Dialogue processing method, dialogue interaction method, electronic equipment, system, storage medium and program product

Abstract

The embodiment of the invention provides a dialogue processing method, a dialogue interaction method, electronic equipment, a dialogue processing system, a dialogue interaction system, a storage medium and a program product. In the dialogue processing method, under the condition that a plurality of task agents of different task types exist, the plurality of task agents can be utilized to perform interactive perception on task processing results of each other, and a single task agent can update a processing result of the single task agent based on perceived processing results of other tasks. Therefore, the correlation between processing results of different tasks can be effectively improved. Besides, a single task agent adopts a mode of gradually constructing the target processing result, after the processing result is generated, an updating mode is adopted to optimize the generated processing result by utilizing other processing results instead of starting to generate the target processing result after other processing results are received, so that the generation speed of the target processing result can be effectively improved, and the generation efficiency of the target processing result is improved. The response fluency of the inquiry request is improved, and the response delay effect is relieved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a dialogue processing, interaction method, electronic device, system, storage medium, and program product. Background Technology

[0002] Advances in natural language processing (NLP) technology have continuously improved the performance of dialogue processing systems. An existing dialogue processing system may provide a question-answering module and a recommendation module. The question-answering module outputs a question-and-answer result based on prompts, while the recommendation module outputs personalized recommendations based on the input prompts. However, the question-and-answer results and recommendations are independent of each other, exhibiting poor content relevance. In an improved dialogue processing system, the recommendation module generates recommendations based on the question-and-answer results after the question-and-answer module outputs them. While this approach increases the relevance between the question-and-answer results and recommendations, the recommendation results are only generated after the question-and-answer results are output, resulting in a significant response delay.

[0003] Therefore, how to balance the relevance and response effectiveness of both the consultation and recommendation results is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] This application provides a dialogue processing, interaction method, electronic device, system, storage medium, and program product to improve the relevance of processing results for different types of tasks while reducing response latency when outputting processing results for different tasks.

[0005] This application provides a dialogue processing method, including: responding to an inquiry request and obtaining inquiry condition information; utilizing multiple task agents corresponding to different task types to obtain target processing results based on the inquiry condition information; among the multiple task agents, any first task agent is used to: generate a first processing result by calling a preset large language model based on the inquiry condition information; and to receive, in real time, a second processing result generated by a second task agent among the multiple task agents based on the inquiry condition information, and update the first processing result based on the second processing result to obtain the target processing result corresponding to the first task agent; and outputting a response result corresponding to the inquiry request based on the target processing results obtained by the multiple task agents respectively.

[0006] Optionally, obtaining query condition information includes: obtaining target query information corresponding to the query request; retrieving a preset knowledge base based on the target query information to obtain target knowledge text; and generating the query condition information based on the target knowledge text and the target query information.

[0007] Optionally, updating the first processing result based on the second processing result to obtain the target processing result corresponding to the first task agent includes: when the correlation between the second processing result and the first processing result is less than a set threshold, updating the first processing result based on the second processing result to obtain the target processing result corresponding to the first task agent.

[0008] Optionally, when the correlation between the second processing result and the first processing result is less than a set threshold, the first processing result is updated according to the second processing result to obtain the target processing result corresponding to the first task agent. This includes: constructing an update prompt word based on the first processing result, the second processing result, and the task description corresponding to the task class of the first task agent; inputting the update prompt word into a preset large language model to obtain an updated first processing result, which is used as the target processing result corresponding to the first task agent; wherein, the update prompt word is used to prompt the large language model to calculate the correlation between the first processing result and the second processing result, and to update the first processing result according to the second processing result when the correlation is less than a set threshold.

[0009] Optionally, it further includes: in the process of constructing the updated prompt word, using the first processing result as a fixed context and the second processing result as a dynamic context.

[0010] Optionally, it further includes: using the first task agent to send the service registration information of the first task agent to the second task agent, so that the second task agent sends the second processing result to the first task agent; and receiving the service registration information sent by the second task agent, so that the first task agent sends the first processing result to the second task agent.

[0011] Optionally, it further includes: responding to the operation of adding a new task agent, obtaining a newly added third task agent; using the third task agent, sending the service registration information of the third task agent to the plurality of task agents, so that the plurality of task agents send processing results to the third task agent; and receiving the service registration information sent by the plurality of task agents, so that the third task agent sends processing results to the plurality of task agents.

[0012] This application embodiment also provides a dialogue processing method, including: responding to an inquiry request and obtaining inquiry condition information; using a question-answering task agent and a recommendation task agent to obtain a target question-answering result and a target recommendation result respectively based on the inquiry condition information; wherein, the question-answering task agent is used to: call a preset large language model to obtain an initial question-answering result based on the inquiry condition information, and provide the initial question-answering result to the recommendation task agent; and receive the initial recommendation result provided by the recommendation task agent, and update the initial question-answering result based on the initial recommendation result to obtain a target question-answering result; the recommendation task agent is used to: call a preset large language model to obtain an initial recommendation result based on the inquiry condition information, and provide the initial recommendation result to the task agent; and receive the initial question-answering result provided by the question-answering task agent, and update the initial recommendation result based on the initial question-answering result to obtain a target recommendation result; and output the response result corresponding to the inquiry request based on the target question-answering result and the target recommendation result.

[0013] This application embodiment also provides a dialogue interaction method, including: responding to an inquiry operation initiated through a first dialogue interaction interface and obtaining target inquiry information; sending an inquiry request to a server based on the target inquiry information, so that the server returns multiple target processing results based on the inquiry request, the multiple target processing results being obtained by the dialogue processing method provided by this application embodiment; and outputting a response result corresponding to the inquiry operation through a second dialogue interaction interface based on the multiple target processing results.

[0014] This application embodiment also provides a dialogue system, including: a terminal device and a server; the terminal device is configured to: respond to an inquiry operation initiated through a first dialogue interaction interface and obtain target inquiry information; send an inquiry request to the server according to the target inquiry information; receive multiple target processing results returned by the server according to the inquiry request, and output a response result corresponding to the inquiry operation through a second dialogue interaction interface based on the multiple target processing results; the server is configured to: respond to the inquiry request and obtain inquiry condition information; utilize multiple task agents of different task types to obtain target processing results according to the inquiry condition information respectively; among the multiple task agents, any first task agent is configured to: call a preset large language model to generate a first processing result according to the inquiry condition information; and receive in real time a second processing result generated by a second task agent among the multiple task agents according to the inquiry condition information, and update the first processing result according to the second processing result to obtain the target processing result corresponding to the first task agent; and return the target processing results obtained by each of the multiple task agents to the terminal device.

[0015] This application also provides an electronic device, including: a memory and a processor; the memory is used to store one or more computer instructions; the processor is used to execute the one or more computer instructions to perform the steps in the method provided in this application.

[0016] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps in the method provided in this application.

[0017] This application also provides a computer program product, including: a computer program / instructions, which, when executed by a processor, can implement the steps in the method provided in this application.

[0018] In the dialogue processing method provided in this application embodiment, when there are multiple task agents of different task types, the multiple task agents can interactively perceive each other's task processing results. A single task agent can update its own processing result based on the perceived processing results of other tasks, thereby effectively improving the correlation between the processing results of different tasks. Secondly, in this implementation, the multiple task agents are interdependent, thus reducing the time difference between different task agents in outputting the target processing result. In addition, a single task agent adopts a step-by-step approach to construct the target processing result. After generating the processing result, it uses an update method to optimize the generated processing result using other processing results, rather than waiting to receive other processing results before starting to generate the target processing result. This effectively improves the generation speed of the target processing result, enhances the smoothness of response to query requests, and alleviates the effect of response latency. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 A flowchart illustrating a dialogue processing method provided in an exemplary embodiment of this application;

[0021] Figure 2 This is a schematic diagram illustrating message interaction between multiple task agents provided in an exemplary embodiment of this application;

[0022] Figure 3 A schematic diagram of the framework structure of a dialogue processing apparatus provided in an exemplary embodiment of this application;

[0023] Figure 4 A schematic diagram of the structure of an electronic device provided in an exemplary embodiment of this application. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one.

[0026] It should be understood that the term "and / or" used in this article 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. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0027] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0028] A dialogue system is a computer system capable of interacting with users through natural language, simulating human conversation. Dialogue systems are typically implemented using techniques such as Natural Language Processing (NLP), Information Retrieval (IR), and Machine Learning (ML). Their goal is to return answers based on input prompts to help users complete tasks, obtain information, or solve problems. Prompts are text used to guide a large language model to output results according to set conditions or instructions. Prompts can be simple instructions or complex text containing background information, context, and expected output styles.

[0029] In an existing dialogue processing system, a question-answering module and a recommendation module are provided. The question-answering module outputs question-answering results based on prompts, and the recommendation module outputs personalized recommendations based on input prompts. However, the question-answering results and recommendation results output for prompts are independent of each other, resulting in poor content relevance. In an improved dialogue processing system, after the question-answering module outputs question-answering results for prompts, the recommendation module can generate recommendation results based on the question-answering results. Although this approach increases the relevance between question-answering results and recommendation results, the recommendation results are only generated after the question-answering results are output, thus exhibiting a significant response delay. To address the technical problem that the above-mentioned dialogue system cannot simultaneously balance the relevance of question-answering results and recommendation results with response timeliness, some embodiments of this application provide a solution. The technical solutions provided by various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0030] Figure 1 This is a flowchart illustrating a dialogue processing method provided in an exemplary embodiment of this application. The method may include, for example: Figure 1 The steps shown are as follows:

[0031] Step 101: Respond to the query request and obtain the query condition information.

[0032] Step 102: Using multiple task agents corresponding to different task types, obtain the target processing result according to the query condition information respectively; among the multiple task agents, any first task agent is used to: call a preset large language model to generate a first processing result according to the query condition information; and receive a second processing result generated by a second task agent among the multiple task agents according to the query condition information, and update the first processing result according to the second processing result to obtain the target processing result corresponding to the first task agent.

[0033] Step 103: Based on the target processing results obtained by the multiple task agents, output the response results corresponding to the query request.

[0034] This embodiment can be executed by a dialogue processing device, which can run on a terminal device or on a server, where the server can be a conventional server or a cloud server. In some embodiments, the dialogue processing device may run partly on the terminal device and partly on the server; this embodiment is not limited to this. The dialogue processing device provides dialogue processing services, can receive external query requests, respond to the query requests using a Large Language Model (LLM), and return the response results. Based on this, the dialogue processing device can also be called a dialogue server. The Large Language Model is a large-scale trained Natural Language Processing (NLP) model. Large Language Models are typically built based on deep learning techniques and trained on large-scale training datasets, thus exhibiting powerful performance in processing natural language tasks. The number of parameters in a Large Language Model is greater than a set threshold, which is typically on the order of millions or billions.

[0035] The dialogue processing device can be applied to question-and-answer scenarios in various fields, such as customer service, e-commerce, teaching assistance, financial / policy / legal consultation, and intelligent search. In these application scenarios, the dialogue processing device can be deployed as different forms of products, such as customer service chatbots, virtual digital humans, intelligent search tools, and intelligent dialogue tools. These different forms of products can use the methods provided in the embodiments of this application to automatically answer input questions or search for target information based on input content.

[0036] In step 101, the query request can be initiated by the dialogue client or by other applications or devices besides the client; this embodiment does not impose any limitations. For example, in some embodiments, the dialogue client can provide a human-computer interaction interface and receive user query instructions through the human-computer interaction interface. If a user query instruction is received, the dialogue client can initiate a query request to the dialogue processing device. The human-computer interaction interface may include at least one of the following: graphical user interface (GUI), text user interface (TUI), voice user interface (VUI), touch user interface (TUI), gesture user interface (Gestural UI), keyboard and mouse interface, and sensor and biometric interface. For example, in some embodiments, a data analysis application running on a device or server, when performing an analysis task on specified data, can initiate a query request to the dialogue processing device so that the dialogue processing device can analyze the specified data and return the analysis results to the data analysis application. For example, in some embodiments, when a timed event (such as an alarm) occurs on the device where the dialogue processing device is located or on the client device, the operating system of the device can send an inquiry request to the dialogue processing device. The inquiry request may be a request for information such as weather conditions, news summaries, and traffic conditions, so as to provide users with personalized timed reminders.

[0037] The query condition information is obtained based on the query request. This query condition information may include at least the target query information input by the user. In addition to the target query information, the query condition information may also include one or more of the following: preprocessing results of the question, semantic understanding results of the question, and knowledge retrieval results.

[0038] In one optional embodiment, the dialogue processing device can obtain the target query information corresponding to the query request and directly use the target query information as query condition information. The target query information can be a question, request, or query description information raised by the questioner or responder, used to guide the dialogue processing device to find an answer or perform a corresponding operation.

[0039] In some alternative embodiments, the dialogue processing device may utilize Retrieval-Augmented Generation (RAG) technology to enhance the target query information corresponding to the query request, thereby obtaining query condition information. This will be explained in detail below. Optionally, the target query information corresponding to the query request may be obtained, and a preset knowledge base may be searched based on the target query information to obtain target knowledge text. Query condition information is then generated based on the target knowledge text and the target query information.

[0040] The pre-defined knowledge base is obtained by pre-processing a large amount of knowledge text. This knowledge text refers to textual materials containing specific information, data, facts, or concepts. Knowledge text is used to convey knowledge, explain concepts, provide guidance, or illustrate a process. In this embodiment, after obtaining the knowledge text, pre-processing can be performed to obtain the knowledge base. The knowledge text can be obtained by extracting text data from various formats of articles, web pages, copywriting, and other knowledge carriers. After obtaining the knowledge text, the knowledge text data can be cleaned, including deduplication, noise removal, format standardization, and stop word removal. After cleaning, the knowledge text can be cut into fixed-length paragraphs according to fixed lengths and punctuation marks, and each paragraph can be numbered. Then, based on the punctuation marks, each paragraph can be processed into multiple clauses, and each clause can be assigned a number. After obtaining the paragraphs and clauses, each paragraph can be encoded into a fixed-length vector, and each clause can be numbered into a fixed-length vector. Based on the paragraph vectors, paragraph numbers, clause vectors, and clause numbers, an index of the knowledge text can be constructed.

[0041] Optionally, when searching a preset knowledge base based on the target query information, the target query information can be vectorized to obtain a question-and-answer vector. Then, the question-and-answer vector is compared with the paragraph vectors and clause vectors in the knowledge text index to calculate the vector similarity, resulting in a matching score between the question-and-answer vector and the paragraph and clause vectors. Based on this matching score, a specified number of knowledge texts can be recalled. In some embodiments, a large number of knowledge texts can be coarsely recalled, and further filtered and sorted based on the coarse recall to obtain a small number of finely ranked knowledge texts with higher precision and more accurate matching scores. For example, the top 20-100 knowledge texts with the highest matching scores can be coarsely recalled, and the top 1-5 knowledge texts can be finely ranked as target knowledge texts. After obtaining the target knowledge text based on the above implementation method, query condition information can be generated based on the target knowledge text and the target query information.

[0042] Based on this implementation method, knowledge texts related to the target query information are retrieved and provided as additional information sources to the large language model, which helps the large language model generate more accurate answers.

[0043] In this embodiment, the dialogue processing device may include multiple task agents corresponding to different task types, and may invoke multiple task agents to implement dialogue processing. After obtaining the query condition information, the dialogue processing device may execute step 102, providing the query condition information to multiple task agents corresponding to different task types. A task agent is an autonomous software entity used to perceive information about the surrounding environment and autonomously make decisions and take actions based on the perceived information.

[0044] The task type is determined based on the nature of the problem that the task agent solves. The task type for any task agent can include: question-and-answer, recommendation, data visualization, multilingual conversion, or multimodal retrieval.

[0045] Among them, the question-answering type task agent (hereinafter referred to as the question-answering task agent) is used to obtain target query information, use natural language processing, knowledge base, machine learning and other technologies to understand the target query information, and return accurate and useful question-answering results.

[0046] Recommendation-type task agents (hereinafter referred to as recommendation task agents) focus on providing suggestions or recommendations. Based on the input target query information, they return relevant recommended content, which may be topics or content that the user is interested in. Recommendation task agents can combine users' historical question-and-answer data, preferences, etc., using recommendation algorithms and data mining techniques to provide personalized recommended content to guide users in making new choices. For example, if the user's input target query information is question A0, the recommendation task agent could return related questions such as question A1, question A2, and question A3 to further help the user explore information of interest.

[0047] Data visualization processing task agents (hereinafter referred to as data visualization task agents) are used to return visual information such as images, graphs, or charts based on the input target query information. For example, if the target query information is: "What is the sales trend of a certain company in the third quarter?", the data visualization task agent can return a sales trend chart for the company in the third quarter, as well as a comparison chart of sales changes between the third quarter and the second quarter.

[0048] A multilingual translation task agent (hereinafter referred to as a multilingual task agent) is used to return translations in multiple languages ​​based on the input target query information. The multilingual task agent returns answers in different languages ​​to meet the question-and-answer needs of users who use different languages. For example, the target query information is: Please explain the meaning of the specified text by considering the cultural attributes of different types of languages. The multilingual task agent can return the interpretation results of the specified text in different languages.

[0049] Multimodal retrieval task agents (hereinafter referred to as multimodal retrieval task agents) are used to return at least one of the following: relevant audio, video, and text / image content, based on the input target query information, to provide richer question-and-answer services based on multimodal output results. For example, if the target query information is: Please provide the main attractions in city C, the multimodal retrieval task agent can return results such as pictures, promotional videos, and text / image travel guides of the main attractions in city C.

[0050] In a multi-task agent system, any two agents can exchange information before outputting their processing results. This exchange allows them to reference each other's processing results for the query conditions and update their own, thereby improving the relevance between their results. For example, a question-answering agent and a recommendation agent can exchange information before outputting their results and update their own based on the exchanged information to obtain their respective target results. This approach improves the relevance between the target question-answer result output by the question-answering agent and the target recommendation result output by the recommendation agent. This relevance characterizes the degree of matching between the target question-answer result and the target recommendation result. The higher the relevance between the processing results of different tasks, the higher the overall quality of the task processing results.

[0051] The following will use either the first task agent or the second task agent as an example for illustrative purposes.

[0052] After the query conditions are input into the first task agent, the first task agent can generate a first processing result based on the query conditions. The first task agent can receive a second processing result from the second task agent in real time and update the first processing result accordingly to obtain the target processing result corresponding to the first task agent. The second task agent can be any one or more task agents other than the first task agent. Correspondingly, the second task agent can receive the first processing result from the first task agent in real time and update the second processing result accordingly to obtain the target processing result corresponding to the second task agent.

[0053] In the above implementation, the output process of any single task agent may include at least two rounds. The first round is used to generate a first processing result based on the target query information, and the second round is used to update the first processing result based on the second processing result received from other task agents. This can increase the correlation between the output results of different task agents on the one hand, and improve the generation efficiency of the target processing result by gradually building the target processing result on the other hand.

[0054] In some optional embodiments, among multiple task agents, a single task agent can invoke a large language model to complete the first round of output based on the corresponding task type and query condition information. Continuing with the first task agent as an example, when generating the first processing result, the first task agent can generate a task prompt word based on the query condition information and the task type corresponding to the first task agent, and provide the task prompt word to a preset large language model to obtain the first processing result output by the large language model.

[0055] In this embodiment, task agents for different task types have different task descriptions. For example, the task description for a question-and-answer task agent could be: "Please answer the input target query information," while the task description for a recommendation task agent could be: "Please generate recommended topics based on the provided target query information." Optionally, any task agent can combine the query condition information and the task description corresponding to the task type according to a set format or template to obtain task prompt words; or, the semantics of the query condition information can be optimized based on the task description to generate task prompt words. This embodiment does not impose any limitations.

[0056] For example, in some embodiments, the multiple task agents may include a question-answering task agent and a recommendation task agent. In this implementation, after obtaining the query condition information, the query condition information can be input into the question-answering task agent and the recommendation task agent respectively.

[0057] In the first round of processing, the question-answering task agent generates question-answering task prompts based on the query conditions and the task description corresponding to the question-answering task type. The question-answering agent inputs these prompts into the large language model to execute the question-answering task and obtains the question-answering results returned by the large language model based on the prompts. Similarly, the recommendation task agent generates recommendation task prompts based on the query conditions and the task description corresponding to the recommendation task type. The recommendation agent inputs these prompts into the large language model to execute the topic retrieval task and obtains the topic retrieval results returned by the large language model based on the prompts. The question-answering task agent provides the obtained question-answering results to the recommendation task agent, and in turn, the recommendation task agent provides the topic retrieval results to the question-answering task agent.

[0058] During the second round of processing, the question-answering task agent can update the question-answering results obtained in the first round based on the retrieved search results. Correspondingly, the recommendation task agent can update the topic search results obtained in the first round based on the retrieved question-answering results.

[0059] For example, in some embodiments, the multiple task agents may include: a question-answering task agent, a data visualization task agent, and a multimodal retrieval task agent. In this implementation, after obtaining the query condition information, the query condition information can be input into the question-answering task agent, the data visualization task agent, and the multimodal retrieval task agent respectively.

[0060] In the first round of processing, the question-answering task agent can invoke the large language model to perform the question-answering task based on the query conditions and obtain the question-answering results returned by the large language model based on the query conditions. The question-answering task agent can then provide the obtained question-answering results to the data visualization task agent and the multimodal retrieval task agent.

[0061] The data visualization task agent can invoke a large language model to perform data visualization tasks based on query conditions and obtain the data visualization results returned by the large language model based on the query conditions. The data visualization task agent can then provide the obtained data visualization results to the question-answering task agent and the multimodal retrieval task agent.

[0062] The multimodal retrieval task agent can invoke a large language model to perform multimodal retrieval tasks based on query conditions and obtain the multimodal retrieval results returned by the large language model based on the query conditions. The multimodal retrieval task agent can then provide these multimodal retrieval results to question-answering task agents and data visualization task agents.

[0063] During the second round of processing, the question-answering task agent can update the question-answering results obtained in the first round based on the acquired data visualization results and multimodal retrieval results. Similarly, the data visualization task agent can update the data visualization results obtained in the first round based on the acquired question-answering results and multimodal retrieval results.

[0064] After obtaining the target processing results acquired by each of the multiple task agents based on the above steps, the dialogue processing device can execute step 103, outputting the response result corresponding to the query request based on the target processing results acquired by each of the multiple task agents. In some deployment scenarios, when the dialogue processing device runs on a server, it can send the response result to the device or application corresponding to the query request, so that the device or application can output the response result. In other deployment scenarios, when the dialogue processing device runs on a terminal device, it can output the response result through the human-computer interaction interface provided by the terminal device; this embodiment does not impose any restrictions. Optionally, the dialogue processing device can output the target processing results of each of the multiple task agents separately, or it can concatenate the target processing results of each of the multiple task agents before outputting them; this embodiment does not impose any restrictions.

[0065] In this implementation, when the dialogue processing device includes multiple task agents of different task types, these agents can interactively perceive each other's task processing results. A single task agent can update its own processing result based on the perceived results of other tasks, effectively improving the correlation between the processing results of different tasks. Secondly, in this implementation, the multiple task agents are interdependent, thus reducing the time difference in outputting the target processing result between different task agents. Furthermore, a single task agent adopts a step-by-step approach to constructing the target processing result. After generating the processing result, it uses an update method to optimize the generated result using other processing results, rather than waiting to receive other processing results before generating the target processing result. This effectively improves the generation speed of the target processing result, enhances the smoothness of response to query requests, and alleviates response latency. Response smoothness refers to the ability to output the corresponding response result to the query request smoothly and coherently, reducing the questioner's perception of the time difference between different processing results, making the dialogue more natural.

[0066] Furthermore, by reducing the time difference between the output target processing results of different task agents, the content displayed by the dialogue client can be more complete, reducing the risk of temporary missing or incomplete content due to the delayed output of a certain target processing result, and improving the readability of the overall output results.

[0067] In some optional embodiments, the operation of the first task agent updating the first processing result based on the second processing result can be triggered when the correlation between the first processing result and the second processing result is low, in order to save computing power and reduce question-and-answer time. An exemplary description follows.

[0068] Optionally, a first task agent can be used to obtain a second processing result sent by a second task agent. When the correlation between the second processing result and the first processing result is less than a set threshold, the first processing result is updated based on the second processing result to obtain the target processing result corresponding to the first task agent. Here, correlation refers to the degree to which the first processing result and the second processing result are related in content.

[0069] In some optional embodiments, the first task agent may employ a content similarity algorithm to calculate the relevance between the first processing result and the second processing result. If the relevance is less than a set threshold, the agent may invoke a large language model to update the first processing result based on the second processing result. For example, the first task agent may obtain the semantic vectors corresponding to the first processing result and the second processing result, and calculate the similarity between the two semantic vectors as the content relevance between the first and second processing results. As another example, the first task agent may obtain the number of identical keywords in the first and second processing results, and determine the content relevance between the first and second processing results based on the number of identical keywords.

[0070] In some alternative embodiments, the first task agent can construct prompt words to guide the large language model to calculate the correlation between the first processing result and the second processing result, and update the first processing result based on the second processing result when the correlation is less than a set threshold. Further illustrative examples will follow.

[0071] Optionally, the first task agent can construct an update prompt word based on the first processing result, the second processing result, and the task description corresponding to the task class of the first task agent. This update prompt word is used to prompt the large language model to calculate the correlation between the first and second processing results, and when the correlation is less than a set threshold, the first processing result is updated based on the second processing result. The first task agent can input the constructed update prompt word into the large language model to obtain the updated first processing result, which serves as the target processing result for the first task agent. The first task agent can directly return the target processing result. In this implementation, by constructing prompt words, the large language model can automatically determine whether to update the generated processing result. This fully utilizes the powerful knowledge understanding and performance advantages of the large language model to accurately determine whether an update operation is needed. Furthermore, it reduces the dependence on the computing power of the first task agent, thus reducing the development and operating costs of the first task agent.

[0072] Optionally, during the process of constructing the update prompt, the first task agent may use the first processing result as a fixed context and the second processing result as a dynamic context.

[0073] Fixed context refers to information associated with the target query and remaining unchanged throughout the conversation or interaction in information processing, dialogue processing, or recommendation systems. Fixed context allows the model to establish a stable reference point when processing the target query, thus facilitating the coherence and consistency of the dialogue. Dynamic context, on the other hand, refers to contextual information that changes during information processing, dialogue, or recommendation. Furthermore, with the development of dialogue between task agents and large language models, large language models can absorb more information from dynamic context and incorporate it into subsequent generation or responses, improving the accuracy of dialogue results.

[0074] In some search enhancement scenarios, optionally, the first task agent, during the construction of update prompts, can also use the target knowledge text retrieved from a preset knowledge base based on the target query information as known relevant information. Furthermore, the large language model can refer to the information provided by the target knowledge text during the update process of the second processing result, returning a more accurate processing result.

[0075] Based on this, the format of the update prompt can include at least the following fields: target query information, known relevant information, fixed context, and dynamic context. Among them, the known relevant information can be used to store target knowledge text.

[0076] For example, the update prompts used by the question-and-answer task agent (qa_agent) could be formatted as follows:

[0077] qa_agent = 'The relevant information is as follows:\n\n

[0078] The fixed context is as follows: {id_content}\n

[0079] The dynamic context is as follows: {dy_content}\n\n

[0080] Please answer the question "{query}" in a structured manner, without using any other information, based on the information above. If you cannot answer, please output: "Unable to answer based on the given information."

[0081] For example, the format of the update prompt words used by the recommended task agent (sug_agent) could be:

[0082] sug_agent = 'The following information is known:'

[0083] Fixed context: {id_content}

[0084] Dynamic context: {dy_content}\n\n

[0085] Task Description: Based on the provided fixed and dynamic context information, and the given question and answer structure "{query}", generate 3 to 10 related questions as recommendations. Please generate recommended topics solely based on the provided context information; do not introduce external information.

[0086] Based on updated prompts, large language models can learn more knowledge from dynamically updated contextual information, thereby improving the relevance between processing results of different task types. Furthermore, by using the second processing result as a dynamic context and the first processing result as a fixed context, the large language model can use the generated first processing result as a stable reference point and expand or adjust it based on the dynamically input second processing result. In other words, the large language model does not need to re-understand all the context information and reconstruct the processing result. This effectively improves the speed at which the large language model optimizes processing results, increases the efficiency of task agents outputting target processing results for different task types, and further reduces the response latency of query requests.

[0087] In this embodiment, the multiple task agents of the dialogue processing device are created based on a multi-agent framework. The multi-agent framework is a distributed computing model in which multiple autonomous software entities (called agents) collaborate to solve complex problems. In the multi-agent framework, each agent has its own task or goal, and the agents can interact, cooperate, negotiate, compete, or communicate with each other to optimize overall performance.

[0088] In some optional embodiments, after creating different task agents based on task type, a data interaction channel between multiple task agents can be established through service registration. Any task agent can register its service with other task agents among these multiple task agents. The following explanation will continue using the first and second task agents as examples.

[0089] Optionally, the first task agent may send its service registration information to the second task agent, enabling the second task agent to send a second processing result to the first task agent. Correspondingly, the first task agent may also receive service registration information sent by the second task agent, enabling it to send a first processing result to the second task agent.

[0090] The service registration information of any task agent may include: the task agent's unique identifier, IP (Internet Protocol) address, port number and other necessary communication information, and may also include the task agent's version number and security credentials (such as keys), etc. This embodiment does not impose any restrictions.

[0091] Optionally, the first and second task agents can transmit their processing results to each other in real time based on a message-based data exchange mechanism. The message encoding method, the fields included in the message, and the specific meanings of the fields can be agreed upon in the service registration information and will not be elaborated further. In addition to the message-based data exchange mechanism, the first and second task agents can also transmit their processing results based on sockets or message queues stored in the kernel. When the first and second task agents run on the same device, they can transmit their processing results based on shared memory.

[0092] In this implementation, multiple task agents can collaborate through service registration, which facilitates mutual reference between multiple different types of tasks, thereby increasing the relevance of task processing results.

[0093] It's worth noting that in a multi-agent architecture, new task agents can be flexibly added according to task requirements to meet question-and-answer needs in different scenarios. Optionally, in response to the addition of a task agent, a new third task agent can be obtained. Using the third task agent, its service registration information is sent to the multiple created task agents, enabling these multiple task agents to send processing results to the third task agent. Correspondingly, the third task agent can receive the service registration information sent by these multiple task agents, allowing it to send processing results to them.

[0094] In this implementation, when a new task agent is added, it can collaborate with existing task agents to improve the quality of question-and-answer results. Furthermore, the task agents can be dynamically expanded flexibly according to the needs of the application scenario, enhancing the scalability of the dialogue processing device and its efficiency in adapting to new tasks.

[0095] Figure 2 Taking the first, second, and third task agents as examples, the process of message interaction between multiple task agents is illustrated. Figure 2 As shown, the first task agent can send the generated processing results to the second and third task agents via a first message, and can receive the second message sent by the second task agent and the third message sent by the third task agent. The first task agent can use the processing results in the second and third messages as a dynamic context, use the generated processing results as a fixed context, and call the large language model to return new processing results based on the fixed and dynamic contexts.

[0096] Accordingly, the second task agent can send the generated processing results to the first and third task agents via a second message, and can receive the first message sent by the first task agent and the third message sent by the third task agent. The second task agent can use the processing results in the first and third messages as a dynamic context, and the generated processing results as a fixed context, and call the large language model to return new processing results based on the fixed and dynamic contexts. The processing procedure of the third task agent is similar to that of the first and second task agents, and will not be described further.

[0097] After multiple task agents obtain the new processing results, they can output the task processing results synchronously to reduce response latency.

[0098] In addition to the aforementioned embodiments, this application also provides a question-answering method, which can be implemented by a dialogue processing device including a question-answering task agent and a recommendation task agent, as will be described exemplarily below.

[0099] The dialogue processing device can respond to query requests, obtain query condition information, and provide the query condition information to the question-answering task agent and the recommendation task agent to obtain the target question-answering result fed back by the question-answering task agent and the target recommendation result returned by the recommendation task agent. Specifically, the question-answering task agent is used to: based on the query condition information, call a preset large language model to obtain an initial question-answering result, and provide the initial question-answering result to the recommendation task agent; and receive the initial recommendation result provided by the recommendation task agent, and update the initial question-answering result based on the initial recommendation result to obtain the target question-answering result. The recommendation task agent is used to: based on the query condition information, call a preset large language model to obtain an initial recommendation result, and provide the initial recommendation result to the task agent; and receive the initial question-answering result provided by the question-answering task agent, and update the initial recommendation result based on the initial question-answering result to obtain the target recommendation result; the dialogue processing device can output the response result corresponding to the query request based on the target question-answering result and the target recommendation result. The following will, in conjunction with the accompanying drawings, take a retrieval enhancement scenario as an example to further illustrate the dialogue processing device provided in this application embodiment.

[0100] Figure 3 A schematic diagram of the framework structure of a dialogue processing apparatus provided in an exemplary embodiment of this application is shown below. Figure 3 As shown, the dialogue processing device may include an article ranking and retrieval module, which is mainly used to build a knowledge base required for retrieval enhancement. The article ranking and retrieval module can perform article preprocessing operations, article indexing operations, and article retrieval and ranking operations. For details, please refer to... Figure 3 The illustrations and descriptions of the knowledge base construction embodiments in the foregoing embodiments will not be repeated here.

[0101] The dialogue processing device includes a multi-task agent framework based on a large language model. Specifically, this framework may include a retrieval-enhanced question-answering agent and a retrieval-enhanced recommendation agent. The retrieval-enhanced question-answering agent retrieves enhanced content from the knowledge base generated by the article ranking retrieval module based on the input target query, and then answers the target query based on the enhanced content and other dynamic contextual information. The retrieval-enhanced recommendation agent retrieves enhanced content from the knowledge base generated by the article ranking retrieval module based on the input target query, and then generates recommended content relevant to the target query based on the enhanced content and other dynamic contextual information.

[0102] like Figure 3 As shown, the question-answering task agent and the recommendation task agent can transmit their respective processing results through bidirectional message exchange, and use the received processing results from each other as dynamic context. For example... Figure 3 As shown, the question-answering task agent can send the generated answer to the recommendation task agent via a first message, and the recommendation task agent can send the generated recommendation result to the question-answering task agent via a second message. The question-answering task agent can use the generated question-answering result as a fixed context and the received recommendation result as a dynamic context to call a preset large language model to generate a new answer. Similarly, the recommendation task agent can use the generated recommendation result as a fixed context and the received question-answering result as a dynamic context to call a preset large language model to generate a new recommendation result. For example... Figure 3 As shown, new recommendation results and question-and-answer results can be output simultaneously.

[0103] Based on this implementation, in the dialogue processing device, the correlation between the processing results of different tasks can be effectively improved through data interaction between the question-answering task agent and the recommendation task agent. Furthermore, the question-answering agent and the recommendation task agent are interdependent, thus reducing the time difference in outputting the target processing result between different task agents. In addition, each task agent adopts a step-by-step approach to construct the target processing result. After generating the processing result, it uses an update method to optimize the generated result using other processing results, rather than waiting to receive other processing results before generating the target processing result. This effectively improves the generation speed of the target processing result, enhances the smoothness of response to query requests, and alleviates response latency.

[0104] In addition to the dialogue processing method provided in the foregoing embodiments, this application also provides a dialogue interaction method, which can be executed by a terminal device, and will be described in detail below.

[0105] The terminal device can respond to an inquiry operation initiated through the first dialogue interaction interface, obtain target inquiry information, and send an inquiry request to the server based on the target inquiry information, so that the server returns multiple target processing results according to the inquiry request and the dialogue processing methods described in the foregoing embodiments. The terminal device can output the response result corresponding to the inquiry operation through the second dialogue interaction interface based on the multiple target processing results.

[0106] In this implementation, the terminal device may include a mobile phone, tablet computer, smart wearable device, etc., on the user side. The terminal device can provide one or more human-computer interaction interfaces for the user to interact with the dialogue device. The first dialogue interaction interface and the second dialogue interaction interface are the human-computer interaction interfaces for dialogue with the user. The first dialogue interaction interface and the second dialogue interaction interface may be the same or different. In some embodiments, the first dialogue interaction interface may include at least one of: a graphical user interface, a text user interface, a voice user interface, a touch user interface, a gesture user interface, a keyboard and mouse interface, and a sensor and biometric interface. In some embodiments, when the first dialogue interaction interface is a text user interface, the terminal device can respond to an inquiry operation, obtain the question information or search information input by the user through the text user interface, and use it as the target inquiry information. In other embodiments, when the first dialogue interaction interface is a voice user interface, the terminal device can respond to an inquiry operation, obtain the voice data input by the user through the voice user interface, perform semantic recognition on the voice data, and use the result of the semantic recognition as the target inquiry information. When the terminal device outputs the response result corresponding to the query operation based on the processing results of multiple targets, it can output the multiple processing results sequentially, or it can concatenate the multiple processing results and output them together. This embodiment does not impose any restrictions.

[0107] In this implementation, the server runs the dialogue processing apparatus described in the foregoing embodiments. Within this apparatus, data interaction between the question-answering task agent and the recommendation task agent effectively enhances the relevance between the processing results of different tasks, ensuring that the responses output by the terminal device are content-related. Furthermore, in the dialogue processing apparatus, individual task agents optimize already generated processing results through updates, rather than waiting to receive other processing results before generating the target processing result. This effectively improves the generation speed of the target processing result, allowing the terminal device to quickly respond to a query operation after detection, thus enhancing dialogue fluency.

[0108] In addition to the dialogue processing method and dialogue interaction method provided in the foregoing embodiments, this application also provides a dialogue system, which includes a terminal device and a server. The terminal device runs a dialogue client, and the server runs the dialogue processing device described in the foregoing embodiments. The terminal device can be a user-side mobile phone, tablet computer, or smart wearable device, and the server can be a conventional server or a cloud server.

[0109] The terminal device is used to: respond to an inquiry operation initiated through a first dialogue interaction interface and obtain target inquiry information; send an inquiry request to the server based on the target inquiry information; receive multiple target processing results returned by the server based on the inquiry request, and output the response result corresponding to the inquiry operation through a second dialogue interaction interface based on the multiple target processing results.

[0110] The server is configured to: respond to the query request and obtain query condition information; utilize multiple task agents of different task types to obtain target processing results based on the query condition information; among the multiple task agents, any first task agent is configured to: generate a first processing result by calling a preset large language model based on the query condition information; and receive in real time a second processing result generated by a second task agent among the multiple task agents based on the query condition information, and update the first processing result based on the second processing result to obtain the target processing result corresponding to the first task agent; and return the target processing results obtained by each of the multiple task agents to the terminal device. For details, please refer to the description in the foregoing embodiments, which will not be repeated here.

[0111] It should be noted that the execution subject of each step of the method provided in the above embodiments can be the same device, or the method can be executed by different devices. For example, the execution subject of steps 101 to 104 can be device A; or the execution subject of steps 101 and 102 can be device A, and the execution subject of step 103 can be device B; and so on.

[0112] Furthermore, some processes described in the above embodiments and accompanying drawings include multiple operations appearing in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not represent any execution order. Additionally, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0113] It should be noted that 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, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0114] Figure 4 This illustration shows a structural diagram of an electronic device provided in an exemplary embodiment of this application, which is applicable to the dialogue processing method provided in the foregoing embodiments. For example... Figure 4 As shown, the electronic device includes: a memory 401, a processor 402, and a communication component 403.

[0115] Memory 401 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device.

[0116] In some embodiments, Figure 4 The illustrated electronic device can be used to execute a dialogue processing method, wherein a processor 402, coupled to a memory 401, is used to execute a computer program in the memory 401 for: responding to an inquiry request and obtaining inquiry condition information; utilizing multiple task agents corresponding to different task types to obtain target processing results according to the inquiry condition information; among the multiple task agents, any first task agent is used to: generate a first processing result by calling a preset large language model according to the inquiry condition information; and receive in real time a second processing result generated by a second task agent among the multiple task agents according to the inquiry condition information, and update the first processing result according to the second processing result to obtain the target processing result corresponding to the first task agent; and output a response result corresponding to the inquiry request according to the target processing results obtained by the multiple task agents respectively.

[0117] Optionally, when acquiring query condition information, the processor 402 is specifically configured to: acquire target query information corresponding to the query request; search a preset knowledge base based on the target query information to obtain target knowledge text; and generate the query condition information based on the target knowledge text and the target query information.

[0118] Optionally, when the processor 402 calls a preset large language model to generate a first processing result based on the query condition information, it is specifically used to: use the first task agent to generate a task prompt word based on the query condition information and the task type corresponding to the first task agent; provide the task prompt word to the preset large language model to obtain the first processing result output by the large language model.

[0119] Optionally, when the processor 402 updates the first processing result according to the second processing result to obtain the target processing result corresponding to the first task agent, it is specifically used to: update the first processing result according to the second processing result to obtain the target processing result corresponding to the first task agent when the correlation between the second processing result and the first processing result is less than a set threshold.

[0120] Optionally, when the processor 402 updates the first processing result based on the second processing result to obtain the target processing result corresponding to the first task agent when the correlation between the second processing result and the first processing result is less than a set threshold, the processor 402 is specifically configured to: construct an update prompt word based on the first processing result, the second processing result, and the task description corresponding to the task class of the first task agent; input the update prompt word into a preset large language model to obtain the updated first processing result, which is used as the target processing result corresponding to the first task agent; wherein, the update prompt word is used to prompt the large language model to calculate the correlation between the first processing result and the second processing result, and update the first processing result based on the second processing result when the correlation is less than a set threshold.

[0121] Optionally, the processor 402 is further configured to: use the first processing result as a fixed context and the second processing result as a dynamic context during the construction of the updated prompt word.

[0122] Optionally, the processor 402 is further configured to: use the first task agent to send the service registration information of the first task agent to the second task agent, so that the second task agent sends the second processing result to the first task agent; and receive the service registration information sent by the second task agent, so that the first task agent sends the first processing result to the second task agent.

[0123] Optionally, the processor 402 is further configured to: respond to the operation of adding a new task agent, obtain a newly added third task agent; use the third task agent to send the service registration information of the third task agent to the plurality of task agents, so that the plurality of task agents send processing results to the third task agent; and receive the service registration information sent by the plurality of task agents, so that the third task agent sends processing results to the plurality of task agents.

[0124] In other embodiments, Figure 4 The illustrated electronic device can be used to perform the following dialogue processing method, wherein a processor 402 is coupled to a memory 401 and is used to execute a computer program in the memory 401 for: responding to an inquiry request and obtaining inquiry condition information; using a question-answering task agent and a recommendation task agent, respectively obtaining a target question-answering result and a target recommendation result based on the inquiry condition information; wherein the question-answering task agent is used to: based on the inquiry condition information, call a preset large language model to obtain an initial question-answering result and provide the initial question-answering result to the recommendation task agent; and receive the initial recommendation result provided by the recommendation task agent and update the initial question-answering result based on the initial recommendation result to obtain a target question-answering result; the recommendation task agent is used to: based on the inquiry condition information, call a preset large language model to obtain an initial recommendation result and provide the initial recommendation result to the task agent; and receive the initial question-answering result provided by the question-answering task agent and update the initial recommendation result based on the initial question-answering result to obtain a target recommendation result; and output the response result corresponding to the inquiry request based on the target question-answering result and the target recommendation result. Further, as... Figure 4 As shown, the electronic device also includes other components such as a power supply component 404, a display component 405, and an audio component 406. Figure 4 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 4 The components shown. Figure 4 In this embodiment, the components within the dashed boxes are optional, not mandatory, and their specific requirements depend on the product form of the electronic device. The electronic device in this embodiment can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT device, or a server-side device such as a conventional server, cloud server, or server array. If the electronic device in this embodiment is a terminal device such as a desktop computer, laptop computer, or smartphone, it may include... Figure 4 The components within the dashed box; if the electronic device in this embodiment is implemented as a conventional server, cloud server, or server array, etc., it may be omitted. Figure 4 The component within the dashed box.

[0125] like Figure 4If the illustrated electronic device is implemented as a terminal device, then the electronic device can be used to execute the following dialogue interaction method, wherein the processor 402 is coupled to the memory 401 and is used to execute the computer program in the memory 401 for: responding to an inquiry operation initiated through a first dialogue interaction interface and obtaining target inquiry information; sending an inquiry request to a server according to the target inquiry information, so that the server returns multiple target processing results according to the inquiry request, the multiple target processing results being obtained by the dialogue processing method provided by the server in this application embodiment; and outputting the response result corresponding to the inquiry operation through a second dialogue interaction interface according to the multiple target processing results.

[0126] The memory 401 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0127] The communication component 403 is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as Wi-Fi, 2G (e.g., Global System for Mobile Communications (GSM)), 3G (e.g., Wideband Code Division Multiple Access (WCDMA), 4G (e.g., Long Term Evolution (LTE)), 4G+ (e.g., LTE-Advanced (LTE-A)), or 5G (5th Generation Mobile Communication Technology), or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component may be implemented based on Near Field Communication (NFC), Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wide Band (UWB), Bluetooth (BT), and other technologies.

[0128] The power supply component 404 is used to provide power to various components of the device in which the power supply component is located. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply component is located.

[0129] The display component includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from a user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation.

[0130] An audio component may be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals may be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0131] In this embodiment, when multiple task agents of different task types exist, they can interactively perceive each other's task processing results. A single task agent can update its own processing result based on the perceived results of other tasks, effectively improving the correlation between the processing results of different tasks. Secondly, in this implementation, the multiple task agents are interdependent, thus reducing the time difference between different task agents outputting the target processing result. Furthermore, a single task agent adopts a step-by-step approach to constructing the target processing result. After generating the processing result, it uses an update method to optimize the generated result using other processing results, rather than waiting to receive other processing results before generating the target processing result. This effectively improves the generation speed of the target processing result, enhances the smoothness of response to query requests, and alleviates response latency.

[0132] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed, can implement the steps in the method provided in the above-described method embodiments.

[0133] This application also provides a computer program product, including: a computer program / instructions, which, when executed by a processor, can implement the steps in the method provided in this application.

[0134] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM (Compact Disc Read-Only Memory), optical storage, etc.) containing computer-usable program code.

[0135] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0136] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0137] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0138] In a typical configuration, a computing device includes one or more processors (Central Processing Unit, CPU), input / output interfaces, network interfaces, and memory.

[0139] 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.

[0140] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using 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 Video Disc (DVD) or other optical storage, magnetic tape, 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 herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0141] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, product, or apparatus that includes said element.

[0142] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A dialog processing method characterized by, Comprise: In response to an inquiry request, obtain inquiry condition information; Using a plurality of task agents corresponding to different task types, respectively according to the inquiry condition information to obtain target processing result; Among the plurality of task agents, any first task agent is used to: according to the inquiry condition information, call a preset large language model to generate a first processing result; And, real-time receiving the second processing result generated by the second task agent in the plurality of task agents according to the inquiry condition information, and updating the first processing result according to the second processing result to obtain the target processing result corresponding to the first task agent; According to the target processing result obtained by the plurality of task agents respectively, output the reply result corresponding to the inquiry request.

2. The method of claim 1, wherein, Obtaining inquiry condition information comprises: Obtain target inquiry information corresponding to the inquiry request; According to the target inquiry information, the preset knowledge base is searched to obtain the target knowledge text; According to the target knowledge text and the target inquiry information, the inquiry condition information is generated.

3. The method of claim 1, wherein, According to the second processing result, the first processing result is updated to obtain the target processing result corresponding to the first task agent, comprising: When the relevance between the second processing result and the first processing result is less than a set threshold, the first processing result is updated according to the second processing result to obtain the target processing result corresponding to the first task agent.

4. The method of claim 3, wherein, When the relevance between the second processing result and the first processing result is less than a set threshold, the first processing result is updated according to the second processing result to obtain the target processing result corresponding to the first task agent, comprising: According to the first processing result, the second processing result and the task description corresponding to the task class of the first task agent, the update prompt word is constructed; The update prompt word is input into the preset large language model to obtain the updated first processing result as the target processing result corresponding to the first task agent; wherein, the update prompt word is used to prompt the large language model to calculate the relevance between the first processing result and the second processing result, and when the relevance is less than a set threshold, the first processing result is updated according to the second processing result.

5. The method of claim 4, wherein, Further comprising: In the process of constructing the update prompt word, the first processing result is used as a fixed context, and the second processing result is used as a dynamic context.

6. The method according to any one of claims 1 to 5, characterized in that, Further comprising: Using the first task agent, the service registration information of the first task agent is sent to the second task agent, so that the second task agent sends the second processing result to the first task agent; and receiving the service registration information sent by the second task agent, so that the first task agent sends the first processing result to the second task agent.

7. The method of claim 6, wherein, Further comprising: In response to the operation of adding a task agent, a third task agent is added The third task agent is used for sending service registration information of the third task agent to the plurality of task agents, so that the plurality of task agents send processing results to the third task agent; and receiving service registration information sent by the plurality of task agents, so that the third task agent sends processing results to the plurality of task agents.

8. A dialogue processing method characterized by, Comprise: In response to an inquiry request, obtain inquiry condition information; Using a question and answer task agent and a recommendation task agent, target question and answer results and target recommendation results are respectively obtained according to the inquiry condition information; wherein the question and answer task agent is used for: according to the inquiry condition information, calling a preset large language model to obtain an initial question and answer result, and providing the initial question and answer result to the recommendation task agent; and receiving the initial recommendation result provided by the recommendation task agent, and updating the initial question and answer result according to the initial recommendation result to obtain the target question and answer result; the recommendation task agent is used for: according to the inquiry condition information, calling a preset large language model to obtain an initial recommendation result, and providing the initial recommendation result to the task agent; and receiving the initial question and answer result provided by the question and answer task agent, and updating the initial recommendation result according to the initial question and answer result to obtain the target recommendation result; According to the target question and answer result and the target recommendation result, a reply result corresponding to the inquiry request is output.

9. A conversational interaction method, characterized by, Comprise: In response to an inquiry operation initiated through a first dialogue interaction interface, target inquiry information is obtained; According to the target inquiry information, an inquiry request is sent to a server, so that the server returns a plurality of target processing results according to the inquiry request, the plurality of target processing results being obtained by the server according to the method of any one of claims 1-7; According to the plurality of target processing results, a reply result corresponding to the inquiry operation is output through a second dialogue interaction interface.

10. A dialog system, characterized by Comprise: Terminal device and server; The terminal device is used for: in response to an inquiry operation initiated through a first dialogue interaction interface, target inquiry information is obtained; According to the target inquiry information, an inquiry request is sent to a server; A plurality of target processing results returned by the server according to the inquiry request are received, and a reply result corresponding to the inquiry operation is output according to the plurality of target processing results through a second dialogue interaction interface; The server is used for: in response to the inquiry request, inquiry condition information is obtained; Using a plurality of task agents of different task types, target processing results are respectively obtained according to the inquiry condition information; any first task agent in the plurality of task agents is used for: according to the inquiry condition information, a preset large language model is called to generate a first processing result; And a second processing result generated by a second task agent in the plurality of task agents according to the inquiry condition information is received in real time, and the first processing result is updated according to the second processing result to obtain a target processing result corresponding to the first task agent; The target processing results respectively obtained by the plurality of task agents are returned to the terminal device.

11. An electronic device, comprising: Comprise: Memory and processor; The memory is configured to store one or more computer instructions; The processor is configured to execute the one or more computer instructions for performing the steps of the method of any one of claims 1-9.

12. A computer readable storage medium storing a computer program, characterized in that, A computer program which, when executed by a processor, is capable of implementing the steps of the method of any one of claims 1-9.

13. A computer program product, characterised in that, comprising: A computer program / instructions which, when executed by a processor, is capable of implementing the steps of the method of any one of claims 1-9.

Images