Intelligent question and answer processing method and storage medium

Abstract

The invention discloses an intelligent question and answer processing method, which comprises the steps of acquiring an intention recognition result corresponding to user input information recognized based on a current first intention recognition model, determining reply content corresponding to the user input information based on the intention recognition result, and determining feedback information corresponding to the reply content based on the reply content, and taking the user input information, the reply content and the feedback information as training samples, and performing increment fine tuning on the first intention recognition model based on the training samples under the condition that a preset condition is met so as to generate a second intention recognition model which is used for intention recognition of the next user input information. According to the method, the intention recognition accuracy of the information input by the user can be improved, and the intention recognition model is continuously optimized based on feedback in the interaction process, so that continuous evolution of the intention recognition capability is realized.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method for processing intelligent question answering and a storage medium. Background Technology

[0002] In enterprise collaborative work scenarios, intelligent question-answering robots based on instant messaging platforms are widely used to answer common questions.

[0003] In existing technologies, common intelligent question answering methods mainly include question answering robots based on keyword matching. This approach typically listens to user messages in the conversation and matches the user input information with pre-configured keywords or regular expression rules one by one. When a corresponding rule is matched, a preset fixed response is returned. The relevant keyword library or rule configuration is usually stored in the form of a configuration file and loaded once when the system starts.

[0004] However, this solution can only process user input that closely matches preset keywords. It struggles to accurately identify the true intent of users whose input differs in expression or semantics but with different keywords. Furthermore, updating the keyword database or rule configuration typically requires manual maintenance and service restarts, making it difficult to continuously optimize the system based on user feedback during actual use. Summary of the Invention

[0005] To address the aforementioned issues, this invention proposes an intelligent question-answering processing method and storage medium that can improve the accuracy of intent recognition of user input information and enable the intent recognition model to be continuously optimized based on feedback during the interaction process, thereby achieving continuous evolution of intent recognition capabilities.

[0006] This invention provides a method for intelligent question answering, the method comprising: obtaining an intent recognition result corresponding to user input information based on a current first intent recognition model; determining a response content corresponding to the user input information based on the intent recognition result; determining feedback information corresponding to the response content based on the response content; using the user input information, the response content, and the feedback information as training samples, and, under preset conditions, incrementally fine-tuning the first intent recognition model based on the training samples to generate a second intent recognition model, wherein the second intent recognition model is used for intent recognition of the next user input information.

[0007] In one embodiment, the step of determining the response content corresponding to the user input information based on the intent recognition result includes: when it is confirmed that the user input information requires private domain knowledge retrieval based on the intent recognition result, obtaining target private domain knowledge according to the user input information, the intent recognition result, and the private domain knowledge; and obtaining the response content corresponding to the user input information based on the target private domain knowledge, the user input information, and the response model.

[0008] In one embodiment, obtaining target private domain knowledge based on the user input information, the intent recognition result, and private domain knowledge includes: obtaining retrieval features based on the user input information and the intent recognition result; performing a hybrid retrieval based on the private domain knowledge and the retrieval features to obtain first candidate private domain knowledge; reordering the first candidate private domain knowledge based on the user input information and the intent recognition result to obtain second candidate private domain knowledge; and performing permission filtering on the second candidate private domain knowledge to obtain the target private domain knowledge.

[0009] In one embodiment, the retrieval features include dense retrieval features and sparse retrieval features; the step of performing a hybrid retrieval based on the private domain knowledge and the retrieval features to obtain the first candidate private domain knowledge includes: performing a dense index retrieval based on the private domain knowledge and the dense retrieval features to obtain dense index candidate private domain knowledge; performing a sparse index retrieval based on the private domain knowledge and the sparse retrieval features to obtain sparse index candidate private domain knowledge; and obtaining the first candidate private domain knowledge based on the dense index candidate private domain knowledge and the sparse index candidate private domain knowledge.

[0010] In one embodiment, based on the user input information and the intent recognition result, the first candidate private domain knowledge is reordered to obtain the second candidate private domain knowledge, including: obtaining a relevance score for each private domain knowledge in the first candidate private domain knowledge based on the relevance between the first candidate private domain knowledge and the user input information and the intent recognition result; sorting the private domain knowledge in descending order based on the relevance scores of the private domain knowledge; and setting the top N private domain knowledge in the sorting as the second candidate private domain knowledge.

[0011] In one embodiment, the second candidate private domain knowledge is subjected to permission filtering to obtain the target private domain knowledge, including: obtaining the user permission level corresponding to the user input information; obtaining the visibility permission level of each private domain knowledge in the second candidate private domain knowledge; and setting the private domain knowledge in the second candidate private domain knowledge whose visibility permission level is equal to or less than the user permission level as the target private domain knowledge.

[0012] In one embodiment, obtaining the response content corresponding to the user input information based on the target private domain knowledge, the user input information, and the response model includes: generating a guidance prompt based on the target private domain knowledge; obtaining context data corresponding to the user input information according to historical question and answer information; and inputting the guidance prompt, the context data, and the user input information into the response model to obtain the response content corresponding to the user input information output by the response model.

[0013] In one embodiment, the user input information, the response content, and the feedback information are used as training samples. Under the condition of satisfying a preset condition, the first intent recognition model is incrementally fine-tuned based on the training samples to generate a second intent recognition model. The second intent recognition model is used for the intent recognition of the next user input information. The steps include: satisfying the preset condition when the number of training samples is greater than or equal to the training threshold, or when the current time reaches a preset periodic trigger time point.

[0014] In one embodiment, the method includes: acquiring incremental knowledge based on positive feedback training samples in the training samples; and updating private domain knowledge based on the incremental knowledge.

[0015] The present invention also provides a storage medium storing a computer program, which, when executed by a processor, implements the steps of the intelligent question-answering processing method described above.

[0016] The intelligent question-answering processing method and storage medium provided by this invention acquire the intent recognition result corresponding to the user input information based on the current first intent recognition model, determine the response content corresponding to the user input information based on the intent recognition result, determine the feedback information corresponding to the response content based on the response content, use the user input information, response content, and feedback information as training samples, and incrementally fine-tune the first intent recognition model based on the training samples under preset conditions to generate a second intent recognition model, wherein the second intent recognition model is used for intent recognition of the next user input information. This invention can improve the accuracy of intent recognition of user input information and enable the intent recognition model to be continuously optimized based on feedback during the interaction process, thereby achieving continuous evolution of intent recognition capabilities. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating an intelligent question-answering processing method according to an embodiment of the present invention.

[0018] Figure 2 for Figure 1 A flowchart illustrating step S12.

[0019] Figure 3 This is a schematic diagram of the structure of an intelligent question-answering processing system according to an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram of the structure of a computing device according to an embodiment of the present invention.

[0021] Figure labeling: Event access layer - 301; Preprocessing layer - 302; AI agent layer - 303; Answer rendering layer - 304; Feedback learning layer - 305; Private domain knowledge layer - 306; Processor - 410; Storage - 411; Network interface - 412; Bus system - 413. Detailed Implementation

[0022] The foregoing and other technical contents, features, and effects of the present invention will be clearly presented in the following detailed description of preferred embodiments with reference to the accompanying drawings. Through the description of the specific embodiments, a more in-depth and specific understanding can be gained of the technical means and effects adopted by the present invention to achieve its intended purpose. However, the accompanying drawings are for reference and illustration only and are not intended to limit the present invention.

[0023] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0024] Figure 1 This is a flowchart illustrating an intelligent question-answering processing method according to an embodiment of the present invention.

[0025] like Figure 1 As shown, the intelligent question-answering processing method provided in this embodiment includes the following steps: Step S11: Obtain the intent recognition result corresponding to the user input information based on the current first intent recognition model.

[0026] Specifically, a long connection is maintained with the server (such as Lark server, DingTalk server, etc.) to listen for message events on the server (such as group chat messages, one-on-one chat messages, card messages, etc.). After deduplicating the listened message events, the deduplicated message events are formatted to obtain user input information in a preset format. The current first intent recognition model is then used to recognize the user input information to obtain the intent recognition result corresponding to the user input information.

[0027] Step S12: Based on the intent recognition result, determine the response content corresponding to the user input information.

[0028] like Figure 2 As shown, step S12 includes: Step S121: Based on the intent recognition result, when it is confirmed that the user input information requires private domain knowledge retrieval, the target private domain knowledge is obtained according to the user input information, the intent recognition result, and the private domain knowledge.

[0029] Specifically, when the response type corresponding to the intent recognition result is a response type that requires the use of private domain knowledge for generation, it is determined that the user input information needs to be retrieved using private domain knowledge; when the response type is a general generative response type, it is determined that the user input information does not need to be retrieved using private domain knowledge.

[0030] Specifically, based on user input information and intent recognition results, retrieval features are obtained. Based on private domain knowledge, a hybrid retrieval is performed on the retrieval features to obtain first candidate private domain knowledge. Based on user input information and intent recognition results, the first candidate private domain knowledge is reordered according to relevance to obtain second candidate private domain knowledge. The second candidate private domain knowledge is then filtered for permissions to obtain the target private domain knowledge.

[0031] In one embodiment, contextual data corresponding to the user input information is obtained based on historical question-and-answer information corresponding to the user input information; retrieval features are obtained based on the user input information, intent recognition results, and contextual data.

[0032] Specifically, the retrieval features include dense retrieval features and sparse retrieval features; based on private domain knowledge and dense retrieval features, dense index retrieval is performed to obtain a first preset number (e.g., 10) of dense index candidate private domain knowledge; based on private domain knowledge and sparse retrieval features, sparse index retrieval is performed to obtain a second preset number (e.g., 10) of sparse index candidate private domain knowledge; the dense index candidate private domain knowledge and the sparse index candidate private domain knowledge are set as the first candidate private domain knowledge.

[0033] Specifically, based on the correlation between each private knowledge in the first candidate private knowledge and the user input information and intent recognition results, a correlation score is obtained for each private knowledge in the first candidate private knowledge. Based on the correlation score of the private knowledge, the private knowledge in the first candidate private knowledge is sorted in descending order, and the top N (e.g. 5) private knowledge in the sorted order are set as the second candidate private knowledge. This can significantly improve the recall accuracy of private knowledge, thereby solving the large model illusion problem.

[0034] In one embodiment, a relevance score is obtained for each piece of private knowledge in the first candidate private knowledge based on its relevance to user input information, intent recognition results, and contextual data.

[0035] Specifically, the system obtains the user permission level corresponding to the user input information, obtains the visibility permission level of each private domain knowledge in the second candidate private domain knowledge, and sets the private domain knowledge in the second candidate private domain knowledge whose visibility permission level is equal to or less than the user permission level as the target private domain knowledge, which can prevent the leakage of private domain data due to unauthorized access.

[0036] Step S122: Based on the target private domain knowledge, the user input information, and the response model, obtain the response content corresponding to the user input information.

[0037] Specifically, based on target private domain knowledge, a guided response model is generated to provide guidance prompts for the response. According to the historical question and answer information corresponding to the user input information, the context data corresponding to the user input information is obtained. The guidance prompts, context data and user input information are input into the response model to obtain the response content corresponding to the user input information output by the response model.

[0038] Step S13: Based on the response content, determine the feedback information corresponding to the response content.

[0039] Specifically, after determining the response content corresponding to the user's input information, the format of the response content is converted into a server-supported format, and the converted response content is streamed to the server. Preferably, the response content is converted into a server-supported card message format, specifically set to schema 2.0, update_multi=true, element_id=streaming_txt, and the button value of the card message contains the message_id. Since the response model outputs the response content word by word, rather than outputting the complete response content at once, a streaming transmission is triggered when the response content output by the response model exceeds a preset content block.

[0040] Users can provide feedback and evaluation on the response from the server (e.g., by clicking "helpful" or "not helpful" under the response). Therefore, they can receive feedback information from the server regarding the response to determine the corresponding feedback information.

[0041] In one embodiment, a feedback setting is implemented for the card message format. This setting allows the same user to provide feedback on the same response only once within a preset time. After receiving the user's first feedback, the server grays out and disables the "helpful / not helpful" button on the user's card via the UpdateCard interface. This avoids training sample skew and training data contamination caused by users providing multiple evaluations of the same response.

[0042] In one embodiment, display permission settings are configured for the card message format. These display permission settings are used to control the display scope of the response content, ensuring that the response content is only displayed to users whose user permission level is greater than or equal to the visibility permission level corresponding to the response content. This further ensures the access security of private domain knowledge during the display stage after the response content is generated, and avoids unauthorized visibility of private domain knowledge due to message forwarding or group chat scenarios.

[0043] Step S14: Use the user input information, the response content, and the feedback information as training samples, and, under the condition of meeting preset conditions, incrementally fine-tune the first intent recognition model based on the training samples to generate a second intent recognition model, wherein the second intent recognition model is used for intent recognition of the next user input information.

[0044] Specifically, the feedback information, the corresponding user input information, and the response content are written into training samples in triplet format and saved to a specific location. When the number of training samples is greater than or equal to the training threshold (e.g., 2000), or when the current time reaches a preset periodic trigger time (e.g., 3 AM every Monday), the first intent recognition model is incrementally fine-tuned based on the existing training samples to generate a second intent recognition model. The second intent recognition model is then used for the intent recognition of subsequent user input information. After generating the second intent recognition model, the existing training samples are set as historical training samples.

[0045] In one implementation, the specific parameters for incremental fine-tuning are rank=8 and learning rate lr=3e-4.

[0046] Step S13 and beyond also includes: based on feedback information, selecting positive feedback training samples (i.e. training samples generated by users clicking "help"), extracting user input information and response content from the positive feedback training samples as incremental knowledge, embedding the question and answer content corresponding to the incremental knowledge into a vector representation to obtain an incremental knowledge vector, and appending the incremental knowledge vector to the existing index of the private domain knowledge to update the private domain knowledge, thereby realizing the non-stop expansion and backflow of private domain knowledge.

[0047] In summary, the intelligent question-answering processing method provided by this invention obtains the intent recognition result corresponding to the user input information based on the current first intent recognition model, determines the response content corresponding to the user input information based on the intent recognition result, determines the feedback information corresponding to the response content based on the response content, uses the user input information, response content, and feedback information as training samples, and incrementally fine-tunes the first intent recognition model based on the training samples under preset conditions to generate a second intent recognition model. The second intent recognition model is used for intent recognition of the next user input information. This invention can improve the accuracy of intent recognition of user input information and enable the intent recognition model to be continuously optimized based on feedback during the interaction process, thereby achieving continuous evolution of intent recognition capabilities.

[0048] Based on the same inventive concept as the foregoing embodiments, this invention provides an intelligent question-answering processing system, such as... Figure 3 As shown, the intelligent question-answering processing system includes: event access layer 301, preprocessing layer 302, AI intelligent agent layer 303, answer rendering layer 304, feedback learning layer 305, and private domain knowledge layer 306.

[0049] The event access layer 301 is used to maintain a connection with the server via a long connection and listen for message events from the server.

[0050] The preprocessing layer 302 is used to deduplicatize the listened message events and then convert the format of the deduplicated message events to obtain user input information in a preset format.

[0051] The AI ​​intelligent agent layer 303 is used to obtain target private domain knowledge based on the user input information, the intent recognition result, and the private domain knowledge when it is confirmed that the user input information requires private domain knowledge retrieval based on the intent recognition result. Based on the target private domain knowledge, the user input information, and the response model, the response content corresponding to the user input information is obtained.

[0052] The answer rendering layer 304 is used to convert the format of the reply content into a format supported by the server, and then stream the converted reply content to the server.

[0053] The feedback learning layer 305 is used to take user input information, response content and feedback information as training samples, and, under the condition of meeting the preset conditions, to incrementally fine-tune the first intent recognition model based on the training samples to generate the second intent recognition model.

[0054] The private domain knowledge layer 306 is used to filter positive feedback training samples based on feedback information, and extract user input information and response content from the positive feedback training samples as incremental knowledge. The question and answer content corresponding to the incremental knowledge is embedded into a vector representation to obtain the incremental knowledge vector. The incremental knowledge vector is appended to the existing index of the private domain knowledge to update the private domain knowledge.

[0055] The intelligent question-answering processing system in this embodiment is used to implement the intelligent question-answering processing method described above. The functions implemented by each module of the intelligent question-answering processing system can be found in [reference needed]. Figure 1 , Figure 2 The descriptions in the embodiments are not repeated here.

[0056] Based on the same inventive concept as the foregoing embodiments, this embodiment of the invention provides a computing device, such as... Figure 4 As shown, the computing device includes: a processor 410 and a memory 411 storing computer programs; wherein, Figure 4 The processor 410 shown in the diagram does not indicate that there is only one processor 410, but only indicates the positional relationship of the processor 410 relative to other devices. In practical applications, there can be one or more processors 410; similarly, Figure 4 The memory 411 shown in the diagram has the same meaning, that is, it is only used to indicate the positional relationship of memory 411 relative to other devices. In practical applications, there can be one or more memories 411. When the processor 410 runs the computer program, the above-mentioned intelligent question-answering processing method is implemented.

[0057] The computing device may also include at least one network interface 412. Various components of the computer device are coupled together via a bus system 413. It is understood that the bus system 413 is used to implement communication between these components. In addition to a data bus, the bus system 413 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 4 The general designated all buses as Bus System 413.

[0058] The memory 411 can be volatile or non-volatile, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memory 411 described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0059] The memory 411 in this embodiment of the invention is used to store various types of data to support the operation of the computing device. Examples of this data include any computer programs used to operate on the computing device, such as operating systems and applications. The operating system includes various system programs, such as framework layers, core library layers, and driver layers, used to implement various basic business functions and handle hardware-based tasks. Applications can include various applications, such as media players and browsers, used to implement various application functions. Here, the program implementing the method of this embodiment of the invention can be included in the application.

[0060] Based on the same inventive concept as the foregoing embodiments, this embodiment also provides a storage medium storing a computer program. The storage medium can be a magnetic random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM), etc.; it can also be various devices including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc. When the computer program stored in the storage medium is run by a processor, it implements the above-described intelligent question-answering processing method. For the specific steps implemented when the computer program is executed by the processor, please refer to [link to relevant documentation]. Figure 1 , Figure 2 The description of the illustrated embodiments will not be repeated here.

[0061] The intelligent question-answering processing method and storage medium provided by this invention acquire the intent recognition result corresponding to the user input information based on the current first intent recognition model, determine the response content corresponding to the user input information based on the intent recognition result, determine the feedback information corresponding to the response content based on the response content, use the user input information, response content, and feedback information as training samples, and incrementally fine-tune the first intent recognition model based on the training samples under preset conditions to generate a second intent recognition model, wherein the second intent recognition model is used for intent recognition of the next user input information. This invention can improve the accuracy of intent recognition of user input information and enable the intent recognition model to be continuously optimized based on feedback during the interaction process, thereby achieving continuous evolution of intent recognition capabilities.

[0062] It should be understood that although the steps in the flowcharts of the accompanying drawings are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple sub-steps or multiple stages, which are not necessarily completed at the same time, but may be executed at different times, and their execution order is not necessarily sequential, but may be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that the embodiments of the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions of the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the embodiments of the present invention.

[0064] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for intelligent question answering, characterized in that, The method includes: Obtain the intent recognition result corresponding to the user input information based on the current first intent recognition model; Based on the intent recognition result, the response content corresponding to the user input information is determined; Based on the content of the response, determine the feedback information corresponding to the content of the response; The user input information, the response content, and the feedback information are used as training samples. Under preset conditions, the first intent recognition model is incrementally fine-tuned based on the training samples to generate a second intent recognition model. The second intent recognition model is used for intent recognition of the next user input information.

2. The intelligent question-answering processing method as described in claim 1, characterized in that, The step of determining the response content corresponding to the user input information based on the intent recognition result includes: Based on the intent recognition result, when it is confirmed that the user input information requires private domain knowledge retrieval, the target private domain knowledge is obtained according to the user input information, the intent recognition result, and the private domain knowledge. Based on the target private domain knowledge, the user input information, and the response model, the response content corresponding to the user input information is obtained.

3. The intelligent question-answering processing method as described in claim 2, characterized in that, Based on the user input information, the intent recognition result, and private domain knowledge, target private domain knowledge is obtained, including: Based on the user input information and the intent recognition result, the retrieval features are obtained; A hybrid retrieval is performed based on the private domain knowledge and the retrieval features to obtain the first candidate private domain knowledge. Based on the user input information and the intent recognition result, the first candidate private domain knowledge is reordered to obtain the second candidate private domain knowledge; The second candidate private domain knowledge is filtered for permissions to obtain the target private domain knowledge.

4. The intelligent question-answering processing method as described in claim 3, characterized in that, The search features include dense search features and sparse search features; The step of performing a hybrid search based on the private domain knowledge and the search features to obtain the first candidate private domain knowledge includes: Based on the aforementioned private domain knowledge and dense retrieval features, dense index retrieval is performed to obtain dense index candidate private domain knowledge. Based on the aforementioned private domain knowledge and sparse retrieval features, sparse index retrieval is performed to obtain sparse index candidate private domain knowledge. Based on the dense index candidate private domain knowledge and the sparse index candidate private domain knowledge, the first candidate private domain knowledge is obtained.

5. The intelligent question-answering processing method as described in claim 3, characterized in that, Based on the user input information and the intent recognition result, the first candidate private domain knowledge is reordered to obtain the second candidate private domain knowledge, including: Based on the correlation between the first candidate private domain knowledge and the user input information and the intent recognition result, a correlation score is obtained for each private domain knowledge in the first candidate private domain knowledge. Based on the relevance score of the private domain knowledge, the private domain knowledge is sorted in descending order; The top N private domain knowledge items are designated as the second candidate private domain knowledge items.

6. The intelligent question-answering processing method as described in claim 3, characterized in that, To obtain the target private domain knowledge, the second candidate private domain knowledge is subjected to permission filtering, including: Obtain the user permission level corresponding to the user input information; Obtain the visibility permission level of each private domain knowledge in the second candidate private domain knowledge; Among the second candidate private domain knowledge, the private domain knowledge whose visibility permission level is equal to or less than the user permission level is set as the target private domain knowledge.

7. The intelligent question-answering processing method as described in claim 2, characterized in that, Based on the target private domain knowledge, the user input information, and the response model, the response content corresponding to the user input information is obtained, including: Based on the target private domain knowledge, a guidance prompt is generated; Based on historical question and answer information, obtain the context data corresponding to the user input information; The guidance prompts, the context data, and the user input information are input into the response model to obtain the response content corresponding to the user input information output by the response model.

8. The intelligent question-answering processing method as described in claim 1, characterized in that, Using the user input information, the response content, and the feedback information as training samples, and under preset conditions, incrementally fine-tuning the first intent recognition model based on the training samples to generate a second intent recognition model, wherein the second intent recognition model is used for the intent recognition of the next user input information includes: The preset condition is met when the number of training samples is greater than or equal to the training threshold, or when the current time reaches a preset periodic trigger time point.

9. The intelligent question-answering processing method as described in claim 1, characterized in that, The method includes: Incremental knowledge is obtained based on the positive feedback training samples in the training samples; Based on the incremental knowledge, update the private domain knowledge.

10. A storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the intelligent question-answering processing method as described in any one of claims 1 to 9.

Images