in processing similar information, semantic matching of service description information is used for further screening. This allows for the determination of the intrinsic relationship between candidate services and input information from the perspective of service self-positioning and functional description, avoiding selection bias caused by insufficient historical data. When semantic matching still cannot determine the most suitable service, preset rules are introduced as a fallback strategy. These rules comprehensively consider indicators such as service response efficiency and user feedback, which are not related to capability confidence levels, ensuring that the selection of target services is both scientific and reasonable and practically feasible. This allows for the stable selection of the most suitable processing service even in complex scenarios, laying the foundation for generating high-quality preliminary response information.

[0087] Step S140: Invoke the solution module of the target service to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server.

[0088] Once the target service is identified, the master server initiates a call request to the solution module of that target service. This call request includes input information and may also include related information. Upon receiving the call request, the solution module can generate preliminary response information corresponding to the input information based on its internal preset business logic, processing flow, and relevant knowledge bases, models, and other resources. This preliminary response information is then sent back to the master server for further processing or direct feedback to the user.

[0089] In one specific embodiment, the solution module for the target service can generate information using a generative artificial intelligence system. This generative AI system, based on input information and related information, combined with knowledge and language patterns learned from its training data, can generate coherent, relevant, and scenario-specific preliminary response information to the input information. This generative solution module can not only handle structured problems but also generate targeted responses to unstructured, open-ended input information through semantic understanding and reasoning, greatly expanding the scope and flexibility of content generation.

[0090] In step S140, the solution module can generate preliminary response information corresponding to the input information according to the following steps S141 to S142.

[0091] Step S141: Provide the following information to the solution module: at least the input information and the associated information of the input information.

[0092] Alternatively, you can provide only the aforementioned input information to the solution module without providing related information. Whether or not to provide related information depends on the actual needs of the solution module for the target service and the complexity of the input information. For example, when the input information is simple and clear, such as "query the current time," providing only the input information is sufficient to generate a preliminary response. However, when the input information involves personalized content such as the user's historical orders or membership level, you can provide the solution module with related information including user identifiers so that the module can accurately query and generate the corresponding response.

[0093] Step S142: The solution module performs semantic understanding on the input information based on the information provided to it and a pre-trained dialogue generation model, and generates preliminary response information corresponding to the input information.

[0094] Dialogue generation models can be deep learning language models with a large parameter scale, but are not limited to a specific scale. These models are trained on large-scale text data and possess powerful natural language understanding and generation capabilities. When generating initial response information, the dialogue generation model first performs semantic parsing on the input and related information, identifying elements such as intent, key entities, and contextual logical relationships, and then generates initial response information based on the parsing results.

[0095] In this embodiment, a generative artificial intelligence system is configured in the solution module. Through a pre-trained dialogue generation model, the model can be fully utilized to efficiently obtain preliminary response information.

[0096] Step S150: Determine the response information corresponding to the input information based on the preliminary response information.

[0097] Specifically, the initial response information can be directly used to determine the response information corresponding to the input information, or the initial response information can be further processed to obtain the response information. For example, if the initial response information is not clear enough, the logic is not coherent, or it contains redundant or sensitive information, further desensitization processing and content supplementation can be performed on the initial response information to improve the accuracy of the response information.

[0098] In one embodiment, the following step S150a may be included before step S150.

[0099] Step S150a: Detect whether the preliminary response information falls within the preset constraints.

[0100] The preset constraints can include, but are not limited to, at least one of the following: the initial response contains sensitive words; the initial response meets preset low-quality conditions; or the relevance between the initial response and the input information is lower than a preset relevance threshold. Sensitive words can be queried based on a pre-defined list of sensitive words set by the system. Preset low-quality conditions can include a large number of repetitive statements in the response, missing key information (e.g., the response does not include step-by-step instructions when the user asks for them), and logical contradictions (e.g., giving conflicting suggestions simultaneously). The preset relevance threshold can be determined by calculating indicators such as semantic similarity and intent matching between the initial response and the input information. For example, if a user asks "how to change my password," and the initial response only mentions "account security is important" without mentioning any steps to change the password, the relevance is low, and the constraint will be deemed not to be met.

[0101] Step S150 can be implemented by following steps S151~S152.

[0102] Step S151: If the preliminary response information is detected to fall within the preset constraints, the input information is sent to the human customer service client so that the human client can obtain the response information corresponding to the input information and send the response information to the server.

[0103] When the initial response falls within preset constraints, it indicates that the response may have quality issues or potential risks and is not suitable for direct feedback to the user. In this case, the main control server forwards the input information to a human customer service representative for processing. The human representative can perform a more in-depth analysis and judgment based on the input information and possible contextual relationships, generating an accurate, compliant, and user-expected response manually and sending it back to the server, which then provides feedback to the user. This human-machine collaborative approach effectively compensates for the limitations of artificial intelligence in handling complex scenarios, special needs, or high-risk content, ensuring the quality and security of the final response and improving the user experience.

[0104] Step S152: If it is detected that the preliminary response information does not fall within the preset constraint conditions, the preliminary response information is determined as the response information corresponding to the input information.

[0105] When the initial response information does not fall into the preset constraints after detection, it indicates that it meets the system requirements in terms of content quality, compliance, and relevance. At this time, the initial response information can be directly used as the final response information corresponding to the input information and fed back to the client.

[0106] In one implementation, when there are no services among the candidate services that meet the above preset rules, that is, when the confidence difference between the highest and second-highest capability confidence scores in each candidate service is less than or equal to a preset difference, when there are no services among the candidate services that have processed similar information, when there are no services among the candidate services whose semantic matching degree between the service description information and the input information meets the preset matching conditions, and when there are no services among the candidate services that meet the above preset rules, in this case, it is impossible to select a target service that meets the requirements from among the candidate services. In this case, the response information corresponding to the input information can be generated through a pre-configured fallback scheme.

[0107] The aforementioned fallback solution may include at least one of the following: sending the input information to a human customer service client so that the human client obtains the response information corresponding to the input information and sends the response information to the server; generating the response information corresponding to the input information through a pre-configured content generation service in the server; and randomly selecting one of the candidate services as the target service.

[0108] After the input information is sent to the human customer service client, the human customer service representative can obtain the user's input and, based on their professional knowledge and service experience, enter a response on their client. The human customer service client then sends the response to the main control server. This method effectively handles complex and special scenarios that are difficult for machines to process, ensuring that users receive effective service support under any circumstances, further improving the reliability of overall content generation and user satisfaction.

[0109] The response information corresponding to the input information is generated through a pre-configured content generation service on the server. Specifically, a content generation service is pre-configured as a backup in the main control server. When none of the candidate services meet the selection criteria, this basic content generation service is invoked to generate preliminary response information. The content generation service can be a question-and-answer model capable of providing standardized answers to common questions. While it may not reach the processing depth of specific professional services, it ensures the basic usability of the response information.

[0110] Randomly selecting one of the candidate services as the target service can also avoid long waiting times for users due to service selection stagnation, thus ensuring the continuity of the service process.

[0111] This embodiment, by setting a fallback plan, can still ensure that input information is responded to and processed in a timely manner when all candidate services fail to meet the selection criteria, thus avoiding service interruptions or unanswered user requests.

[0112] In one embodiment, prior to step S130, the above method may further include the following step S130a.

[0113] Step S130a: Based on the input information and the service description information corresponding to each task processing service, determine the second processing capability information of each task processing service for the input information.

[0114] The second processing capability information is the processing capability of the task processing service determined by the main control server for the input information. The main control server can obtain the second processing capability information by analyzing the text content of the service description information, such as functional keywords, processing scenario descriptions, and applicable scope descriptions, and by combining a semantic understanding model to quantitatively evaluate the matching degree between the service description information and the input information; alternatively, the main control server can also determine the relevance and matching degree between the service description information and the input information according to preset rules or models, and use this as the second processing capability information. The second processing capability information can serve as an auxiliary basis for screening candidate services.

[0115] The content of the second processing capability information can be the same type as that of the first processing capability information, for example, both being capability confidence values, so as to facilitate comparison and screening in a unified dimension; or it can include different types of evaluation indicators, such as the relevance score between the service description and the input information in the second processing capability information, and the functional matching level in the first processing capability information, so as to reflect the potential processing capability of the service from multiple perspectives.

[0116] Correspondingly, step S130 can be specifically performed by selecting the target service as follows: step S133.

[0117] Step S133: Based on the first processing capability information and the second processing capability information, select a target service from each of the task processing services whose first and second processing capability information both indicate the ability to process the input information.

[0118] For example, when the first processing capability information is the capability confidence score returned by the task processing service itself, and the second processing capability information is the relevance score obtained by the main control server through semantic matching of service description information, if the preset first processing capability information is 0.7, where a capability confidence score greater than 0.6 indicates the ability to process input information, and the preset second processing capability information is 0.65, where a relevance score greater than 0.7 indicates the ability to process input information, then the first processing capability information of the task processing service meets the condition (0.7>0.6), but the second processing capability information does not reach the preset threshold (0.65<0.7). In this case, the service will not be selected as the target service. Only when the capability confidence score (first processing capability information) of a task processing service is higher than 0.6 and the relevance score (second processing capability information) is higher than 0.7 will it be considered to simultaneously meet the requirements of the first and second processing capability information, and thus be included in the candidate range of target services.

[0119] This dual-dimensional screening mechanism combines the task processing service's assessment of its own capabilities with the main control server's independent judgment based on objective descriptive information. It can effectively avoid the bias that may exist in a single evaluation dimension, significantly improve the accuracy of candidate service screening, and ensure the quality and reliability of subsequent content generation.

[0120] In one embodiment, the above method may further include the following steps S160-S170.

[0121] Step S160: When it is detected that there is a fault state task processing service among the task processing services that meets the preset fault conditions, the circuit breaker process is performed on the fault state task processing service, that is, the invocation of the fault state task processing service is stopped.

[0122] The preset fault conditions include at least one of the following: within a preset time period, the proportion of response content generated by the task processing service that meets the preset unqualified conditions is greater than a preset proportion threshold; within a preset time period, the response time of the task processing service is greater than a preset response time threshold; within a preset number of times, the number of times the task processing service fails to call is greater than a preset failure number threshold.

[0123] For example, if more than 30% of the responses generated by a task processing service within one hour are judged to be low-quality responses (such as missing key information or logical contradictions), that is, the proportion that meets the preset unqualified conditions is greater than the preset proportion threshold (30%), then the circuit breaker will be triggered.

[0124] The purpose of circuit breaking is to isolate abnormal services in a timely manner to prevent them from negatively impacting the stability and processing efficiency of the overall content generation system, and to ensure that the system can continuously and reliably provide services to users.

[0125] In step S120, when calling the intent recognition module of each task processing service, the external task processing service that is not in the circuit breaker state is called.

[0126] Step S170: For a fault-state task processing service that has undergone circuit breaker processing, when it is detected that the task processing service has returned to normal status, the circuit breaker processing of the task processing service is lifted, and the call to the task processing service is resumed.

[0127] The task processing service has returned to normal operation when all its metrics have met normal operating standards within a preset recovery period. For example, after technical personnel have fixed the issue, if the response time of the task processing service is less than a preset response time threshold (e.g., 2 seconds) in 10 consecutive test calls, the generated response content does not fall under preset disqualification conditions, and the number of call failures is 0, then the task processing service can be considered to have returned to normal operation. The system can periodically send test requests to services in circuit breaker mode and check their return results and performance parameters to automatically detect whether the task processing service has recovered. When the task processing service returns to normal operation, the circuit breaker is lifted, allowing it to be called as a task processing service, thereby restoring the overall service capacity and resource utilization of the system.

[0128] This circuit breaker and recovery mechanism can dynamically ensure the health of each task processing service in the system, and improve the system's fault tolerance and stability.

[0129] The content generation method provided in this application integrates multiple task processing services with different professional capabilities through a server, forming a processing system where the server schedules multiple external services to work collaboratively. When faced with user input information, the server can determine the corresponding first processing capability information by evaluating the capabilities of each task processing service's intent recognition module. This allows the server to accurately locate the target service with the capability to process the input information and the most suitable service for processing the current input information. This multi-service collaboration and optimal selection mechanism effectively compensates for the limitations of a single model in terms of capability range and knowledge coverage. By calling the solution module of the target service to generate preliminary response information, the solution module of the task processing service most suitable for processing the complex and diverse input information can be selected for targeted processing, thereby generating preliminary response information that is more in line with user needs and more accurate and effective. Based on this, the server determines the final response information based on the preliminary response information, further ensuring the quality and reliability of the response. Compared with traditional single-model intelligent question answering systems, this application embodiment significantly improves the processing capability and response accuracy for complex and diverse user input through multi-service collaboration and dynamic selection mechanisms, thereby effectively improving the user experience.

[0130] Furthermore, in this application, each task processing service, through modular processing of the intent recognition module and the solution module, can quickly identify whether the input information belongs to its own processing capabilities and quickly generate the corresponding response. This modular design not only improves the response efficiency of each task processing service but also facilitates system expansion and maintenance. Moreover, this application allows for the flexible integration of new task processing services or the removal of unnecessary services based on actual business needs, enabling the system to quickly adapt to content generation requirements in different domains and scenarios, demonstrating strong scalability and adaptability.

[0131] Example 2 The second embodiment of this application also provides a content generation system. Since the system embodiment is basically similar to the method embodiment, it is described simply. For details of the relevant technical features and their effects, please refer to the corresponding descriptions of the content generation method embodiments provided above. The following mainly describes the differences from the above. For example... Figure 1 As shown, the content generation system provided in this embodiment includes an input / output interface system 200, an intelligent gateway 100, and a task processing server cluster 300.

[0132] The input / output interface system 200 is used to allow users to input information and sends the user's input information to the smart gateway 100. The smart gateway 100 is used to invoke the intent recognition modules 301 of each task processing service in the task processing server cluster 300 and send the input information to the intent recognition modules 301. The intent recognition modules 301 of each task processing service in the task processing server cluster 300 are used to determine the first processing capability information of the task processing service for the input information and send the first processing capability information to the smart gateway 100. The smart gateway 100 is also used to select a target service from the task processing services capable of processing the input information based on the first processing capability information and invoke the solution module 302 of the target service. The solution module 302 of the target service is used to generate preliminary response information corresponding to the input information and send the preliminary response information to the smart gateway 100. The smart gateway 100 is also used to determine the response information corresponding to the input information based on the preliminary response information and send the response information to the input / output interface system 200. The input / output interface system 200 is also used to display the response information.

[0133] The input / output interface system 200 may include a client, such as a mobile application interface or a computer webpage interface used by the user. The input / output interface system 200 is used to allow the user to input information, receive the user's input information, and display the received response information to the user. After the user submits input information through the input / output interface system 200, the input / output interface system 200 is used to transmit the input information to the smart gateway 100.

[0134] The smart gateway 100 may integrate the main control server 400 in the above method embodiment. The smart gateway 100 is used to receive input information from the input / output interface system 200, call the intent recognition module 301 of each task processing service in the task processing server cluster 300, and send the input information to the intent recognition module 301 of each task processing service.

[0135] In one specific embodiment, the smart gateway 100 can be used to receive input information from the input / output interface system 200 and send the input information to the main control server 400. The main control server 400 can be used to call the intent recognition module 301 of each task processing service in the task processing server cluster 300 and send the input information to the intent recognition module 301 of each task processing service. That is, the smart gateway 100 realizes the data interaction management between the main control server 400 and the input / output interface system 200.

[0136] The task processing server cluster 300 comprises multiple task processing servers, each configured with a specific task processing service. This can be implemented with one task processing server per service, multiple service deployments on the same server, or multiple servers sharing a single service. The specific deployment method can be flexibly configured based on actual resource and performance requirements. For example, a resource-intensive task processing service can be deployed separately on a high-performance server, or different processing modules can be deployed on different servers, with multiple servers handling the resource-intensive task. Conversely, multiple simpler, less resource-intensive task processing services can be combined and deployed on the same server to improve hardware resource utilization.

[0137] The task processing service deployed in the task processing server cluster 300 includes an intent recognition module 301 and a solution module 302. The intent recognition module 301 analyzes the input information, evaluates the corresponding task processing service's ability to process the input information, and generates initial processing capability information. The solution module 302 generates specific preliminary response information based on the input information when invoked by the smart gateway 100 or the main control server 400.

[0138] After receiving the first processing capability information returned by the intent recognition module 301 of each task processing service, the intelligent gateway 100 or the main control server 400 filters out the target service and calls its solution module 302. The solution module 302 of the target service generates preliminary response information and returns it to the intelligent gateway 100. The intelligent gateway 100 then determines the final response information based on the preliminary response information and presents it to the user through the input / output interface system 200. The task processing services configured in the task processing server cluster 300 may include various services external to the intelligent gateway 100.

[0139] The above system architecture centrally schedules the various task processing services in the task processing server cluster 300 through the intelligent gateway 100, realizing the efficient integration and utilization of multi-source service resources and ensuring the professionalism, accuracy and efficiency of content generation services.

[0140] Optionally, such as Figure 1 As shown, the task processing server cluster 300 may also include basic task processing services configured within the smart gateway 100. Figure 1 (The term "basic service" is used here.) For example, a basic task processing service can be integrated within the smart gateway 100. This basic task processing service may include a basic intent recognition module 301 and a basic solution module 302. The basic intent recognition module 301 of the basic task processing service is used to determine the second processing capability information of each external task processing service for the input information based on the input information and the service description information corresponding to each external task processing service. The basic solution module of the basic task processing service is configured with a content generation service for generating response information corresponding to the input information. This serves as a fallback solution to generate response information when none of the external task processing services can meet the processing requirements, thus avoiding situations where user input information goes unanswered. This design, which integrates some basic services into the smart gateway 100, can optimize the allocation and use of system resources while ensuring processing efficiency, further enhancing the flexibility and practicality of the system architecture.

[0141] In one implementation, such as Figure 1 As shown, the content generation system may also include a monitoring server 500, which is used to detect at least one of the following during the content generation process: whether the preliminary response information falls under preset constraints, whether the input information meets preset conditions for manual transfer, whether there are fault state task processing services that meet preset fault conditions in each task processing service, and whether the fault state task processing service that has performed circuit breaker processing has returned to normal status.

[0142] The intelligent gateway 100 is specifically used to generate response information based on the first detection result of the monitoring server 500, and to determine whether to perform a circuit breaker or restore normal operation between the gateway and the task processing service based on the second detection result of the monitoring server 500. The first detection result includes at least one of the detection results of whether the preliminary response information falls within preset constraints and whether the input information meets preset conditions for manual intervention. The second detection result includes the detection result of whether the task processing service meets preset fault conditions and the detection result of whether the fault-state task processing service that has undergone circuit breaker processing has returned to normal operation.

[0143] In one implementation, such as Figure 1 As shown, the content generation system may also include a human customer service system 600; the smart gateway 100 is specifically used to send the input information to the human customer service system 600 when it detects that the preset conditions for transferring to human assistance are met.

[0144] The preset conditions for transferring to human assistance include at least one of the following: the initial response information falls within preset constraints, the input information is detected to meet the preset conditions for transferring to human assistance, and the target service cannot be determined in any of the external task processing servers. The manual customer service system 600 can specifically be a manual customer service terminal, used to display the input information, obtain the reply information corresponding to the input information entered by the customer service personnel, and send the reply information to the smart gateway 100.

[0145] The 600 customer service system enables timely human intervention when the system's automatic processing mechanism is unable to effectively handle user input, ensuring that user needs are properly addressed.

[0146] The execution logic of each device in the content generation system provided in this embodiment is consistent with the execution logic of the corresponding steps in the above method embodiment. For example, the logic of the smart gateway 100 in executing operations such as calling the intent recognition module 301 of the task processing service, selecting the target service, calling the solution module 302, and determining the final response information based on the preliminary response information is the same as the execution logic of steps S110 to S160 in the method embodiment. The logic of the monitoring server 500 in executing operations such as detecting whether the preliminary response information falls within the constraint conditions, detecting whether the task processing service meets the fault conditions, and detecting whether the circuit breaker service has been restored is the same as the logic of the monitoring and detection involved in steps S130, S140, S120, and S170 in the method embodiment. The specific execution process of each device can be referred to the first embodiment, and will not be repeated here. Through collaborative work, each device jointly realizes the functions of multi-service collaboration, dynamic selection, quality monitoring, fault isolation and recovery of the content generation system, thereby improving the system's ability to process complex and diverse user inputs, response accuracy, and overall stability and reliability.

[0147] The following specific examples illustrate the content generation process of this application.

[0148] Example 1 like Figure 3 As shown, the content generation method in this example includes the following steps 1 to 10.

[0149] Step 1: The client sends a user request containing the user's input information to the smart gateway.

[0150] Users can send user requests to the smart gateway through various channels, such as clients, web pages, and mini-programs. Step 2: The smart gateway routes the user request containing the input information to the main control server.

[0151] Step 3: After receiving the user request, the main control server obtains the service list of each external task processing service through its service scheduler, and calls the intent recognition module of each task processing service in the access state in the service list in parallel, and sends the input information to each intent recognition module.

[0152] Step 4: The intent recognition module determines the first processing capability information of the input information and returns it to the main control server.

[0153] Step 5: The decision engine of the main control server selects the target service based on the first processing capability information.

[0154] Step 6: The main control server calls the solution module of the target service to generate preliminary response information to the input information.

[0155] Step 7: The monitoring server monitors in real time whether the initial response information falls within the preset constraints and whether the input information meets the preset conditions for manual intervention.

[0156] Step 8: If the initial response information falls into the preset constraints, or the input information meets the preset conditions for transferring to human intervention, or the target service cannot be determined, the fallback processor generates the response information corresponding to the input information through the pre-configured fallback scheme.

[0157] The fallback plan includes human customer service processing and demotion processing. Demotion processing involves generating a response corresponding to the input information through a pre-configured content generation service on the server, or randomly selecting a target service from the task processing services.

[0158] Step 9: If a preliminary response is successfully generated and it is detected that the preliminary response does not fall under the preset constraints, then the preliminary response is determined as the response.

[0159] Step 10: Adapt the response information to the front-end rendering using the response compositor and send it to the front-end client for display.

[0160] The specific execution process of each step in this example can be found in the relevant content of the first and second embodiments, and will not be repeated here.

[0161] Example 2 This example mainly describes the process by which the active server determines the target service. The remaining processes are basically the same as in Example 1, and the similarities will not be repeated. Figure 4 As shown, the content generation method in this example includes the following steps a~n.

[0162] Step a: The main control server receives the user's input information.

[0163] Step b: Retrieve all connected task processing services from the list of registered services.

[0164] Step c: Call the intent recognition module of each task processing service to output the first processing capability information of the input information and return it to the main control server.

[0165] Step d: The active server obtains the first processing capability information of each task processing service.

[0166] The first processing capability information includes the judgment result of whether the task processing service can process the input information, and the confidence level of its ability to process the input information.

[0167] Step e: Determine the number of task processing services that can be processed based on the judgment result.

[0168] Step f: If the number of task processing services that can be processed is 0, then execute the fallback plan to determine the response information.

[0169] Step g: If there is only one task processing service (i.e., candidate service) that can be processed, then the task processing service is determined as the target service.

[0170] Step h: If there are multiple task processing services (i.e., candidate services) that can be processed, a preset decision strategy is triggered, and the target service is selected from the candidate services according to the decision strategy.

[0171] The preset decision-making strategies include multiple decision-making strategies with different priorities, specifically determining the target service according to the priority of steps h1 to h4.

[0172] h1: When the confidence difference between the highest and second-highest capability confidence scores among the candidate services is greater than a preset difference, the service with the highest confidence score will be selected as the target service.

[0173] Figure 4 The preset difference is 0.2.

[0174] h2: If the confidence difference between the highest and second-highest capability confidence scores among the candidate services is less than or equal to a preset difference, then the service that has processed similar information is selected as the target service.

[0175] h3: If there is no service among the candidate services that has processed similar information, then the service description information of each candidate service is semantically matched with the semantics of the input information, and the service with the highest semantic matching degree that meets the preset matching conditions is selected as the target service.

[0176] h4: If there is no service among the candidate services whose semantic matching degree between the service description information and the input information meets the preset matching conditions, then select the service that meets the preset rules from among the candidate services as the target service.

[0177] Step i: The solution module of the target service generates a response message for the input information and returns it to the main control server.

[0178] Step j: The main control server sends the reply information to the smart gateway, so that the smart gateway can convert the reply information into a preset format or render it before sending it to the client for display.

[0179] Example 3 The third embodiment of this application also provides a content generation apparatus. Since the apparatus embodiment is basically similar to the method embodiment, it is described simply. For details of the relevant technical features and their effects, please refer to the corresponding descriptions of the content generation method embodiments provided above. Figure 5As shown, the content generation apparatus provided in this embodiment includes: Acquisition unit 310 is used to acquire input information; The calling unit 320 is used to call the intent recognition module of each task processing service to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server. Selection unit 330 is configured to select a target service from each of the task processing services capable of processing the input information based on the first processing capability information; The calling unit is also used to call the solution module of the target service to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; The determining unit 340 is used to determine the response information corresponding to the input information based on the preliminary response information.

[0180] The fourth embodiment of this application also provides an electronic device embodiment corresponding to the content generation method provided in the first embodiment. This electronic device is a server, and the following description of the electronic device embodiment is merely illustrative. The electronic device embodiment is as follows: Please refer to Figure 6 Understanding the above electronic devices, Figure 6 This is a schematic diagram of an electronic device. The electronic device provided in this embodiment includes: a processor 1001, a memory 1002, a communication bus 1003, and a communication interface 1004; The memory 1002 is used to store computer instructions for data processing. When these computer instructions are read and executed by the processor 1001, the following steps are performed: Obtain input information; The intent recognition module of each task processing service is invoked to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; Based on the first processing capability information, a target service is selected from each of the task processing services that have the capability to process the input information; The solution module of the target service is invoked to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; The response information corresponding to the input information is determined based on the preliminary response information.

[0181] The fourth embodiment of this application also provides a computer-readable storage medium for implementing the method described in the first embodiment. The embodiments of the computer-readable storage medium provided in this application are described in a relatively simple manner; relevant parts can be found in the corresponding descriptions of the above method embodiments. The embodiments described below are merely illustrative.

[0182] The computer-readable storage medium provided in this embodiment stores computer instructions, which, when executed by a processor, perform the following steps: Obtain input information; The intent recognition module of each task processing service is invoked to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; Based on the first processing capability information, a target service is selected from each of the task processing services that have the capability to process the input information; The solution module of the target service is invoked to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; The response information corresponding to the input information is determined based on the preliminary response information.

[0183] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0184] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0185] 1. Computer-readable media includes both permanent and non-permanent, removable and non-removable media, which can store information by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined in this application, computer-readable media does not include non-transitory computer-readable media, such as modulated data signals and carrier waves.

[0186] 2. Those skilled in the art will understand that embodiments of this application can provide methods, systems, or computer program products. Therefore, embodiments of this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] 3. This application embodiment may involve the use of user data. In practical applications, user-specific personal data may be used within the scope permitted by applicable laws and regulations of the country in which the application is located (e.g., with the user's explicit consent and effective notification to the user, etc.). Furthermore, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. The collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0188] Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be determined by the scope defined in the claims of this application.

Claims

1. A content generation method, characterized in that, Applied to a server, the method includes: Obtain input information; The intent recognition module of each task processing service is invoked to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; Based on the first processing capability information, a target service is selected from each of the task processing services that have the capability to process the input information; The solution module of the target service is invoked to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; The response information corresponding to the input information is determined based on the preliminary response information.

2. The content generation method according to claim 1, characterized in that, The first processing capability information includes at least the confidence level of the task processing service's ability to process the input information; The step of selecting a target service from the task processing services capable of processing the input information based on the first processing capability information includes: From each of the task processing services, select the candidate service whose judgment result for whether it can process the input information is yes, and whose capability confidence is greater than a preset confidence threshold; Among the candidate services, the target service is selected according to a predetermined selection method.

3. The content generation method according to claim 1, characterized in that, The intent recognition module performs intent recognition through a generative artificial intelligence system. Determining the first processing capability information of the task processing service for the input information includes: The intent recognition module is provided with at least the input information and the associated information of the input information; The intent recognition module determines the relationship between the intent of the input information and the processing capability of the task processing service based on the information provided to it and the pre-set task setting statement. Based on the discrimination result, at least one of the following is obtained: the confidence level of the task processing service's ability to process the input information, and the judgment result of whether the task processing service can process the input information.

4. The content generation method according to claim 2, characterized in that, The selection of a target service from the candidate services is based on a predetermined selection method, including the following selection methods: When there are multiple candidate services, the candidate service with the highest capability confidence is selected as the target service. When there is one candidate service, the candidate service is determined as the target service.

5. The content generation method according to claim 2, characterized in that, Selecting a target service from the candidate services according to a predetermined selection method includes: When there are multiple candidate services, and the confidence difference between the highest and second-highest capability confidence in each candidate service is greater than a preset difference, the candidate service with the highest capability confidence is selected as the target service. When there are multiple candidate services, and the confidence difference between the highest and second-highest capability confidence in each candidate service is less than or equal to a preset difference, the service that has processed similar information is selected as the target service from among the candidate services, and the similarity between the similar information and the input information is greater than a preset similarity threshold.

6. The content generation method according to claim 2, characterized in that, Selecting a target service from the candidate services according to a predetermined selection method includes: When there are multiple candidate services, and the confidence difference between the highest and second-highest capability confidence scores in each candidate service is less than or equal to a preset difference, further, if none of the candidate services have processed similar information, then the service description information of each candidate service is semantically matched with the semantics of the input information, and the service whose semantic matching degree meets the preset matching condition is selected as the target service; if none of the candidate services have a service whose service description information and the semantic matching degree meet the preset matching condition, then the service that meets the preset rule is selected from the candidate services as the target service, and the similarity between the similar information and the input information is greater than a preset similarity threshold.

7. The content generation method according to claim 6, characterized in that, The method further includes: If none of the candidate services meet the preset rules, then the response information corresponding to the input information is generated through a pre-configured fallback scheme.

8. The content generation method according to claim 7, characterized in that, The fallback solution includes at least one of the following: The input information is sent to a human customer service client so that the human client can obtain the response information corresponding to the input information and send the response information to the server; The response information corresponding to the input information is generated by the content generation service pre-configured in the server. Randomly select one of the candidate services as the target service.

9. The content generation method according to any one of claims 1 to 8, characterized in that, Also includes: Based on the input information and the service description information corresponding to each task processing service, determine the second processing capability information of each task processing service for the input information; Selecting a target service from each of the task processing services capable of processing the input information includes: selecting a target service from each of the task processing services based on the first processing capability information and the second processing capability information, wherein both the first processing capability information and the second processing capability information indicate that the target service is capable of processing the input information.

10. The content generation method according to claim 1, characterized in that, Before determining the response information corresponding to the input information based on the preliminary response information, the method further includes: Detect whether the preliminary response information falls within the preset constraints; The step of determining the response information corresponding to the input information based on the preliminary response information includes: If the initial response information is detected to fall within the preset constraints, the input information is sent to the human customer service client so that the human client can obtain the response information corresponding to the input information and send the response information to the server. If the preliminary response information is detected to not fall within the preset constraints, the preliminary response information is determined to be the response information corresponding to the input information.

11. The content generation method according to claim 1, characterized in that, Before the intent recognition module that invokes each task processing service determines the first processing capability information of the task processing service for the input information, the method further includes: Detect whether the input information meets the preset conditions for transferring to human intervention; If the conditions are met, the input information is sent to the human customer service client so that the human client can obtain the response information corresponding to the input information and send the response information to the server. If the conditions are not met, then the step of calling the intent recognition module of each task processing service to determine the first processing capability information of the task processing service for the input information is executed.

12. The content generation method according to claim 11, characterized in that, The preset conditions for transferring to a human operator include at least one of the following: the input information indicates that a human customer service representative will respond; the input information contains sensitive information; or the input information indicates that a negative emotion is present.

13. The content generation method according to claim 1, characterized in that, The method further includes: When it is detected that there is a fault state task processing service among the task processing services that meets the preset fault conditions, the circuit breaker is executed on the fault state task processing service, that is, the invocation of the fault state task processing service is stopped. For a faulty task processing service that has undergone circuit breaker processing, when it is detected that the task processing service has returned to normal, the circuit breaker processing for the task processing service is lifted, and the call to the task processing service is resumed.

14. The content generation method according to claim 13, characterized in that, The preset fault conditions include at least one of the following: within a preset time period, the proportion of response content generated by the task processing service that meets the preset unqualified conditions is greater than a preset proportion threshold; within a preset time period, the response time of the task processing service is greater than a preset response time threshold; within a preset number of times, the number of times the task processing service fails to call is greater than a preset failure number threshold.

15. The content generation method according to claim 1, characterized in that, In the step of calling the solution module of the target service to generate preliminary response information corresponding to the input information, the solution module of the target service generates information through a generative artificial intelligence system. The generation of preliminary response information corresponding to the input information includes: The solution module shall provide the following information: at least the input information and the associated information of the input information; The solution module performs semantic understanding on the input information based on the information provided to it and a pre-trained dialogue generation model, and generates preliminary response information corresponding to the input information.

16. A content generation system, comprising: Input / output interface system, intelligent gateway, task processing server cluster; The input / output interface system is used to allow users to input information and send the user's input information to the smart gateway; The smart gateway is used to call the intent recognition module of each task processing service in the task processing server cluster, and send the input information to the intent recognition module; The intent recognition module of each task processing service in the task processing server cluster is used to determine the first processing capability information of the task processing service for the input information, and send the first processing capability information to the smart gateway. The smart gateway is also configured to select a target service from the task processing services that have the ability to process the input information based on the first processing capability information, and call the solution module of the target service; The solution module of the target service is used to generate preliminary response information corresponding to the input information and send the preliminary response information to the smart gateway; The smart gateway is also used to determine the response information corresponding to the input information based on the preliminary response information, and send the response information to the input / output interface system; The input / output interface system is also used to display the response information.

17. The content generation system according to claim 16, characterized in that, Also includes: The monitoring server is used to detect at least one of the following during the content generation process: whether the preliminary response information falls under preset constraints, whether the input information meets preset conditions for transferring to manual processing, whether there are fault state task processing services that meet preset fault conditions, and whether the fault state task processing services that have performed circuit breaker processing have returned to normal status. The smart gateway is specifically used to generate response information based on the first detection result of the monitoring server and to determine whether to disconnect or restore normal operation between the gateway and the task processing service based on the second detection result of the monitoring server. The first detection result includes at least one of the detection results of whether the preliminary response information falls within preset constraints and whether the input information meets preset conditions for manual intervention. The second detection result includes the detection results of whether the task processing service meets preset fault conditions and whether the fault-state task processing service that has undergone circuit breaker processing has returned to normal operation.

18. The content generation system according to claim 17, characterized in that, Also includes: Human customer service system; The smart gateway is specifically used to send the input information to the human customer service system when it detects that the preset conditions for transferring to human assistance are met. The preset conditions for transferring to human assistance include at least one of the following: the initial response information falls into preset constraints, the input information is detected to meet the preset conditions for transferring to human assistance, and the target service cannot be determined in any of the external task processing servers. The human customer service system is used to display the input information, obtain the response information corresponding to the input information entered by the customer service personnel, and send the response information to the smart gateway.

19. A content generation apparatus, characterized in that, Applied to a server, the device includes: The acquisition unit is used to acquire input information; The calling unit is used to call the intent recognition module of each task processing service to determine the first processing capability information of the task processing service for the input information, so that the intent recognition module sends the first processing capability information to the server; The selection unit is configured to select a target service from each of the task processing services capable of processing the input information, based on the first processing capability information. The calling unit is also used to call the solution module of the target service to generate preliminary response information corresponding to the input information, so that the solution module sends the preliminary response information to the server; A determining unit is used to determine the response information corresponding to the input information based on the preliminary response information.

20. An electronic device, characterized in that, include: Processor, memory, and computer program instructions stored in said memory and executable on the processor; When the processor executes the computer program instructions, it implements the method as described in any one of claims 1-15.

21. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-15.

Images