Question and answer method, device and medium under multi-reply form

Abstract

Embodiments of the present disclosure disclose a question and answer method, device, equipment and medium under multiple reply forms. A specific implementation of the method comprises: in response to inputting a health-related question, obtaining user identity information; determining a question reply form; inputting the health-related question into a multi-modal encoding model to obtain question encoding information; inputting the question encoding information into a multi-question and answer form reply model to generate first reply content and second reply content; in response to the first reply form, displaying the first reply content; in response to selecting a one-key upload control, sending page data corresponding to the first reply content and page data corresponding to the second reply content to a personalized intelligent agent. The implementation can flexibly output reply content under the corresponding reply form according to the identity information corresponding to the target user, so as to facilitate the target user to understand offline reply content in the medical field, and support page data transmission to the personalized intelligent agent for information storage.

Description

Technical Field

[0001] Embodiments of this disclosure relate to the field of computer technology, and more particularly to question-and-answer methods, apparatus, electronic devices, and computer-readable media in a multi-response format. Background Technology

[0002] Currently, with the continuous development of artificial intelligence and intelligent agents, applying AI and intelligent agent technologies to the health field has become one of the current development directions. For health-related questions and answers, the common approach is to directly utilize large language models to generate professional answers to health-related questions.

[0003] However, when using the above method, the following technical problems often arise: Answers to professional questions about health issues often contain a large number of medical terms. For people with different levels of medical knowledge, these terms require explanations from professionals or analysis using other intelligent tools, which is inflexible and inefficient.

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure provide question-and-answer methods, apparatuses, electronic devices, and computer-readable media in a multi-response format to address one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a question-and-answer method with multiple response formats, including: in response to a target user inputting a health-related question on a health dialogue page in a target hospital system, obtaining user identity information corresponding to the target user; determining, based on the user identity information, a default question response format to be displayed on the health dialogue page, wherein the question response format is one of the following: a first response format for users without health knowledge, and a second response format for users with health knowledge; inputting the health-related question into a multimodal coding model included in a pre-trained question-and-answer large language model to obtain question coding information; inputting the question coding information into a response model with multiple question-and-answer formats included in the question-and-answer large language model to generate a first response content under the first response format and a second response content under the second response format; in response to the question response format being the first response format, displaying the first response content in the response area corresponding to the health dialogue page, wherein the response area has a one-click upload control; in response to selecting the one-click upload control in the response area, sending the page data corresponding to the first response content and the page data corresponding to the second response content to the personalized intelligent agent corresponding to the target user based on the user identity information.

[0008] Secondly, some embodiments of this disclosure provide a question-answering device with multiple response formats, including: an acquisition unit configured to acquire user identity information corresponding to the target user in response to a target user inputting a health-related question on a health dialogue page in a target hospital system; a determination unit configured to determine, based on the user identity information, a default question response format displayed on the health dialogue page, wherein the question response format is one of the following: a first response format for users without health knowledge, and a second response format for users with health knowledge; a first generation unit configured to input the health-related question into a multimodal coding model included in a pre-trained question-answering large language model to obtain question coding information; and a second generation unit configured to... The system comprises: a unit configured to input the aforementioned question encoding information into the multi-question-answer format response model included in the aforementioned question-answering language model, to generate a first response content under the aforementioned first response format and a second response content under the aforementioned second response format; a display unit configured to, in response to the aforementioned question response format being the aforementioned first response format, display the aforementioned first response content in the response area corresponding to the aforementioned health dialogue page, wherein the response area contains a one-click upload control; and a sending unit configured to, in response to the aforementioned one-click upload control being selected in the response area, send the page data corresponding to the aforementioned first response content and the page data corresponding to the aforementioned second response content to the personalized intelligent agent corresponding to the aforementioned target user, based on the aforementioned user identity information.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, such that when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method as described in any implementation of the first aspect.

[0011] The above embodiments of this disclosure have the following beneficial effects: Through the multi-response question-and-answer method of some embodiments of this disclosure, the response content in the corresponding response format can be flexibly output according to the target user's identity information, so that the target user can understand the offline response content in the medical field, and support the transmission of page data to a personalized intelligent agent for information storage. Specifically, the reason why the responses to questions in the relevant medical field are not flexible enough is that the responses to professional questions about health issues often contain a large number of medical professional terms. For people with different levels of medical knowledge, professional personnel need to explain or use other intelligent tools for analysis, which is inflexible and inefficient. Based on this, the multi-response question-and-answer method of some embodiments of this disclosure firstly, in response to the target user inputting a health-related question on the health dialogue page of the target hospital system, obtains the user identity information corresponding to the target user. Here, by obtaining the user identity information, the subsequent response content display method is determined so that the response content that the user can directly understand is displayed on the health dialogue page. Then, based on the user identity information, the default question response format displayed on the health dialogue page can be accurately determined. The question response format is one of the following: a first response format for users without health knowledge, and a second response format for users with health knowledge. Next, the aforementioned health-related questions are input into the multimodal coding model included in the pre-trained question-and-answer large language model, which accurately obtains the question coding information to facilitate the generation of content in different response formats. Then, the question coding information is input into the multi-question-and-answer format response model included in the aforementioned question-and-answer large language model to accurately generate the first response content under the first response format and the second response content under the aforementioned second response format. Secondly, in response to the question response format being the aforementioned first response format, the first response content is displayed in the response area corresponding to the aforementioned health dialogue page, where a one-click upload control is provided. Here, based on the user's identity, response content of different levels of expertise is displayed, allowing the user to more accurately and clearly understand the response results. Finally, in response to selecting the aforementioned one-click upload control in the response area, based on the aforementioned user identity information, the page data corresponding to the aforementioned first response content and the page data corresponding to the aforementioned second response content are sent to the personalized intelligent agent corresponding to the target user for personalized storage of the individual's response content. In summary, by prioritizing the display of response content in the corresponding response format based on user identity information, the system can flexibly output response content in the corresponding response format, so that target users can understand the offline response content in the medical field, and support the transfer of page data to a personalized intelligent agent for information storage. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 These are flowcharts of some embodiments of the question-and-answer method in the multi-response format according to this disclosure; Figure 2 These are schematic diagrams of some embodiments of a question-and-answer device in a multi-response format according to the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] refer to Figure 1 The flowchart 100 illustrates some embodiments of a question-and-answer method in a multi-response format according to this disclosure. This question-and-answer method in a multi-response format includes the following steps: Step 101: In response to the target user entering a health-related question on the health dialogue page of the target hospital system, obtain the user identity information corresponding to the target user.

[0021] In some embodiments, in response to a target user inputting health-related questions on a health dialogue page within a target hospital system, the executing entity of the aforementioned multi-response question-and-answer method (e.g., an electronic device) can obtain the user's identity information via wired or wireless means. The target user can be a user who raises questions within the target hospital system offline. Specifically, this can be a scenario where a user undergoes offline diagnosis at a hospital, and the target hospital system can be configured to answer relevant health questions for the user's reference. For example, while the user is awaiting diagnosis, the target hospital system can be used to allow the user to consult on relevant health issues in advance. Another example is when a user still has questions about relevant health issues after diagnosis at a hospital, and can use the target hospital system to obtain the information they need. The target hospital system can be the hospital system deployed at the hospital where the target user is receiving treatment. The hospital system supports diverse functions. The target hospital system supports intelligent question-and-answer functionality. The target hospital system can be in the form of an application, supporting installation on mobile devices. Alternatively, the target hospital system can be deployed on medical robots within the hospital, allowing various users within the hospital to engage in medical knowledge question-and-answer sessions. The health dialogue page can be a page for engaging in dialogue about health knowledge and answering user questions. In practice, clicking the question-and-answer control in the target hospital system can redirect to the health dialogue page for health-related Q&A. The health dialogue page includes: a question input box and a domain selection control. Specifically, the user can enter their question in the question input box. The domain selection control allows selecting the domain of the desired resource. If no domain selection control is selected, a search can be performed on various domains for the user's question, but the search speed may be slower than when a domain is selected. Health-related questions can be questions related to health knowledge. For example, a health-related question could be a user's inquiry about precautions for a specific disease, or a question about the diagnosis results from a diagnosing doctor. User identity information can be the user's own identity. Specifically, user identity information can be one of the following: a medical worker, a medical academic, or an ordinary person without medical knowledge.

[0022] As an example, the aforementioned implementing entity can obtain the user identity information corresponding to the target user from the personnel database in the target hospital system. Specifically, the personnel data in the personnel database can be obtained by registering or logging into the target hospital system, as well as acquiring individual user information and personalized intelligent agent information. Here, each user can register and log in to a dedicated personalized intelligent agent for maintaining their own health. The personalized intelligent agent can be an intelligent agent that assists users in maintaining their own health and engaging in related health knowledge dialogues. In practice, the personalized intelligent agent can support health status inquiries, hospital appointment bookings, medical result inquiries, medical knowledge dialogues, and health status alerts.

[0023] Step 102: Based on the user identity information mentioned above, determine the default question and answer format to be displayed on the health dialogue page.

[0024] In some embodiments, the aforementioned executing entity can determine the default question and answer format displayed on the health dialogue page based on the aforementioned user identity information. The question and answer format can be the display format of the answer content on the health dialogue page. Specifically, the question and answer format can be one of the following: a first answer format for users without health knowledge, and a second answer format for users with health knowledge. Users without health knowledge are those who have no knowledge of medical and health-related knowledge or a low level of knowledge. That is, users without health knowledge may be those who cannot understand professional medical knowledge. For example, users without health knowledge may be those who have never received medical and health training or learned about medical and health knowledge. Users with health knowledge are those who have systematically learned or mastered medical and health knowledge. For example, users with health knowledge may be attending physicians in hospitals or medical academics. The default question and answer format can indicate which form of answer content will be displayed on the subsequent health dialogue page. That is, for users whose identity information is that of a medical worker and / or a medical academic, the corresponding question and answer format is the second answer format. For users whose identity information is that of an ordinary person without medical knowledge, the corresponding question and answer format is the first answer format.

[0025] As an example, the aforementioned executing entity can query the question and answer format corresponding to the user's identity information from the mapping table, and use it as the default question and answer format to be displayed. This mapping table represents the mapping relationship between user identity information and question and answer formats. For example, the mapping information could be "Medical Academic Personnel Identity - Second Response Format" or "Ordinary Personnel Identity - First Response Format".

[0026] Step 103: Input the above health-related questions into the multimodal coding model included in the pre-trained question-answering large language model to obtain question coding information.

[0027] In some embodiments, the aforementioned execution entity can input the aforementioned health-related questions into a pre-trained question-answering large language model, which includes a multimodal encoding model, to obtain question encoding information. The question-answering large language model can be a large language model for answering medical and health questions. In practice, the question-answering large language model can be a multimodal large language model. In practice, the question-answering large language model can include: a multimodal encoding model and a multi-question-answer format response model. The multimodal encoding model can be a model that encodes multimodal input content. In practice, the multimodal encoding model can be an encoding model based on the Transformer architecture. Specifically, the modalities of the input content supported by the health dialogue page include: text modality, audio modality, and image modality. The corresponding multimodal encoder can include: a text encoder, an audio encoder, an image encoder, a multimodal alignment layer, an attention-based residual layer, and a multimodal feature encoding layer. The multimodal alignment layer can be a cross-attention layer. The attention-based residual layer can be a network layer based on a multi-head attention mechanism residual architecture. A multi-question-answer format response model can consist of multiple response models corresponding to various question-answer formats. That is, each question-answer format has a corresponding response model. For example, a multi-question-answer format response model may include: a model for outputting the response content corresponding to the first response format and a model for outputting the response content corresponding to the second response format. In practice, each response model can be a response model based on a decoding structure. The network structures corresponding to each response model can be the same or different. In practice, the corresponding network structure can be customized for different response formats.

[0028] Specifically, firstly, in response to the health-related questions being in text and image formats, the health-related questions are input into a text encoder and an image encoder to obtain text-encoded information and image-encoded information, respectively. Then, this text-encoded and image-encoded information is input into a multimodal alignment layer to perform modal alignment of text and image features, obtaining preliminary multimodal alignment feature information. Next, the initial multimodal alignment feature information, the aforementioned text-encoded and image-encoded information, are input into an attention-based residual layer to extract key modal feature information from the text-encoded and image-encoded information and add it to the initial multimodal alignment feature information, thus achieving a residual architecture and obtaining multimodal alignment feature information with richer multimodal content. Finally, the multimodal alignment feature information is input into a multimodal feature encoding layer to integrate multimodal features and further compress features, obtaining question-encoded information.

[0029] In some optional implementations of certain embodiments, the above-described multimodal coding model includes: a first multimodal coding layer corresponding to a first response method and a second multimodal coding layer corresponding to a second response method. The first multimodal coding layer corresponding to the first response method is a multimodal coding layer specifically designed to encode the question content described using colloquial language. In practice, the first multimodal coding layer can be a coding layer based on a multi-head attention mechanism. In practice, the first multimodal coding layer can be a Transformer network layer that does not include a decoding layer. The second multimodal coding layer corresponding to the second response method can be a multimodal coding layer specifically designed to encode the question content described using professional knowledge language. In practice, the second multimodal coding layer can be a coding layer based on a multi-head attention mechanism. In practice, the second multimodal coding layer can be a Transformer network layer that does not include a decoding layer.

[0030] It should be noted that during the training of the large question-answering model, the network parameters of the first and second multimodal coding layers are simultaneously trained based on the question-related dataset. The question-related data can be task data related to question-related processing tasks. For example, the processing task could be a question semantic extraction task or a question keyword extraction task. Correspondingly, the question-related data can include: the question and a summary of its semantics, or it can include: the question and a set of question keywords.

[0031] Here, by setting up a first multimodal coding layer and a second multimodal coding layer to encode questions in the corresponding question-answer format, the question-answer style can be preserved, facilitating semantic understanding and question answering using the answer sub-model in the corresponding answer format. In addition, an encoding mapping layer exists between the first and second multimodal coding layers to transform the encoded content of the first multimodal coding layer into the encoded content of the second multimodal coding layer, or vice versa. In practice, the encoding mapping layer can be trained synchronously with the first and second multimodal coding layers. The encoding mapping layer can be multiple concatenated convolutional layers to achieve the mapping transformation between encoded information.

[0032] Optionally, the aforementioned execution entity can input the aforementioned health-related questions into the multimodal coding model included in the pre-trained question-answering large language model to obtain question coding information, including the following steps: The first step, in response to the above question's answer format being the first answer format, is to input the above health-related questions into the above first multimodal coding layer to obtain first question coding information. The first question coding information can be the result of coding the health-related questions based on the coding style corresponding to the first multimodal coding layer.

[0033] The second step involves inputting the aforementioned first question encoding information into the encoding mapping layer between the first and second multimodal coding layers to obtain the second question encoding information. This second question encoding information can be the result of encoding health-related questions based on the encoding style corresponding to the second multimodal coding layer.

[0034] The third step is to determine the first problem coding information and the second problem coding information as the problem coding information.

[0035] In some alternative implementations of certain embodiments, the above-described question-answering large language model is trained through the following steps: The first step is to acquire a medical processing dataset. This dataset includes medical questions and their responses. The medical processing data can be the content of data related to processing medical questions. Medical questions can be questions within the healthcare field. For example, a medical question could be "Can amoxicillin be taken with ibuprofen?". The response content can be the results of answering these medical questions. For example, a response could be "Generally, amoxicillin (antibiotic) and ibuprofen (pain reliever and fever reducer) can be taken simultaneously; there is no significant drug interaction between them." This medical processing dataset serves as the training dataset for training the initial large-scale language model for the initial question. The initial large-scale language model for the initial question can be a model that has not yet finished training.

[0036] It should be noted that the individual medical treatment data in the medical treatment dataset can be obtained through collection and data augmentation.

[0037] The second step is to select target medical treatment data from the aforementioned medical treatment dataset. This target medical treatment data can be data used to train subsequent models at the current time.

[0038] As an example, the aforementioned executing entity may randomly select medical processing data from the aforementioned medical processing dataset as the target medical processing data.

[0039] The third step is to perform the training steps for the target medical processing data: Sub-step 1 involves inputting the medical questions from the target medical processing data into the initial multimodal coding model included in the initial question-answering large language model to obtain the question coding information. The initial multimodal coding model can be a multimodal coding model that has not yet finished training.

[0040] Sub-step 2 involves inputting the aforementioned question encoding information into a multi-question-answer format initial response model to obtain the first initial response content under the first response format and the second initial response content under the second response format. The multi-question-answer format initial response model can be multiple initial response models that have not yet finished training. The first initial response content can be a candidate first response content whose accuracy is uncertain. The second initial response content can be a candidate second response content whose accuracy is uncertain.

[0041] Sub-step 3 involves acquiring a first semantic understanding model trained with general medical knowledge, a second semantic understanding model not trained with medical knowledge, and a third semantic understanding model trained with specialized medical knowledge. General medical knowledge can be medical knowledge stored in a general medical database. In practice, general medical knowledge can be basic medical knowledge from a user's daily life. The determination of basic medical knowledge can be based on the prevalence of diseases and common sense. The first semantic understanding model can be a model that supports answering questions related to general medical knowledge and outputting the degree of understanding and a summary of that understanding. In practice, the first semantic understanding model can be a small language model; that is, it is often a lightweight language model. This first semantic understanding model can be specifically trained for learning general medical knowledge. The second semantic understanding model can be a model that supports the output of semantic understanding levels without being trained with general medical knowledge. For example, the second semantic understanding model can be trained from knowledge datasets from various non-medical fields. The second semantic understanding model can be a large language model or a small language model obtained from other fields. The third semantic understanding model can be a small language model or a large language model specifically trained with medical and health knowledge. The third semantic understanding model here can output the degree of knowledge understanding and the summary of understanding. This third semantic understanding model can be based on the Transformer architecture. Alternatively, it can be obtained through transfer learning from pre-trained large-scale medical models from other medical systems.

[0042] Sub-step 4 involves determining the first content semantic loss information among the first initial response content, the second initial response content, and the corresponding question response content. This first content semantic loss information characterizes the content semantic differences among the first initial response content, the second initial response content, and the corresponding question response content.

[0043] As an example, the distance loss (e.g., mean squared error loss (MSELoss)) embedded in the pre-trained model can be used to determine the first content semantic loss information between the first initial response content, the second initial response content, and the corresponding question response content.

[0044] Sub-step 5 involves inputting the initial response content into the first semantic understanding model to generate a first semantic understanding score and first semantic understanding content. The first semantic understanding score characterizes the degree to which the first semantic understanding model comprehends the initial response content. In practice, the first semantic understanding score can be a value between 0 and 100. A higher score indicates a higher degree of comprehension of the initial response content. The first semantic understanding content can be the key elements of the initial response content as understood by the first semantic understanding model. In practice, the first semantic understanding content can be a summary of the initial response content (i.e., a content outline) or a set of keywords corresponding to the initial response content.

[0045] Sub-step 6 involves inputting the first initial response into the second semantic understanding model to generate a second semantic understanding score and second semantic understanding content. The explanations of the second semantic understanding score and second semantic understanding content will not be elaborated further.

[0046] Sub-step 7 involves inputting the aforementioned third initial response into the aforementioned third semantic understanding model to generate a third semantic understanding score and third semantic understanding content. The explanations of the third semantic understanding score and third semantic understanding content are not elaborated here.

[0047] Sub-step 8 generates semantic understanding loss information corresponding to the first, second, and third semantic understanding scores. This semantic understanding loss information characterizes the differences between the first, second, and third semantic understanding scores and predetermined score thresholds. That is, the greater the difference from the predetermined score threshold, the higher the corresponding semantic understanding loss information. The predetermined score threshold characterizes the accuracy of the output semantic understanding content.

[0048] As an example, the Mean Squared Error (MSE) loss function can be used to generate semantic understanding loss information corresponding to the first, second, and third semantic understanding scores mentioned above.

[0049] Sub-step 9 involves determining the second content semantic loss information for the first semantically understood content, the second semantically understood content, and the third semantically understood content. Here, the second content semantic loss information characterizes the degree of difference in content semantics among the first, second, and third semantically understood content. The generation of the second content semantic loss information is similar to the generation of the first content semantic loss information.

[0050] Sub-step 10: Based on the first content semantic loss information, semantic understanding loss information, and second content semantic loss information, determine whether the initial question-answering large language model has finished training.

[0051] As an example, the semantic loss information of the first content, the semantic understanding loss information, and the semantic loss information of the second content are weighted to obtain weighted loss information. When the loss information sequence corresponding to the weighted loss information stabilizes, the initial training of the question-answering large language model is considered complete. When the loss information sequence corresponding to the weighted loss information does not stabilize, the initial training of the question-answering large language model is considered incomplete.

[0052] Sub-step 11: In response to determining the end of training, the initial question-answering large language model is identified as the question-answering large language model.

[0053] In some optional implementations of certain embodiments, the aforementioned first content semantic loss information includes: first semantic sub-loss information between the first initial response content and the question response content, and second semantic sub-loss information between the second initial response content and the question response content. The first semantic sub-loss information can characterize the degree of content semantic difference between the first initial response content and the question response content. The second semantic sub-loss information can characterize the degree of content semantic difference between the second initial response content and the question response content.

[0054] Optionally, the steps also include: The first step involves generating first training confirmation information for training the response model corresponding to the first response format, in response to the determination that the first semantic sub-loss information is higher than the first loss value. Here, the first loss value can be a loss threshold used to measure whether the response model corresponding to the first response format is accurate and requires training. If the loss value is higher than the first loss value, it can be directly confirmed that the response model corresponding to the first response format needs training. If the loss value is lower than the first loss value, a decision on whether to train the model can be made based on the third and fourth loss values.

[0055] The second step involves generating second training confirmation information for training the response model corresponding to the second response format, in response to the determination that the second semantic sub-loss information is higher than the second loss value. The second loss value can be a loss threshold used to measure whether the response model corresponding to the second response format is accurate and requires training. The first and second loss values ​​can be the same or different. If the loss value is higher than the second loss value, it can be directly confirmed that the response model corresponding to the second response format needs training. If the loss value is lower than the second loss value, a fourth loss value can be used to determine whether to train the model.

[0056] Thirdly, in response to the determination that the semantic loss information of the second content is higher than the third loss value, third training confirmation information is generated for training the response model corresponding to the first response format. The third loss value can be a loss threshold used to measure whether the response model corresponding to the first response format is accurate and requires training. The first, second, and third loss values ​​can be the same or different. If the loss value is higher than the third loss value, it can be directly confirmed that the response model corresponding to the first response format needs training. If the loss value is lower than the third loss value, it can be determined whether to train the model based on the first and fourth loss values.

[0057] Fourth, in response to the determination that the semantic understanding loss information is higher than the fourth loss value, fourth training confirmation information is generated for training the response models corresponding to the first and second response formats. The fourth loss value can be a loss threshold used to measure whether the response models corresponding to the first and second response formats are accurate and require training. The first, second, third, and fourth loss values ​​can be the same or different. When the semantic understanding loss information is higher than the fourth loss information, it can be directly confirmed that both models need training. When it is not higher than the fourth loss information, the determination of which model needs training can be further based on the judgment of the first, second, and third loss values.

[0058] Fifth, based on the first, second, third, and fourth training confirmation information mentioned above, determine the model architecture information for model training. The model architecture information here can be one of the following: a response model in the form of a first response, a response model in the form of a second response, or a response model in the form of both a first and second response.

[0059] As an example, in response to the determination that model training is required for any one of the representations in the first, third, and fourth training confirmation information, a response model in the first response form is trained. In response to the determination that model training is required for any one of the representations in the second and fourth training confirmation information, a response model in the second response form is trained. Based on the above, the model architecture information used for model training can be either the response model in the first response form or the response model in the second response form.

[0060] The sixth step involves training the sub-models corresponding to the aforementioned model architecture information in the initial question-answering large language model, based on the first content semantic loss information, the semantic understanding loss information, and the second content semantic loss information, to obtain the trained question-answering large model. Further details are omitted.

[0061] Step 7: Use the trained question-answering model as the initial question-answering language model, and re-extract the target medical processing data, and continue to perform the above training steps.

[0062] Step 104: Input the above question encoding information into the multi-question-answer format response model included in the above question-answering big language model to generate the first response content under the first response format and the second response content under the second response format.

[0063] In some embodiments, the executing entity may input the question encoding information into the multi-question-answer format response model included in the question-answering large language model to generate a first response content in the first response format and a second response content in the second response format. The first response content may be a response result output to a user who lacks health knowledge. The second response content may be a response result output to a user who possesses health knowledge.

[0064] In some optional implementations of certain embodiments, the execution entity may input the question encoding information into the multi-question-answer format response model included in the question-answering large language model to generate the first response content under the first response format and the second response content under the second response format, including the following steps: The first step involves inputting the encoded information of the first question into the response sub-model corresponding to the first response format in the multi-question-answer format response model to obtain the first response content. The response sub-model corresponding to the first response format is used for content parsing and answering the encoded semantic content of common questions.

[0065] The second step involves inputting the encoded information of the second question into the response sub-model corresponding to the second response format within the aforementioned multi-question-answer format response model, thereby obtaining the second response content. The response sub-model corresponding to the second response format is used for content parsing and question answering of encoded semantic content, including questions requiring professional knowledge.

[0066] Step 105: In response to the above question being answered in the form of the first answer, the content of the first answer is displayed in the corresponding answer area of ​​the health dialogue page.

[0067] In some embodiments, in response to the aforementioned question answer being in the first answer format, the executing entity can display the first answer content in the answer area corresponding to the health dialogue page. This answer area includes a one-click upload control. The one-click upload control can be a control that allows for one-click uploading of the answer process corresponding to a health-related question.

[0068] Step 106: In response to selecting the one-click upload control in the above reply area, the page data corresponding to the first reply content and the page data corresponding to the second reply content are sent to the personalized intelligent agent corresponding to the target user based on the above user identity information.

[0069] In some embodiments, in response to selecting the one-click upload control in the response area, the executing entity can send the page data corresponding to the first response content and the page data corresponding to the second response content to the personalized intelligent agent corresponding to the target user, based on the user identity information. The page data corresponding to the first response content may be the page data of the health dialogue page when the first response content is displayed. The page data corresponding to the second response content may be the page data of the health dialogue page when the second response content is displayed.

[0070] As an example, firstly, based on the aforementioned user identity information, the corresponding agent identity information is determined. For example, agent identity information can characterize the agent's identity. For instance, it may include: agent identifier and agent transmission information. Then, the page data corresponding to the first response content and the page data corresponding to the second response content are sent to the personalized agent corresponding to the aforementioned agent identity information.

[0071] In some optional implementations of certain embodiments, the response area further includes a response format conversion control. This response format conversion control can be a control that switches the displayed response content in the response area to another response format. For example, if the current response area displays a first response format, clicking the response format conversion control can switch the displayed response format from the first response format to a second response format, and also switch the first response content to the second response content.

[0072] Optionally, after step 106, the steps further include: The first step is to switch the first response content displayed in the response area to the second response content in response area in response to the selection of the response format conversion control in response area.

[0073] The second step is to respond to the above question by adopting the second response format, and then display the second response content in the corresponding response area on the above health dialogue page.

[0074] Third, in response to selecting the above-mentioned response format conversion control in the above-mentioned response area, the second response content displayed in the above-mentioned response area is switched to the above-mentioned first response content.

[0075] In some optional implementations of certain embodiments, after step 106, the steps further include: The first step, in response to the detection that the target user is undergoing medical processing within the target hospital system, is to acquire real-time medical processing data for the target user. Medical processing can refer to handling medical matters. For example, medical processing could be the target user going to the hospital for a medical diagnosis or for a follow-up visit. The real-time medical processing data here can be the processing results of the medical matters that occurred to the target user at the hospital at the current time. For example, for a target user going to the hospital for a follow-up visit, the corresponding real-time medical processing data could be real-time follow-up visit data.

[0076] As an example, the aforementioned executing entity can obtain the real-time medical processing data corresponding to the target user from the real-time medical database corresponding to the target hospital system.

[0077] The second step is to detect whether there is a corresponding personalized intelligent agent for the target user.

[0078] As an example, the aforementioned executing entity can query the agent information registration database to see if there is a personalized agent corresponding to the user identifier of the target user.

[0079] Third, in response to the determination that the personalized intelligent agent does not exist, a prompt message is sent to the target user's corresponding terminal, suggesting that the personalized intelligent agent be installed. The prompt message may be a script prompting the target user to install the personalized intelligent agent on their mobile device. The terminal can be the mobile device used by the target user.

[0080] Fourth, in response to the confirmed existence, the aforementioned real-time medical processing data is sent to the personalized intelligent agent, allowing the target user to view the medical processing results on the corresponding intelligent agent page. The intelligent agent page can be the homepage of the personalized intelligent agent. Here, a brief summary of the medical processing results can be displayed in the real-time message scrolling box on the intelligent agent page. After the user clicks on the summary, a pop-up window displaying the complete medical processing results can appear for the user to view.

[0081] In addressing the technical problems mentioned in the background section, and considering the application scenario where users utilize a personalized AI agent, the following technical issues arise: Common personalized AI agents typically only store the target user's private settings. These settings do not interact with the target hospital's system, preventing the hospital from immediately accessing the user's personal information when they visit the physical hospital. To address the specific steps involved in resolving the information isolation issue between the personalized AI agent and the target hospital system, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the personalized intelligent agent described above is an intelligent agent set up based on the health status of the target user. The health status of the target user can be the user's own health status. For example, the personalized intelligent agent can display the target user's current disease status, physical indicators, and medication status.

[0082] Optionally, the aforementioned personalized intelligent agent interacts with the target hospital system through the following steps: The first step is to respond to the target user's selection to bind information with the target hospital system on the smart agent page corresponding to the personalized smart agent, and package the user information corresponding to the target user and the smart agent information corresponding to the personalized smart agent to obtain the smart agent binding package.

[0083] The second step involves sending the aforementioned agent binding package to the binding unit corresponding to the target hospital system for verification of the user information and agent information. The binding unit can be a unit responsible for establishing data communication between the personalized agent corresponding to the user information and the target hospital system. This verification can be conducted manually to confirm whether the personalized agent corresponding to the user is qualified for data communication and binding.

[0084] The third step, in response to approval, involves completing the data binding between the personalized intelligent agent and the target hospital system. Based on the user's personal information received from the target hospital system, the intelligent agent page is rendered and displayed. This page includes controls for disease precautions, follow-up appointments, medical knowledge Q&A, professional Q&A, and one-click emergency handling. The medical knowledge Q&A controls correspond to the general knowledge language model in the target hospital system, while the professional Q&A controls correspond to specific disease Q&A models within the same system. User personal information can be the target user's personal information. In practice, this may include name, ID number, and relevant medical diagnosis records. The disease precautions control displays disease precautions. The follow-up appointment control displays follow-up appointment-related content. The medical knowledge Q&A control handles medical knowledge questions and answers. The professional Q&A control handles professional medical knowledge questions and answers. The one-click emergency handling control provides a one-click alarm for emergencies, allowing the medical system to immediately address the emergency.

[0085] Fourth, in response to selecting the above disease precautions control, the user is redirected to the content of the disease precautions corresponding to the personalized intelligent agent.

[0086] The fifth step involves responding to the selection of the aforementioned medical common sense Q&A control, and then redirecting to the medical common sense Q&A page corresponding to the aforementioned personalized intelligent agent to conduct common sense knowledge Q&A.

[0087] The sixth step involves responding to the selection of the aforementioned professional Q&A control by redirecting the user to the professional Q&A page corresponding to the personalized intelligent agent for general knowledge Q&A.

[0088] Step 7: In response to selecting the above one-click emergency handling control, instruct the personalized intelligent agent to perform various bound emergency handling operations.

[0089] The aforementioned "Steps 1-7" constitute another inventive point of this disclosure, solving the problem of "information isolation between the personalized intelligent agent used by the target user and the target hospital system." Based on this, this disclosure first achieves information binding between the personalized intelligent agent and the target hospital system by binding information between the intelligent agent page and the target hospital system, and by realizing interaction and verification between user information. Then, after confirming the verification and data binding, various event controls that support user use can be rendered on the intelligent agent page, and each event control is specifically bound to the model corresponding to the target hospital system to enable the invocation of the corresponding model resources of the target hospital system to handle relevant personal matters.

[0090] In adopting technical solutions to address the technical problems mentioned in the background, and considering the application scenario—where users face difficulties in accessing offline follow-up appointments at target hospitals—further technical challenges arise: the user is located far from the target hospital, or the user has mobility impairments, requiring assistance from medical personnel to reach the hospital for the follow-up appointment. Based on the specific requirements of this application scenario and the difficulties users face in accessing offline follow-up appointments, we have decided to adopt the following solution: In some optional implementations of certain embodiments, the steps further include: The first step, in response to the selection of the aforementioned follow-up consultation control, determines whether the follow-up consultation method is online or offline, based on the target user's corresponding disease information. The disease information can be the current condition of the target user's illness. An online follow-up consultation can be achieved through a personalized intelligent agent to diagnose the target user's current disease recovery status online. An offline follow-up consultation can instruct the user to visit a hospital for an in-person follow-up examination.

[0091] The second step, in response to the online follow-up consultation method, involves a pop-up camera to capture images or videos of the recovery process for the target user's corresponding disease. This camera can be a device corresponding to the device used by the target user. The camera can capture images showing the surface recovery of the disease. Recovery images or videos can both demonstrate the recovery status of the disease.

[0092] The third step involves invoking the dialogue model corresponding to the professional question-and-answer control deployed by the aforementioned personalized intelligent agent to conduct a follow-up consultation with the target user and obtain the follow-up consultation experience results. In practice, this dialogue model can be the question-and-answer model corresponding to the professional question-and-answer control, used for answering professional medical knowledge questions. Through the dialogue model, the recovery status and experience of the target user's corresponding disease can be understood. The follow-up consultation experience results can include: changes in the disease. The follow-up consultation dialogue can be a dialogue processing regarding the follow-up consultation content. The follow-up consultation experience results can be the feeling results of the recovery status at the time of the follow-up consultation.

[0093] The fourth step is to extract disease-related information about the aforementioned diseases from the consultations or feedback received by the target users through the personalized intelligent agent. This disease-related information can be disease-related consultation or feedback content. In practice, disease-related information may include: disease symptoms, disease descriptions, and disease-related questions.

[0094] Fifth, in response to the severity of the disease corresponding to the target user being the first severity level, based on the restored image or video, the follow-up visit experience, and the disease-related information, a preliminary follow-up visit result is generated using the specific disease question-and-answer model. The preliminary follow-up visit result includes: the follow-up visit thought process, the current severity of the disease, and the follow-up visit conclusion.

[0095] Step 6: In response to the fact that the severity of the disease corresponding to the previous follow-up visit is higher than the first severity level or the current severity level is higher than the first severity level, the preliminary follow-up visit results, the follow-up visit experience results, the restored images or videos, the disease-related information and user information are sent to the target hospital system so that the accuracy of the preliminary follow-up visit results can be determined by the specific disease question-and-answer model and professionals.

[0096] The seventh step is to display the confirmation result sent by the target hospital system on the aforementioned intelligent agent page.

[0097] The aforementioned "step one to step seven" is another inventive point of this disclosure. This disclosure, through the personalized intelligent agent corresponding to the target user and the interaction between the personalized intelligent agent and the target hospital system, can realize the timely acquisition of the target user's follow-up visit recovery status and the remote and effective processing of the follow-up visit recovery status, so that the target user can carry out follow-up visit processing at home.

[0098] The above embodiments of this disclosure have the following beneficial effects: Through the multi-response question-and-answer method of some embodiments of this disclosure, the response content in the corresponding response format can be flexibly output according to the target user's identity information, so that the target user can understand the offline response content in the medical field, and support the transmission of page data to a personalized intelligent agent for information storage. Specifically, the reason why the responses to questions in the relevant medical field are not flexible enough is that the responses to professional questions about health issues often contain a large number of medical professional terms. For people with different levels of medical knowledge, professional personnel need to explain or use other intelligent tools for analysis, which is inflexible and inefficient. Based on this, the multi-response question-and-answer method of some embodiments of this disclosure firstly, in response to the target user inputting a health-related question on the health dialogue page of the target hospital system, obtains the user identity information corresponding to the target user. Here, by obtaining the user identity information, the subsequent response content display method is determined so that the response content that the user can directly understand is displayed on the health dialogue page. Then, based on the user identity information, the default question response format displayed on the health dialogue page can be accurately determined. The question response format is one of the following: a first response format for users without health knowledge, and a second response format for users with health knowledge. Next, the aforementioned health-related questions are input into the multimodal coding model included in the pre-trained question-and-answer large language model, which accurately obtains the question coding information to facilitate the generation of content in different response formats. Then, the question coding information is input into the multi-question-and-answer format response model included in the aforementioned question-and-answer large language model to accurately generate the first response content under the first response format and the second response content under the aforementioned second response format. Secondly, in response to the question response format being the aforementioned first response format, the first response content is displayed in the response area corresponding to the aforementioned health dialogue page, where a one-click upload control is provided. Here, based on the user's identity, response content of different levels of expertise is displayed, allowing the user to more accurately and clearly understand the response results. Finally, in response to selecting the aforementioned one-click upload control in the response area, based on the aforementioned user identity information, the page data corresponding to the aforementioned first response content and the page data corresponding to the aforementioned second response content are sent to the personalized intelligent agent corresponding to the target user for personalized storage of the individual's response content. In summary, by prioritizing the display of response content in the corresponding response format based on user identity information, the system can flexibly output response content in the corresponding response format, so that target users can understand the offline response content in the medical field, and support the transfer of page data to a personalized intelligent agent for information storage.

[0099] Further reference Figure 2As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a question-and-answer device in a multi-response format. These device embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this question-and-answer device in a multi-response format can be specifically applied to various electronic devices.

[0100] like Figure 2 As shown, a question-answering device 200 with multiple response formats includes: an acquisition unit 201, a determination unit 202, a first generation unit 203, a second generation unit 204, a display unit 205, and a sending unit 206. The acquisition unit 201 is configured to acquire user identity information corresponding to the target user in response to a target user inputting a health-related question on a health dialogue page in a target hospital system; the determination unit 202 is configured to determine the default question response format displayed on the health dialogue page based on the user identity information, wherein the question response format is one of the following: a first response format for users without health knowledge, and a second response format for users with health knowledge; the first generation unit 203 is configured to input the health-related question into a pre-trained question-answering large language model including a multimodal coding model to obtain question coding information; the second generation unit 204 is configured to... The encoded information is input into the multi-question-answer format response model included in the above-mentioned question-answering language model to generate the first response content under the first response format and the second response content under the second response format; the display unit 205 is configured to display the first response content in the response area corresponding to the health dialogue page in response to the question response format being the first response format, wherein the response area has a one-click upload control; the sending unit 206 is configured to send the page data corresponding to the first response content and the page data corresponding to the second response content to the personalized intelligent agent corresponding to the target user in response to the selection of the one-click upload control in the response area, based on the user identity information.

[0101] It is understandable that the units described in the question-and-answer device 200 in the multi-response format are related to the reference. Figure 1 The steps in the described method correspond to each other. Therefore, the operations, features, and beneficial effects described above for the method also apply to the question-and-answer device 200 and its constituent units in the multi-response format, and will not be repeated here.

[0102] The following is for reference. Figure 3 It shows a schematic diagram of the structure of an electronic device (e.g., an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0103] like Figure 3 As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0104] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0105] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0106] It should be noted that, in some embodiments of this disclosure, the computer-readable medium described above may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0107] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0108] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs. When the aforementioned one or more programs are executed by the electronic device, the electronic device causes the following actions: In response to a target user inputting a health-related question on a health dialogue page in a target hospital system, the electronic device obtains the user identity information corresponding to the target user; based on the user identity information, it determines the default question response format displayed on the health dialogue page, wherein the question response format is one of the following: a first response format for users without health knowledge, and a second response format for users with health knowledge; it inputs the health-related question into a multimodal coding model included in a pre-trained question-and-answer large language model to obtain question coding information; it inputs the question coding information into a response model including multiple question-and-answer formats in the question-and-answer large language model to generate a first response content under the first response format and a second response content under the second response format; in response to the question response format being the first response format, it displays the first response content in the response area corresponding to the health dialogue page, wherein the response area contains a one-click upload control; in response to selecting the one-click upload control in the response area, it sends the page data corresponding to the first response content and the page data corresponding to the second response content to the personalized intelligent agent corresponding to the target user, based on the user identity information.

[0109] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0110] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0111] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including an acquisition unit, a determination unit, a first generation unit, a second generation unit, a display unit, and a sending unit. The names of these units do not necessarily limit the specific unit; for example, the acquisition unit may also be described as "a unit that acquires user identity information corresponding to the target user in response to the target user inputting health-related questions on a health dialogue page in a target hospital system."

[0112] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0113] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A question-and-answer method with multiple responses, comprising: In response to a target user entering a health-related question on the health dialogue page of the target hospital system, obtain the user identity information corresponding to the target user; Based on the user identity information, the default question and answer format displayed on the health dialogue page is determined. The question and answer format is one of the following: a first answer format for users who do not have health knowledge, and a second answer format for users who have health knowledge. The health-related questions are input into the multimodal coding model included in the pre-trained question-answering large language model to obtain question coding information; The question encoding information is input into the multi-question-answer format response model included in the question-answering big language model to generate the first response content under the first response format and the second response content under the second response format; In response to the question being answered in the first answer format, the first answer content is displayed in the answer area corresponding to the health dialogue page, wherein the answer area contains a one-click upload control; In response to selecting the one-click upload control in the reply area, the page data corresponding to the first reply content and the page data corresponding to the second reply content are sent to the personalized intelligent agent corresponding to the target user based on the user identity information.

2. The method according to claim 1, wherein, The response area also includes a response format conversion control; and the method further includes: In response to selecting the response format conversion control in the response area, the first response content displayed in the response area is switched to the second response content; In response to the question being answered in the second response format, the content of the second response is displayed in the corresponding response area on the health dialogue page. In response to selecting the response format conversion control in the response area, the second response content displayed in the response area is switched to the first response content.

3. The method according to claim 1, wherein, The method further includes: In response to detecting that the target user is undergoing medical treatment in the target hospital system, real-time medical treatment data for the target user is acquired. Detect whether the target user has a corresponding personalized intelligent agent; In response to the determination that it does not exist, a prompt message will be sent to the target user's corresponding terminal to prompt the target user to install the personalized intelligent agent; In response to the confirmation of existence, the real-time medical processing data is sent to the personalized intelligent agent so that the target user can view the medical processing results on the intelligent agent page corresponding to the personalized intelligent agent.

4. The method according to claim 1, wherein, The question-answering large language model is trained through the following steps: Obtain the medical processing dataset, which includes: medical questions and their responses; Select target medical treatment data from the medical treatment dataset; For the target medical processing data, perform the training steps: The medical questions in the target medical processing data are input into the initial multimodal coding model included in the initial question-answering large language model to obtain question coding information; The question coding information is input into the initial response model in a multi-question-answer format to obtain the first initial response content under the first response format and the second initial response content under the second response format. Acquire a first semantic understanding model trained with conventional medical knowledge, a second semantic understanding model not trained with medical knowledge, and a third semantic understanding model trained with professional medical knowledge; Determine the first content semantic loss information among the first initial response content, the second initial response content, and the corresponding question response content; The first initial response content is input into the first semantic understanding model to generate a first semantic understanding score and a first semantic understanding content; The first initial response content is input into the second semantic understanding model to generate a second semantic understanding score and second semantic understanding content; The third initial response is input into the third semantic understanding model to generate a third semantic understanding score and third semantic understanding content; Generate semantic understanding loss information corresponding to the first semantic understanding score, the second semantic understanding score, and the third semantic understanding score; Determine the second content semantic loss information of the first semantic understanding content, the second semantic understanding content, and the third semantic understanding content; Based on the first content semantic loss information, semantic understanding loss information, and second content semantic loss information, determine whether the initial question-answering large language model has finished training. In response to the conclusion of training, the initial question-answering large language model is defined as the question-answering large language model.

5. The method according to claim 4, wherein, The first content semantic loss information includes: first semantic sub-loss information between the first initial response content and the question response content, and second semantic sub-loss information between the second initial response content and the question response content; The method also includes: In response to determining that the first semantic sub-loss information is higher than the first loss value, first training confirmation information is generated for training the response model corresponding to the first response form; In response to determining that the second semantic sub-loss information is higher than the second loss value, a second training confirmation information is generated for training the response model corresponding to the second response form; In response to determining that the semantic loss information of the second content is higher than the third loss value, a third training confirmation information is generated for training the model of the response model corresponding to the first response form; In response to determining that the semantic understanding loss information is higher than the fourth loss value, a fourth training confirmation information is generated for training the response model corresponding to the first response form and the response model corresponding to the second response form. Based on the first training confirmation information, the second training confirmation information, the third training confirmation information, and the fourth training confirmation information, the model architecture information for model training is determined; Based on the first content semantic loss information, semantic understanding loss information, and second content semantic loss information, the sub-models corresponding to the model architecture information in the initial question-answering large language model are trained to obtain the trained question-answering large model. The trained question-answering model is used as the initial question-answering language model, and the target medical processing data is re-extracted to continue the training steps.

6. The method according to claim 1, wherein, The multimodal coding model includes: a first multimodal coding layer corresponding to the first response method and a second multimodal coding layer corresponding to the second response method; and The process of inputting the health-related questions into a pre-trained question-answering large language model, including a multimodal coding model, to obtain question coding information includes: In response to the question answer being in the first answer format, the health-related question is input into the first multimodal coding layer to obtain the first question coding information; The first question encoding information is input into the encoding mapping layer between the first multimodal coding layer and the second multimodal coding layer to obtain the second question encoding information; The first question coding information and the second question coding information are determined as the question coding information; and The step of inputting the question encoding information into the multi-question-answer format response model included in the question-answering large language model to generate a first response content in the first response format and a second response content in the second response format includes: The first question encoding information is input into the response sub-model corresponding to the first response format in the multi-question-answer format response model to obtain the first response content; The second question encoding information is input into the response sub-model corresponding to the second response form in the response model of the multi-question-answer format to obtain the second response content.

7. A question-and-answer device in a multi-response format, comprising: The acquisition unit is configured to acquire user identity information corresponding to the target user in response to the target user entering a health-related question on the health dialogue page of the target hospital system. The determining unit is configured to determine the default question and answer format displayed on the health dialogue page based on the user identity information. The question and answer format is one of the following: a first answer format for users who do not have health knowledge, and a second answer format for users who have health knowledge. The first generation unit is configured to input the health-related questions into a multimodal coding model included in a pre-trained question-answering large language model to obtain question coding information; The second generation unit is configured to input the question encoding information into the multi-question-answer format response model included in the question-answering big language model, so as to generate the first response content under the first response format and the second response content under the second response format; The display unit is configured to display the first response content in the response area corresponding to the health dialogue page in response to the question answer being in the first response format, wherein the response area contains a one-click upload control; The sending unit is configured to, in response to selecting the one-click upload control in the reply area, send the page data corresponding to the first reply content and the page data corresponding to the second reply content to the personalized intelligent agent corresponding to the target user, based on the user identity information.

8. An electronic device, comprising: One or more processors; Storage device, on which one or more programs are stored, When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.

Images