Reply information generation method, device, client and storage medium

Abstract

An embodiment of the present invention provides a reply information generation method, device, client, and storage medium. The method includes: in response to demand information input by a user in a dialogue interface, parsing the demand information into at least one search enhancement subtask, and displaying the at least one search enhancement subtask in the dialogue interface; searching the at least one search enhancement subtask, generating and displaying at least one search result corresponding to the at least one search enhancement subtask; generating reply information for the demand information based on the at least one search result, and displaying the reply information in the dialogue interface. In the technical solution provided by the embodiment of the present invention, in the reply information generation process, the thinking process of the display model is explicitly displayed, which improves the logical integrity of the reply information and alleviates the illusion problem of the LLM, thereby improving the credibility of the large language model and enhancing the user's dialogue experience.

Description

Technical field

[0001] Embodiments of the present invention relate to the field of artificial intelligence technology, and in particular to a method, device, client, and storage medium for generating reply information. [Background Technology]

[0002] With the continuous development of the field of artificial intelligence (AI) technology, AI-based human-computer dialogue systems have gradually been applied to various fields. In related technologies, AI-based human-computer dialogue systems usually use large language models (LLMs) as support. In the actual application of LLMs, hallucination problems may exist. The specific manifestation of the hallucination problem is that when the content generated by the LLM requires knowledge beyond the training data, the LLM may generate output that is inconsistent with the facts. That is, the LLM may produce hallucinations, resulting in a poor user experience and low authenticity and credibility of the LLM output content.

[0003] The automatic generation capabilities of large language models rely on training data. Due to the limited training data available, large language models may not be able to provide professional responses to highly specialized questions, especially those in vertical fields where knowledge is not readily available. Furthermore, the training data for large language models has expiration dates and may not reflect the latest events or research. For time-sensitive questions, the generated responses may be less timely, resulting in a poor user experience. [Summary of the invention]

[0004] In view of this, embodiments of the present invention provide a reply information generation method, device, client, and storage medium to solve at least some of the above problems. The disclosed technical solutions are as follows:

[0005] In a first aspect, an embodiment of the present invention provides a method for generating a reply message, which is applied to a human-computer dialogue application based on generative artificial intelligence (AIGC). The method includes:

[0006] In response to demand information input by a user in a dialogue interface, parsing the demand information into at least one search enhancement subtask, and displaying the at least one search enhancement subtask in the dialogue interface;

[0007] Searching the at least one search enhancement subtask, generating and displaying at least one search result corresponding to the at least one search enhancement subtask;

[0008] Generate reply information for the demand information according to the at least one search result, and display the reply information in the dialogue interface.

[0009] In a possible implementation, parsing the demand information into at least one search enhancement subtask and displaying the at least one search enhancement subtask in the dialogue interface includes:

[0010] The demand information is parsed into at least one search enhancement subtask and at least one automatically generated subtask, and the at least one search enhancement subtask and the at least one automatically generated subtask are displayed in the dialogue interface.

[0011] In a possible implementation, parsing the demand information into at least one search enhancement subtask and displaying the at least one search enhancement subtask in the dialogue interface includes:

[0012] The demand information is parsed into a plurality of search enhancement subtasks, and the plurality of search enhancement subtasks are displayed in the dialogue interface.

[0013] In a possible implementation, the search result includes a search result title control and site information of the search result, and the method further includes:

[0014] In response to the user clicking the search result title control, the interface is redirected to the search result display interface associated with the search result title control according to the site information of the search result.

[0015] In a possible implementation, the method further includes:

[0016] In response to a user sliding operation of the at least one search result, adjacent search results are displayed.

[0017] In a possible implementation, before displaying the reply information in the conversation interface, the method further includes:

[0018] The process information for generating the reply information is folded, where the process information includes at least one of the at least one search enhancement subtask, the at least one search result, and process marking information, where the process marking information is used to mark the current processing process.

[0019] In a possible implementation, the reply information includes reply content, at least one reference mark control, and the at least one search result, and the reply content is automatically generated by a large language model based on the at least one search result.

[0020] In a possible implementation, after generating reply information for the demand information according to the at least one search result and displaying the reply information on the dialogue interface, the method further includes:

[0021] In response to the user clicking the reference badge control, a reference information display floating layer is displayed in the dialogue interface, and the reference information display floating layer is used to display the reference verification information corresponding to the reference badge control, so that the user can perform reference verification on the reply content according to the reference verification information.

[0022] In a possible implementation, the reply message further includes a display status control, the display status control has a folded state and an expanded state, and the method further includes:

[0023] In response to a user clicking on a display status control in a folded state, unfolding information about a process of generating the reply information; or

[0024] In response to a user clicking on a display status control in an expanded state, folding process information of generating the reply information;

[0025] The process information includes at least one of the at least one search enhancement subtask, the at least one search result, and process marking information, where the process marking information is used to mark the current processing process.

[0026] In a second aspect, an embodiment of the present invention provides a reply information generating device, which is applied to an AIGC human-computer dialogue application, and the device includes:

[0027] a parsing module, configured to, in response to demand information input by a user in a dialogue interface, parse the demand information into at least one search enhancement subtask, and display the at least one search enhancement subtask in the dialogue interface;

[0028] A search module, configured to search the at least one search enhancement subtask, and generate and display at least one search result corresponding to the at least one search enhancement subtask;

[0029] A reply module is used to generate reply information of the demand information according to the at least one search result, and display the reply information in the dialogue interface.

[0030] In a third aspect, an embodiment of the present invention provides a client comprising: a display screen; one or more processors; a memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions that, when executed by the client, cause the client to execute the reply information generation method in the first aspect or any possible implementation of the first aspect.

[0031] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the device where the computer-readable storage medium is located is controlled to execute the reply information generation method in the first aspect or any possible implementation of the first aspect.

[0032] The technical solution provided by the embodiments of this invention parses the user's input requirement and searches for the parsed subtasks to generate a response to the requirement. The model's thought process is explicitly displayed during the response generation process, improving the logical integrity of the response and alleviating the LLM hallucination problem. This enhances the credibility of the large language model and improves the user's conversational experience.

[0033] In an embodiment of the present invention, when the demand information input by the user belongs to a highly professional vertical field, the client can search the demand information through RAG, obtain relevant search results in the vertical field, and generate professional reply information based on the search results, thereby improving the user's conversation experience.

[0034] In an embodiment of the present invention, when the demand information input by the user has high timeliness requirements, the client can obtain the latest search results through RAG and generate reply information based on the latest search results, avoiding the problem of poor timeliness of reply information due to the timeliness limitation of training data of the large language model, thereby improving the user's conversation experience.

Brief Description of the Drawings

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0036] Figure 1 A flowchart of a method for generating a reply message provided by an embodiment of the present invention;

[0037] Figure 2 A schematic diagram of a dialogue interface provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of another dialogue interface provided by an embodiment of the present invention;

[0039] Figure 4a A schematic diagram of another dialogue interface provided by an embodiment of the present invention;

[0040] Figure 4b A schematic diagram of another dialogue interface provided by an embodiment of the present invention;

[0041] Figure 5 A schematic diagram of another dialogue interface provided by an embodiment of the present invention;

[0042] Figure 6 A schematic structural diagram of a reply information generating device provided by an embodiment of the present invention;

[0043] Figure 7 A schematic diagram of a client provided by an embodiment of the present invention. [Specific implementation method]

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0045] It should be understood that the embodiments described are only a portion of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0046] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0047] It should be understood that the term "and / or" as used herein simply describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. Furthermore, the character " / " in this document generally indicates an "or" relationship between the associated objects.

[0048] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0049] Figure 1A flowchart of a method for generating a reply message provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes:

[0050] Step 101: In response to demand information input by a user in a dialogue interface, the demand information is parsed into at least one search enhancement subtask, and the at least one search enhancement subtask is displayed in the dialogue interface.

[0051] Each step of the embodiment of the present invention may be executed by a client, which includes but is not limited to a mobile phone, a tablet computer, a portable personal computer (PC), a desktop computer, etc.

[0052] The embodiment of the present invention is applied to human-computer dialogue applications based on generative artificial intelligence (Artificial Intelligence Generated Content, referred to as AIGC). AIGC is a general artificial intelligence (AGI) that focuses on generating or creating new content. It generates new text, program code, images, videos, sounds and other data based on existing large-scale multimodal data sets through machine learning. It can handle a variety of different tasks and be applied to various different scenarios, such as text summarization, text generation, machine translation, question and answer, etc., with a high level of generalization. The application of AIGC technology can improve the efficiency of content generation, save time and costs for creators and enterprises, and expand the possibilities of content creation. The basic support of AIGC is the large language model (LLM). LLM is a language model containing billions or more parameters. It generates human-like text by learning and understanding large amounts of text data and contextual data. For example, LLM includes Generative Pre-Trained Transformer (GPT), Baichuan2-7B, Baichuan2-13B, etc. The LLM undergoes comprehensive training, including pre-training, supervised fine-tuning (SFT), and reinforcement learning from human feedback (RLHF). During the pre-training phase, the model uses a large amount of text data to learn the language structure, grammar, and factual knowledge contained in the text. During the SFT phase, the pre-trained LLM is fine-tuned using labeled data to make it more suitable for a specific task. In the RLHF phase, reinforcement learning methods are used to directly optimize the language model using human feedback signals.

[0053] After receiving user input, the client processes it through the intent understanding module. This module determines whether to trigger Retrieval-Augmented Generation (RAG) to process the user input, effectively understanding the intent.

[0054] Optionally, the client parses the demand information into at least one search enhancement subtask and at least one automatic generation subtask, and displays the at least one search enhancement subtask and at least one automatic generation subtask in the dialogue interface. The search enhancement subtask is a subtask that cannot generate a response through the automatic generation capability of the large language model and requires triggering RAG, while the automatic generation subtask is a subtask that can generate a response through the automatic generation capability of the large language model and does not require triggering RAG.

[0055] Optionally, the client parses the demand information into multiple search enhancement subtasks and displays the multiple search enhancement subtasks in the dialogue interface. For example, if the user inputs the demand information "What is the current stock price of the top 7 insurance companies in China by revenue as of 2022?", the client parses the demand information into two search enhancement subtasks, namely "Top 7 insurance companies in China by revenue" and "Insurance company stock price," and displays the two search enhancement subtasks in the dialogue interface.

[0056] As an optional solution, when the demand information is a simple entity query or a simple first-degree relationship query of an entity, the client parses the demand information into automatically generated subtasks. Entities are broad entities, which can include not only tangible objective objects (such as people or things), but also abstract events (such as definitions, lists, etc.). For example, entity queries include "Who is Andy Lau?" or "What is Newton's first law?" A first-degree relationship refers to a relationship with a direct connection. For example, first-degree relationship queries of entities include "Who is Jay Chou's wife?" or "Is eating rice dumplings a custom during the Dragon Boat Festival?"

[0057] As another alternative, when the requirement is a complex entity query or a complex first-degree relationship query between entities, the client parses the requirement into a search enhancement subtask. For example, if the requirement is "Can a man and a woman, both registered as residents of Zhejiang Province, register their marriage in Beijing?", the essence of this requirement is "the conditions for registering a marriage in Beijing." Another example is "Can a woman born in the 1860s obtain a one-child certificate?", the essence of this requirement is "the start and end times of the one-child certificate." In these examples, the essence of the requirement is a simple entity query or a simple first-degree relationship query between entities. However, due to the complex expression of the requirement, RAG is also triggered.

[0058] As another alternative, when the requirement information can be parsed into multiple simple entity queries or simple entity-first-degree relationship queries, the client parses the requirement information into multiple search enhancement subtasks. For example, if the requirement information is "Who was the coach of the National Basketball Association (NBA) runner-up in the year Yao Ming was born?", it can be parsed into "What year was Yao Ming born?" and "NBA runner-up coach query." For another example, if the requirement information is "Which river flows through more provinces, the Yangtze River or the Yellow River?", it can be parsed into "How many provinces does the Yangtze River and the Yellow River flow through respectively?" and "Compare the number of provinces."

[0059] As another option, when the demand information is open-ended subjective question-and-answer information, the client parses the demand information into search enhancement subtasks. For example, open-ended subjective question-and-answer information includes subjective suggestions (such as investment advice and career advice), experience, planning, and development trends (such as industry development trends), subjective predictions (predictions of uncertain facts), causal speculation, recommendations (such as shopping recommendations and film and television entertainment recommendations), and comparisons (such as comparing scenarios and judging positions).

[0060] In this step, the client displays at least one search enhancement subtask using streaming output. Streaming output is a method of processing data that treats data as a continuous stream rather than a static dataset. With traditional output methods, the client outputs all data to the front-end all at once after all data is generated. This traditional output method is suitable for situations with small data volumes or fast processing speeds. However, for situations with large data volumes or slow processing speeds, this method may result in longer data generation times and a poor user experience. With streaming output, data is processed and output immediately upon generation, without waiting for all data to be fully prepared. This makes streaming output particularly suitable for processing large and real-time data, providing real-time feedback and results. For example, if the client needs to display two parsed search enhancement subtasks, with traditional output methods, it would wait until both search enhancement subtasks are parsed before displaying them. With streaming output, once the client has parsed a search enhancement subtask, it will be displayed directly, without waiting for both to be parsed. Users can see changes in the output data in real time, thus improving the user experience. The client breaks down the requirement information into at least one search-enhanced subtask. When displaying subtasks, a maximum of N subtasks are displayed. Subtasks exceeding the limit are not displayed. Where N is a positive integer, for example, N = 3. Each subtask is limited to one line. Content exceeding one line is indicated by "...".

[0061] Optionally, the process marking information includes parsing title information. Specifically, the client may further display parsing title information above at least one search enhancement subtask, the parsing title information including text information and icon information, and the parsing title information is first parsing title information or second parsing title information. When the parsing title information is the first parsing title information, it indicates that the client is executing step 101; when the parsing title information is the second parsing title information, it indicates that the client has successfully executed step 101. For example, the first parsing title information includes text information of "subtask parsing in progress" and icon information of the ongoing status; the second parsing title information includes text information of "subtask parsing completed" and icon information of the completed status.

[0062] Step 102: Search for at least one search enhancement subtask, and generate and display at least one search result corresponding to the at least one search enhancement subtask.

[0063] In this step, the search results include a search results title control and the site information of the search results. Optionally, when a user clicks on the search results title control, the client, in response to the user clicking on the search results title control, jumps to a search results display interface associated with the search results title control based on the site information of the search results. The search results display interface is a Hypertext Markup Language 5 (HTML5) interface. The site information of the search results includes the site URL, site icon, and site name. For example, the site may include official websites such as Zhihu, Baidu Knows, Sogou Q&A, and Sohu News to ensure the authenticity and accuracy of the search results.

[0064] Optionally, the process marking information also includes search title information. Specifically, the client can also display the search title information above at least one search result. The search title information includes text information and icon information, and the search title information is either the first search title information or the second search title information. When the search title information is the first search title information, it indicates that the client is currently executing step 102; when the search title information is the second search title information, it indicates that the client has successfully executed step 102. For example, the first search title information includes text information such as "Retrieving relevant information" and an icon indicating the ongoing status; the second search title information includes text information such as "Retrieval of relevant information completed" and an icon indicating the completed status.

[0065] Step 103: Generate reply information for the demand information based on at least one search result, and display the reply information in the dialogue interface.

[0066] In this step, the reply information includes the reply content, at least one reference corner control, and at least one search result. The reply content is automatically generated by the large language model based on the at least one search result, and the reply content may include at least one of text content, image content, audio and video content, and link content.

[0067] In an embodiment of the present invention, before displaying the reply message in the conversation interface, the method further includes: folding process information of generating the reply message. The process information includes at least one of at least one search enhancement subtask, at least one search result, and process marking information, where the process marking information is used to mark the current processing process.

[0068] In an embodiment of the present invention, at least one search result is displayed when displaying a reply message, so that the user can also check at least one search result when the process information is folded, which facilitates the user to further understand the reply message, thereby improving the user's conversation experience.

[0069] Figure 2 A schematic diagram of a dialogue interface provided by an embodiment of the present invention is shown as follows: Figure 2 As shown, the reply message includes the reply content, two reference badge controls, and three search results. The two reference badge controls are the first reference badge control and the second reference badge control, and the three search results are the first search result, the second search result, and the third search result. When the number of search results is large and the dialog interface cannot display all the search results, the user slides the search results on the dialog interface, and the client displays adjacent search results in response to the user's sliding operation.

[0070] After step 103, the process may further include: in response to a user clicking a citation badge control, displaying a citation information display overlay in the conversation interface. The citation information display overlay is used to display citation verification information corresponding to the citation badge control, allowing the user to perform citation verification on the reply content based on the citation verification information. The citation verification information includes a citation title control, citation verification content, the publication time of the citation verification content, and the site information of the citation verification content. Optionally, the citation verification information also includes a citation sequence number.

[0071] In this embodiment of the present invention, citation verification is specifically implemented by generating a response corresponding to user input based on the RAG, inputting the response corresponding to the user input into a trained citation verification classification model, and outputting citation verification content. RAG provides additional information to the LLM from an external knowledge source, enabling the LLM to generate more accurate and contextually relevant content, while reducing the generation of erroneous information, alleviating the LLM's hallucination problem, and improving the authenticity and reliability of the LLM's output.

[0072] Specifically, the citation classification training model includes a Transformer module, a token classification layer, a BIO annotation layer, a citation classification layer, and a conditional random field (CRF) layer. Among them, the Transformer module is used to process the input sequence completely based on the self-attention mechanism. The token classification layer is used to divide the text data into at least one token, and the token can be understood as the smallest unit in the text. The BIO annotation layer is used to perform sequence annotation on the text data using the BIO annotation method to mark out the entities in the text. The citation classification layer is used to classify the text data as a whole according to whether there is a citation relationship between the text data, and divide the text data into two categories: those with citation relationships and those without citation relationships. The CRF layer is used to automatically learn constraint information from the training data, and add some constraints to the labels predicted by the model based on the constraint information, so as to ensure the validity and legitimacy of the predicted labels.

[0073] Figure 3 A schematic diagram of another dialogue interface provided by an embodiment of the present invention, such as Figure 3 As shown, the reply information includes reply content, two reference badge controls and five search results. Among them, the two reference badge controls are the first reference badge control and the second reference badge control. The user clicks on the second reference badge control, and the client responds to the user's operation of clicking on the second reference badge control by displaying a reference information display pop-up layer in the dialogue interface, and displays the reference verification information corresponding to the second reference badge control on the reference information display pop-up layer. The reference information display pop-up layer is displayed at the bottom of the dialogue interface, and the reference information display pop-up layer displays the words "reference verification information", and the reference verification content is "According to the latest data from Tonghuashun, as of the closing time on December 12, 2023, the latest share price of Ping An of China is 45.89, the highest price during the trading session is 46, and the lowest is 44." The publication time of the reference verification content is 2023-12-13.

[0074] Optionally, the reply message also includes a display status control, which has a collapsed state and an expanded state. In response to a user clicking the display status control in the collapsed state, the reply message's progress information is expanded; or, in response to a user clicking the display status control in the expanded state, the reply message's progress information is collapsed. The progress information includes at least one of at least one search enhancement subtask, at least one search result, and process tag information, the process tag information being used to mark the current processing step.

[0075] Figure 4a A schematic diagram of another dialogue interface provided by an embodiment of the present invention, such as Figure 4aAs shown, the demand information input by the user is "What are the average heights of Chinese and Japanese people?", and the reply information of the demand information includes the reply content and three search results, which are the first search result, the second search result and the third search result.

[0076] Figure 4b A schematic diagram of another dialogue interface provided by an embodiment of the present invention, such as Figure 4b As shown, the user clicks a collapsed display control. In response to this, the client expands the reply message. The client parses the request into two search enhancement subtasks: "What is the average height of Chinese people?" and "What is the average height of Japanese people?" The process information includes the second parsed title information, two search enhancement subtasks, the second search title information, and three search results.

[0077] Figure 5 A schematic diagram of another dialogue interface provided by an embodiment of the present invention, such as Figure 5 As shown, the user inputs the requirement "How to write a year-end summary." The client parses the requirement into an automatically generated subtask, which is "How to write a year-end summary." The process information includes the second parsed title information, an automatically generated subtask, the second search title information, and three search results.

[0078] The technical solution provided by the embodiments of this invention parses the user's input requirement and searches for the parsed subtasks to generate a response to the requirement. The model's thought process is explicitly displayed during the response generation process, improving the logical integrity of the response and alleviating the LLM hallucination problem. This enhances the credibility of the large language model and improves the user's conversational experience.

[0079] In an embodiment of the present invention, when the demand information input by the user belongs to a highly professional vertical field, the client can search the demand information through RAG, obtain relevant search results in the vertical field, and generate professional reply information based on the search results, thereby improving the user's conversation experience.

[0080] In an embodiment of the present invention, when the demand information input by the user has high timeliness requirements, the client can obtain the latest search results through RAG and generate reply information based on the latest search results, avoiding the problem of poor timeliness of reply information due to the timeliness limitation of training data of the large language model, thereby improving the user's conversation experience.

[0081] Figure 6 A schematic diagram of the structure of a reply information generating device provided by an embodiment of the present invention is shown as follows: Figure 6As shown, the device includes: a parsing module 11, a search module 12, and a reply module 13. The parsing module 11 is connected to the search module 12, and the search module 12 is connected to the reply module 13. The parsing module 11 is used to respond to the demand information input by the user in the dialogue interface, parse the demand information into at least one search enhancement subtask, and display the at least one search enhancement subtask in the dialogue interface; the search module 12 is used to search for the at least one search enhancement subtask, generate and display at least one search result corresponding to the at least one search enhancement subtask; and the reply module 13 is used to generate reply information for the demand information based on the at least one search result, and display the reply information in the dialogue interface.

[0082] In the embodiment of the present invention, the parsing module 11 is specifically configured to parse the demand information into at least one search enhancement subtask and at least one automatically generated subtask, and display the at least one search enhancement subtask and the at least one automatically generated subtask in the dialogue interface.

[0083] In the embodiment of the present invention, the parsing module 11 is specifically configured to parse the demand information into a plurality of search enhancement subtasks, and display the plurality of search enhancement subtasks in the dialogue interface.

[0084] In this embodiment of the present invention, the device further includes a redirect module 14, which is connected to the search module 12. The search results include a search result title control and the site information of the search result. Redirect module 14 is configured to, in response to a user clicking on the search result title control, redirect to a search result display interface associated with the search result title control based on the site information of the search result.

[0085] In the embodiment of the present invention, the device further includes a display module 15, which is connected to the search module 12. The display module 15 is configured to display adjacent search results in response to a user sliding operation on at least one search result.

[0086] In an embodiment of the present invention, the apparatus further includes a folding module 16, which is connected to the reply module 13. The folding module 16 is configured to fold process information for generating the reply information, where the process information includes at least one of at least one search enhancement subtask, at least one search result, and process marking information, where the process marking information is used to mark the current processing process.

[0087] In an embodiment of the present invention, the device further includes a verification module 17, which is connected to the reply module 13. The verification module 17 is configured to display a citation information display pop-up window in the dialog interface in response to a user clicking a citation badge control. The citation information display pop-up window is configured to display citation verification information corresponding to the citation badge control, allowing the user to perform citation verification on the reply content based on the citation verification information.

[0088] In an embodiment of the present invention, the device further includes a state switching module 18, which is connected to the reply module 13. The reply information further includes a display state control, which has a folded state and an expanded state. The state switching module 18 is configured to expand the process information for generating the reply information in response to a user clicking on the display state control in the folded state; or to collapse the process information for generating the reply information in response to a user clicking on the display state control in the expanded state; the process information includes at least one search enhancement subtask, at least one search result, and at least one process marking information, wherein the process marking information is used to mark the current processing process.

[0089] The technical solution provided by the embodiments of this invention parses the user's input requirement and searches for the parsed subtasks to generate a response to the requirement. The model's thought process is explicitly displayed during the response generation process, improving the logical integrity of the response and alleviating the LLM hallucination problem. This enhances the credibility of the large language model and improves the user's conversational experience.

[0090] In an embodiment of the present invention, when the demand information input by the user belongs to a highly professional vertical field, the client can search the demand information through RAG, obtain relevant search results in the vertical field, and generate professional reply information based on the search results, thereby improving the user's conversation experience.

[0091] In an embodiment of the present invention, when the demand information input by the user has high timeliness requirements, the client can obtain the latest search results through RAG and generate reply information based on the latest search results, avoiding the problem of poor timeliness of reply information due to the timeliness limitation of training data of the large language model, thereby improving the user's conversation experience.

[0092] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program. When the program is executed, the device where the computer-readable storage medium is located is controlled to execute an embodiment of the reference verification method described above.

[0093] Figure 7 A schematic diagram of a client provided for an embodiment of the present invention includes: the client 3 of this embodiment includes: a processor 31, a memory 32, and a computer program 33 stored in the memory 32 and executable on the processor 31. When the computer program 33 is executed by the processor 31, the reference verification method in the embodiment is implemented. To avoid repetition, they are not described here one by one.

[0094] The client 3 includes, but is not limited to, a processor 31 and a memory 32. Those skilled in the art will understand that Figure 7This is only an example of the client 3 and does not constitute a limitation on the client 3. The client 3 may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the network device may also include input and output devices, network access devices, buses, etc.

[0095] The processor 31 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0096] Memory 32 can be an internal storage unit of client 3, such as a hard drive or memory on client 3. Memory 32 can also be an external storage device on client 3, such as a plug-in hard drive, a SmartMedia Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. Furthermore, memory 32 can include both an internal storage unit and an external storage device on client 3. Memory 32 is used to store computer programs and other programs and data required by network devices. Memory 32 can also be used to temporarily store data that has been output or is about to be output.

[0097] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0098] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0099] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention pertain.

[0100] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0101] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, which may be electrical, mechanical or other forms.

[0102] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A reply message generation method, applied to a human-computer dialogue application based on generative artificial intelligence AIGC, characterized in that: The method comprises: In response to demand information input by a user in a dialogue interface, parsing the demand information into at least one search enhancement subtask, and displaying the at least one search enhancement subtask in the dialogue interface in a streaming output manner; Searching the at least one search enhancement subtask, generating and displaying at least one search result corresponding to the at least one search enhancement subtask, the search result including a search result title control and site information of the search result; Generating reply information for the requirement information based on the at least one search result, folding process information for generating the reply information, and displaying the reply information in the dialogue interface, wherein the reply information includes reply content, at least one reference badge control, and the at least one search result, and the process information includes at least one of the at least one search enhancement subtask, the at least one search result, and process marking information, wherein the process marking information is used to mark the current processing process; In response to a user clicking on the reference badge control, a reference information display floating layer is displayed in the dialog interface, wherein the reference information display floating layer is used to display reference verification information corresponding to the reference badge control, so that the user can perform reference verification on the reply content according to the reference verification information; In response to a user clicking on the search result title control, jumping to a search result display interface associated with the search result title control according to the site information of the search result; In response to a user sliding operation of the at least one search result, adjacent search results are displayed.

2. The method according to claim 1, characterized in that The step of parsing the demand information into at least one search enhancement subtask and displaying the at least one search enhancement subtask in the dialogue interface includes: The demand information is parsed into at least one search enhancement subtask and at least one automatically generated subtask, and the at least one search enhancement subtask and the at least one automatically generated subtask are displayed in the dialogue interface.

3. The method according to claim 1, characterized in that The step of parsing the demand information into at least one search enhancement subtask and displaying the at least one search enhancement subtask in the dialogue interface includes: The demand information is parsed into a plurality of search enhancement subtasks, and the plurality of search enhancement subtasks are displayed in the dialogue interface.

4. The method according to claims 1-3, characterized in that The reply message further includes a display state control, wherein the display state control has a folded state and an expanded state. The method further includes: In response to a user clicking on a display status control in a folded state, unfolding the process information of the reply information; or In response to a user clicking on a display status control in an expanded state, folding the process information of the reply information; The process information includes at least one of the at least one search enhancement subtask, the at least one search result, and process marking information, where the process marking information is used to mark the current processing process.

5. A reply information generating device, used in AIGC human-computer dialogue application, characterized in that: The device comprises: a parsing module, configured to, in response to demand information input by a user in a dialogue interface, parse the demand information into at least one search enhancement subtask, and display the at least one search enhancement subtask in the dialogue interface in a streaming output manner; A search module, configured to search for the at least one search enhancement subtask, generate and display at least one search result corresponding to the at least one search enhancement subtask, the search result including a search result title control and site information of the search result; a reply module, configured to generate reply information for the requirement information based on the at least one search result, fold process information for generating the reply information, and display the reply information in the dialogue interface, wherein the reply information includes reply content, at least one reference badge control, and the at least one search result, and the process information includes at least one of the at least one search enhancement subtask, the at least one search result, and process marking information, wherein the process marking information is used to mark the current processing process; a jump module, configured to jump to a search result display interface associated with the search result title control in response to a user clicking the search result title control and according to the site information of the search result; a verification module, configured to display a citation information display floating layer in the dialog interface in response to a user clicking on a citation badge control, wherein the citation information display floating layer is configured to display citation verification information corresponding to the citation badge control, so that the user can perform citation verification on the reply content according to the citation verification information; The display module is configured to display adjacent search results in response to a user sliding the at least one search result.

6. A client, characterized in that: include: Display screen; one or more processors; Memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions, which, when executed by the client, enable the client to perform the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 4.

Images