Enhanced feedback-based medical interactive large model training method and system

Abstract

The invention relates to a medical interactive large model training method based on enhanced feedback, and relates to the field of medical interaction, and the method comprises the steps: building a medical interactive large model based on an input data sequence and an output data sequence; and after one-time medical interaction is completed, determining the number of supplementary training times of the medical interactive large model according to a score value fed back for the latest reply content of the dialogue reply box so as to complete enhanced training of the large model. The invention also relates to a medical interactive large model training system based on enhanced feedback. According to the invention, aiming at the technical problems that the medical interactive large model cannot quantitatively reinforce feedback training and the internal structure is not perfect, the structure of the medical interactive large model is perfected by introducing the deep neural network architecture for dynamically selecting the number of hidden layers and the customized data structure; and meanwhile, the supplementary training frequency of the large model is determined according to the score value fed back for the latest reply content, so that the technical problem is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical interaction, and in particular to a medical interactive large model training method and system based on reinforcement feedback. BACKGROUND

[0002] In recent years, artificial intelligence (AI) technology, especially large models, has made significant progress in the medical field. These models perform well in static medical question and answer tasks, and even approach human experts in some cases. By adopting a large model-based medical interaction method, users can quickly receive professional-like responses, saving medical units labor and time costs, while users are transferred between various medical units, for example, users can complete diagnosis, diagnosis and prescription at home through a large model-based medical interaction service under a charging framework. However, medical diagnosis is not a single static task, but a dynamic and complex process involving multiple rounds of interaction and information collection, so the large model-based medical interaction method needs to be reinforced.

[0003] For example, Chinese invention patent publication CN118335360A proposes an interactive clinical decision support system and method based on large model knowledge enhancement. The present application updates the interaction graph based on the interaction content at the end of each round of interaction with the large model; the embedding representation of the interaction graph is taken as the state, and the inquiry, physical examination or auxiliary examination is taken as the executable action, the diagnosis process is optimized through the reinforcement learning method, and the accuracy and efficiency of the diagnosis are improved; the changes in the interaction graph of the multi-round inquiry are analyzed to generate a reasoning path, the counterfactual state of the interaction graph is generated based on the reasoning path, the importance of each reasoning path is calculated based on the counterfactual state, and the important reasoning path is input into the large model to generate a natural language description, which can improve the quality of knowledge retrieval from the medical knowledge graph in multi-round inquiry, provide more reliable knowledge representation for large model input, and improve the quality and reliability of large model generated content.

[0004] For example, Chinese invention patent publication CN118098530A proposes an intelligent triage method and device based on a medical large language model and electronic equipment. The present application relates to the field of artificial intelligence technology, and in particular to the fields of large models, intelligent medical care and intelligent recommendation. The specific scheme of the present application is: based on the medical large language model, a multi-round interactive dialogue is carried out with the target object to determine the illness information of the target object; the illness information is matched with the medical resources in the medical knowledge base to obtain triage information matched with the target object, which includes the departments and doctors recommended for the target object. The scheme of the present application can improve the accuracy of the triage information, and is helpful to provide personalized medical advice and high-quality medical services for the target object.

[0005] However, none of the above-mentioned technical solutions in the existing technology completes the quantitative reinforcement processing of the medical interactive big model based on the user feedback of the medical interaction, such as the user rating score, resulting in the inability to realize the reinforcement feedback of the medical interactive big model, making it impossible for the medical interactive big model to be dynamically improved and grown according to the user feedback after each medical interaction, affecting the reliability and stability of the responses in subsequent medical interactions. At the same time, the internal structure of the medical interactive big model in the existing technology needs to be further improved. Summary of the Invention

[0006] In order to solve the technical problems in the prior art, the present invention provides a medical interactive large model training method and system based on reinforced feedback. By introducing a deep neural network architecture with dynamic selection of the number of hidden layers, the input content and output content of the customized data structure are designed to complete the improvement of the structure of the medical interactive large model. What is particularly critical is that after completing a medical interaction by executing the dialogue input box and the dialogue reply box of the medical interaction, the number of supplementary training of the medical interactive large model is determined according to the score value feedback of the latest reply content of the dialogue reply box to complete the reinforced training of the medical interactive large model. The determined number of supplementary training times is inversely correlated with the score value, thereby realizing dynamic reinforced training of the medical interactive large model based on the score value, and improving the reliability and stability of the automatic reply of the medical interactive large model.

[0007] According to a first aspect of the present invention, a medical interactive large model training method based on enhanced feedback is provided, the method comprising: Capturing ASCII codes corresponding to respective characters input into a dialogue input box for executing medical interaction, and concatenating the ASCII codes end to end in the order of input to obtain an input data sequence, wherein the dialogue input box has a character quantity limit of a first quantity threshold; capturing ASCII codes corresponding to respective characters of an automatic reply in a dialogue reply box for executing a medical interaction, and concatenating the ASCII codes end to end in the order of input to obtain an output data sequence, wherein the dialogue reply box has a character quantity limit of a second quantity threshold; Searching a preset number of medical keywords from each character input in the dialogue input box as a plurality of input keywords, and searching a preset number of medical keywords from each character automatically replied in the dialogue reply box as a plurality of reply keywords; The medical interactive big model is established by using the first quantity threshold, the second quantity threshold, the preset quantity, the multiple input keywords, the multiple response keywords, and the input data sequence as the input content of the medical interactive big model item by item, and using the output data sequence as the single output content of the medical interactive big model; After a medical interaction is completed by the conversation input box for performing the medical interaction and the conversation reply box for performing the medical interaction, the number of supplementary training times for the medical interactive large model is determined according to the score value of the feedback on the latest reply content of the conversation reply box to complete the reinforcement training of the medical interactive large model, and the determined number of supplementary training times is inversely related to the score value.

[0008] According to a second aspect of the present application, a medical interactive large model training system based on reinforcement feedback is provided, and the system comprises: A first capturing device is configured to capture each part of ASCII code corresponding to each character input by the conversation input box for performing the medical interaction, and concatenate each part of ASCII code in a head-to-tail manner according to the input sequence to obtain an input data sequence, and the conversation input box has a first number threshold of character quantity limit. A second capturing device is configured to capture each part of ASCII code corresponding to each character automatically replied by the conversation reply box for performing the medical interaction, and concatenate each part of ASCII code in a head-to-tail manner according to the input sequence to obtain an output data sequence, and the conversation reply box has a second number threshold of character quantity limit. A character searching device is connected with the first capturing device and the second capturing device respectively, and is configured to search a preset number of medical keywords from each character input by the conversation input box as a plurality of input keywords, and search a preset number of medical keywords from each character automatically replied by the conversation reply box as a plurality of reply keywords. A model establishing device is connected with the first capturing device, the second capturing device and the character searching device respectively, and is configured to take the first number threshold, the second number threshold, the preset number, the plurality of input keywords, the plurality of reply keywords and the input data sequence as the item-by-item input content of the medical interactive large model, take the output data sequence as the single output content of the medical interactive large model, and establish the medical interactive large model. A reinforcement feedback device is connected with the model establishing device, and is configured to determine the number of supplementary training times for the medical interactive large model after a medical interaction is completed by the conversation input box for performing the medical interaction and the conversation reply box for performing the medical interaction, according to the score value of the feedback on the latest reply content of the conversation reply box to complete the reinforcement training of the medical interactive large model, and the determined number of supplementary training times is inversely related to the score value.

[0009] Therefore, the present application has at least the following six main points: Invention point A: a medical interactive large model is established for a dialogue input box and a dialogue reply box for performing a medical interaction to complete automatic reply of the dialogue reply box, and after the dialogue input box and the dialogue reply box for performing the medical interaction complete a medical interaction, a number of supplementary training of the medical interactive large model is determined according to a score value fed back according to latest reply content of the dialogue reply box to complete reinforcement training of the medical interactive large model, the number of supplementary training is inversely related to the score value, thereby realizing dynamic reinforcement training of the medical interactive large model based on the score value, and improving reliability and stability of automatic reply of the medical interactive large model; Invention point B: a medical interactive large model of a customized structure is established, the medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and a mean value of a first number threshold and a second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the mean value of the first number threshold and the second number threshold, thereby completing intelligent conversion of input content of the dialogue input box to reply content of the dialogue reply box, and realizing automatic medical interaction, and saving a large amount of labor cost and time cost; Invention point C: the input content and the output content of the established medical interactive large model adopt a data structure of targeted screening, specifically, the input content of the medical interactive large model is a first number threshold, a second number threshold, a preset number, a plurality of input keywords, a plurality of reply keywords and an input data sequence, and the output content of the medical interactive large model is an output data sequence, thereby completing data structure design of the medical interactive large model, and ensuring intelligent conversion from the input content to the output content; Invention point D: the input data sequence and the output data sequence are specifically obtained as follows: each part of ASCII codes corresponding to each character input by the dialogue input box for performing the medical interaction is captured, and each part of the ASCII codes is connected at the head and tail in the order of input to obtain the input data sequence, the dialogue input box has a character number limit of the first number threshold, the number of ASCII codes in the input data sequence is equal to the first number threshold, and when the number of input characters is less than the first number threshold, one or more ASCII codes representing zero are supplemented at the tail of the input data sequence to make the input data sequence equal to the first number threshold, and each part of ASCII codes corresponding to each character of automatic reply of the dialogue reply box for performing the medical interaction is captured, and each part of the ASCII codes is connected at the head and tail in the order of input to obtain the output data sequence, the dialogue reply box has a character number limit of the second number threshold, the number of ASCII codes in the output data sequence is equal to the second number threshold, and when the number of output characters is less than the second number threshold, one or more ASCII codes representing zero are supplemented at the tail of the output data sequence to make the output data sequence equal to the second number threshold; Inventive point E: The specific steps of obtaining multiple input keywords and multiple reply keywords are as follows: searching a preset number of multiple medical keywords from each character input in the dialogue input box as multiple input keywords, and searching a preset number of multiple medical keywords from each character automatically replied in the dialogue reply box as multiple reply keywords, including: when the number of medical keywords searched from each character input in the dialogue input box exceeds a preset number, using the preset number of multiple medical keywords searched first as multiple input keywords, and when the number of medical keywords searched from each character automatically replied in the dialogue reply box exceeds a preset number, using the preset number of multiple medical keywords searched first as multiple reply keywords; Invention point F: The initial training times of the medical interactive big model are fixed and positively correlated with the average of the first quantity threshold and the second quantity threshold. At the same time, for each training of the medical interactive big model, an output data sequence corresponding to a known past medical interaction is used as the single output content of the medical interactive big model. The first quantity threshold, the second quantity threshold, the preset quantity, multiple input keywords corresponding to a past medical interaction, multiple reply keywords corresponding to a past medical interaction, and the input data sequence corresponding to a past medical interaction are used as the item-by-item input content of the medical interactive big model to complete this training. In addition, a past medical interaction that has not been used in previous training is used in each supplementary training, thereby completing the initial training of the medical interactive big model and the specific design of each supplementary training. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The embodiments of the present invention will be described below with reference to the accompanying drawings, in which: Figure 1 This is a technical flow chart of a medical interactive large model training method and system based on enhanced feedback according to the present invention.

[0011] Figure 2 The present invention is a flowchart of the steps of the medical interactive large model training method based on enhanced feedback according to Example 1 of the present invention.

[0012] Figure 3 This is a flowchart of the steps of the medical interactive large model training method based on enhanced feedback according to Example 2 of the present invention.

[0013] Figure 4 This is a flowchart of the steps of the medical interactive large model training method based on enhanced feedback according to Example 3 of the present invention.

[0014] Figure 5 This is a diagram showing the internal structure of a medical interactive large model training system based on enhanced feedback according to Example 4 of the present invention.

[0015] Figure 6 This is a diagram showing the internal structure of a medical interactive large model training system based on enhanced feedback according to Example 5 of the present invention.

[0016] Figure 7 This is a diagram showing the internal structure of a medical interactive large model training system based on enhanced feedback according to Example 6 of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, a technical flow chart of the medical interactive large model training method and system based on enhanced feedback according to the present invention is given.

[0018] The specific technical process of the present invention is as follows: Technical Process 1: Build a customized medical interactive model for the dialogue input box and dialogue response box to automatically respond to the dialogue response box. like Figure 1 As shown, as the execution subject of the medical interactive large model, the dialogue input box and dialogue response box for executing the medical interaction involve the target patient associated with the user; Specifically, the structural customization of the medical interactive large model is mainly reflected in the following aspects: First: the medical interactive large model is based on a deep neural network architecture, and the deep neural network includes multiple hidden layers, a single input layer, and a single output layer; Second: the number of hidden layers of the deep neural network is proportional to the average of the first number threshold and the second number threshold; Technical process 2: Targeted screening of input and output content for the medical interactive big model; Specifically, the input content and output content of the established medical interactive big model adopt a targeted screening data structure. The input content of the medical interactive big model is a first quantity threshold, a second quantity threshold, a preset quantity, multiple input keywords, multiple response keywords, and an input data sequence. The output content of the medical interactive big model is an output data sequence, thereby completing the data structure design of the medical interactive big model and ensuring intelligent conversion from input content to output content. Further specifically, the input data sequence and the output data sequence are obtained as follows: capturing respective ASCII codes corresponding to respective characters input in the conversation input box for performing the medical interaction, and concatenating the respective ASCII codes in a head-to-tail manner according to the input order to obtain the input data sequence, the conversation input box having a character quantity limit of a first quantity threshold, and when the number of input characters is less than the first quantity threshold, one or more ASCII codes representing zero are supplemented to the tail of the input data sequence to make the number of ASCII codes in the input data sequence equal to the first quantity threshold, and capturing respective ASCII codes corresponding to respective characters automatically replied in the conversation reply box for performing the medical interaction, and concatenating the respective ASCII codes in a head-to-tail manner according to the input order to obtain the output data sequence, the conversation reply box having a character quantity limit of a second quantity threshold, and when the number of output characters is less than the second quantity threshold, one or more ASCII codes representing zero are supplemented to the tail of the output data sequence to make the number of ASCII codes in the output data sequence equal to the second quantity threshold; Further specifically, the plurality of input keywords and the plurality of reply keywords are obtained as follows: searching for a preset number of medical keywords from respective characters input in the conversation input box as the plurality of input keywords, and searching for a preset number of medical keywords from respective characters automatically replied in the conversation reply box as the plurality of reply keywords, including: when the number of medical keywords searched from the respective characters input in the conversation input box exceeds the preset number, taking the first searched preset number of medical keywords as the plurality of input keywords, and when the number of medical keywords searched from the respective characters automatically replied in the conversation reply box exceeds the preset number, taking the first searched preset number of medical keywords as the plurality of reply keywords In this way, the intelligent conversion of the input content of the conversation input box to the reply content of the conversation reply box is completed, and the automated medical interaction is realized, saving a large amount of labor cost and time cost; Technical process three: after completing a medical interaction in the conversation input box and the conversation reply box for performing the medical interaction, determining the number of supplementary training of the medical interactive large model according to the score value fed back to the latest reply content of the conversation reply box to complete the reinforcement training of the medical interactive large model; For example, the medical interactive large model has a fixed initial training number, which is positively correlated with the mean of the first quantity threshold and the second quantity threshold, and on this basis, after completing a medical interaction in the conversation input box and the conversation reply box for performing the medical interaction, the number of supplementary training of the corresponding medical interactive large model can be determined according to the score value fed back by the user, so as to obtain the medical interactive large model for subsequent medical interaction; and for example, each training of the medical interactive large model adopts the output data sequence corresponding to the certain past medical interaction as the single output content of the medical interactive large model, adopts the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords corresponding to the certain past medical interaction, the plurality of reply keywords corresponding to the certain past medical interaction and the input data sequence corresponding to the certain past medical interaction as the item-by-item input content of the medical interactive large model, completes the training, and each time the supplementary training is used for a certain past medical interaction that is not used in the past training; Technical process four: using the medical interactive large model after the supplementary training to execute subsequent medical interactions, that is, to execute automatic reply of the subsequent dialogue reply box; In this way, since the determined number of supplementary training and the score value are inversely associated, the dynamic reinforcement training of the medical interactive large model based on the score value is realized, and the reliability and stability of the automatic reply of the medical interactive large model are improved.

[0019] The key point of the present application is that the dynamic reinforcement training of the medical interactive large model based on the score value, the customized structure design of the medical interactive large model and the targeted screening data structure of the input content and output content of the medical interactive large model.

[0020] In the following, a medical interactive large model training method and system based on reinforcement feedback according to an embodiment of the present application will be described in detail in the form of an embodiment.

[0021] Embodiment 1

[0022] Figure 2 The step flowchart of the medical interactive large model training method based on reinforcement feedback according to Embodiment 1 of the present application is shown.

[0023] As Figure 2 shown, the medical interactive large model training method based on reinforcement feedback includes the following steps: Step S201: capturing each ASCII code corresponding to each character input in the dialogue input box for executing medical interaction, and concatenating the ASCII codes at the head and tail in the order of input to obtain an input data sequence, the dialogue input box having a character quantity limit of the first quantity threshold; For example, the dialogue input box and the dialogue reply box for executing medical interaction are the interactive subjects of the medical interactive large model, after the user inputs the inquiry statement through the dialogue input box, the automatic reply of the reply statement in the dialogue reply box is automatically triggered under the running of the medical interactive large model; Step S202: capture each ASCII code corresponding to each character of the automatic reply of the dialogue reply box performing the medical interaction, and concatenate the ASCII codes in the order of input to obtain an output data sequence, the dialogue reply box having a character number limit of a second number threshold value; Specifically, capturing each ASCII code corresponding to each character of the automatic reply of the dialogue reply box performing the medical interaction, and concatenating the ASCII codes in the order of input to obtain an output data sequence, the dialogue reply box having a character number limit of a second number threshold value includes that the number of bits of the ASCII code corresponding to each character is a fixed value. Step S203: search for a preset number of medical keywords from the characters input from the dialogue input box as a plurality of input keywords, and search for a preset number of medical keywords from the characters of the automatic reply of the dialogue reply box as a plurality of reply keywords; Specifically, searching for a preset number of medical keywords from the characters input from the dialogue input box as a plurality of input keywords, and searching for a preset number of medical keywords from the characters of the automatic reply of the dialogue reply box as a plurality of reply keywords includes that the search processing of the medical keywords is performed in the order of input or reply in the dialogue input box and the dialogue reply box. Step S204: taking the first number threshold value, the second number threshold value, the preset number, the plurality of input keywords, the plurality of reply keywords, and the input data sequence as the item-by-item input content of the medical interactive large model, and taking the output data sequence as the single output content of the medical interactive large model, establishing the medical interactive large model; Specifically, taking the first number threshold value, the second number threshold value, the preset number, the plurality of input keywords, the plurality of reply keywords, and the input data sequence as the item-by-item input content of the medical interactive large model, and taking the output data sequence as the single output content of the medical interactive large model, establishing the medical interactive large model includes that one medical interactive large model corresponds to one interactive subject composed of the dialogue input box and the dialogue reply box performing the medical interaction. Step S205: after completing a medical interaction in the dialogue input box performing the medical interaction and the dialogue reply box performing the medical interaction, determining the number of supplementary training of the medical interactive large model according to the score value of the feedback to the latest reply content of the dialogue reply box to complete the reinforcement training of the medical interactive large model, and the determined number of supplementary training is inversely related to the score value. Specifically, after the conversation input box for performing the medical interaction and the conversation reply box for performing the medical interaction complete a medical interaction, the number of supplementary training of the medical interactive large model is determined according to the score value of the feedback of the latest reply content of the conversation reply box to complete the reinforcement training of the medical interactive large model, and the determined number of supplementary training is inversely related to the score value, including: when the score value is higher than a set score threshold, the reinforcement training of the medical interactive large model can not be performed, that is, the determined number of supplementary training is zero. The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first number threshold and the second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average. For example, the medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first number threshold and the second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average, including: when the average of the first number threshold and the second number threshold is 200, the corresponding number of hidden layers of the deep neural network is 4, when the average of the first number threshold and the second number threshold is 300, the corresponding number of hidden layers of the deep neural network is 6, when the average of the first number threshold and the second number threshold is 400, the corresponding number of hidden layers of the deep neural network is 8, when the average of the first number threshold and the second number threshold is 500, the corresponding number of hidden layers of the deep neural network is 10, and so on. The initial training number of the medical interactive large model is fixed and positively related to the average of the first number threshold and the second number threshold. For example, the initial training number of the medical interactive large model is fixed and positively related to the average of the first number threshold and the second number threshold, including: when the average of the first number threshold and the second number threshold is 200, the initial training number of the medical interactive large model is 500 times, when the average of the first number threshold and the second number threshold is 300, the initial training number of the medical interactive large model is 600 times, when the average of the first number threshold and the second number threshold is 400, the initial training number of the medical interactive large model is 700 times, when the average of the first number threshold and the second number threshold is 500, the initial training number of the medical interactive large model is 800 times, and so on. Wherein, each training of the medical interactive large model adopts the output data sequence corresponding to a certain past medical interaction as the single output content of the medical interactive large model, adopts the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords corresponding to the certain past medical interaction, the plurality of reply keywords corresponding to the certain past medical interaction and the input data sequence corresponding to the certain past medical interaction as the item-by-item input content of the medical interactive large model, completes the training, and each time the supplementary training is used is a certain past medical interaction that is not used in the past training; Wherein, each character corresponding to each ASCII code of the conversation input box input during the medical interaction is captured, and the ASCII codes are connected in sequence to obtain the input data sequence, and the character quantity limit of the conversation input box with the first quantity threshold includes: when the number of input characters is less than the first quantity threshold, one or more ASCII codes representing zero are supplemented to the tail of the input data sequence to make the number of ASCII codes in the input data sequence equal to the first quantity threshold; And wherein, each character corresponding to each ASCII code of the conversation reply box automatic reply during the medical interaction is captured, and the ASCII codes are connected in sequence to obtain the output data sequence, and the character quantity limit of the conversation reply box with the second quantity threshold includes: when the number of output characters is less than the second quantity threshold, one or more ASCII codes representing zero are supplemented to the tail of the output data sequence to make the number of ASCII codes in the output data sequence equal to the second quantity threshold.

[0024] Embodiment 2

[0025] Figure 3 The step flow chart of the medical interactive large model training method based on reinforcement feedback according to embodiment 2 of the present application is shown.

[0026] As Figure 3 shown, after the conversation input box and the conversation reply box during the medical interaction complete a medical interaction, the number of supplementary training of the medical interactive large model is determined according to the score value of the feedback to the latest reply content of the conversation reply box to complete the reinforcement training of the medical interactive large model, and after the determined number of supplementary training is inversely associated with the score value, i.e. after step S205, the medical interactive large model training method based on reinforcement feedback further comprises: Step S206: receiving the medical interactive large model after the reinforcement training is completed, and realizing the model storage of the medical interactive large model after the reinforcement training is completed by storing the model parameters of the medical interactive large model after the reinforcement training is completed; For example, the medical interactive large model after completing the reinforcement training is received, and the model storage of the medical interactive large model after completing the reinforcement training is realized by storing the model parameters of the medical interactive large model after completing the reinforcement training. For example, the MMC storage chip, the TF storage chip or the FLASH flash memory can be selected to receive the medical interactive large model after completing the reinforcement training, and the model storage of the medical interactive large model after completing the reinforcement training is realized by storing the model parameters of the medical interactive large model after completing the reinforcement training.

[0027] Embodiment 3

[0028] Figure 4 The step flow chart of the reinforcement feedback-based medical interactive large model training method according to Embodiment 3 of the present application is shown.

[0029] As shown in Figure 4 After completing a medical interaction in the dialogue input box for performing the medical interaction and the dialogue reply box for performing the medical interaction, the number of supplementary training of the medical interactive large model is determined according to the score value fed back for the latest reply content of the dialogue reply box to complete the reinforcement training of the medical interactive large model. After the determined number of supplementary training is inversely associated with the score value, that is, after step S205, the reinforcement feedback-based medical interactive large model training method further comprises: Step S207: receiving the score value fed back for the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model, and synchronously displaying the score value fed back for the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model. Specifically, the score value fed back for the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model can be received and synchronously displayed by selecting the LED display array or the LCD display array.

[0030] In any of the above embodiments 1-3, optionally, in the reinforcement feedback-based medical interactive large model training method: The searching of the preset number of medical keywords from the characters input in the dialogue input box as the input keywords and the searching of the preset number of medical keywords from the characters automatically replied in the dialogue reply box as the reply keywords comprises: when the number of the medical keywords searched from the characters input in the dialogue input box exceeds the preset number, taking the preset number of the medical keywords searched first as the input keywords, and when the number of the medical keywords searched from the characters automatically replied in the dialogue reply box exceeds the preset number, taking the preset number of the medical keywords searched first as the reply keywords; For example, when the number of the medical keywords searched from the characters input in the dialogue input box exceeds the preset number, taking the preset number of the medical keywords searched first as the input keywords, and when the number of the medical keywords searched from the characters automatically replied in the dialogue reply box exceeds the preset number, taking the preset number of the medical keywords searched first as the reply keywords comprises: the preset number can be selected as 8; The medical keywords comprise disease names, drug names, surgical names, patient symptoms, and physiological information. The medical interactive large model is determined to be retrained for a number of times to complete the reinforcement training of the medical interactive large model according to a score value fed back for the latest reply content of the dialogue reply box after the dialogue input box for performing the medical interaction and the dialogue reply box for performing the medical interaction complete a medical interaction, and the determined number of times of retraining of the medical interactive large model is inversely associated with the score value, which comprises: a numerical mapping function is used to represent a numerical mapping relationship between the determined number of times of retraining of the medical interactive large model and the score value fed back for the latest reply content of the dialogue reply box. For example, the numerical mapping relationship between the determined number of times of retraining of the medical interactive large model and the score value fed back for the latest reply content of the dialogue reply box represented by the numerical mapping function comprises: when the score value fed back for the latest reply content of the dialogue reply box is 60, the number of times of retraining of the medical interactive large model is determined to be 250, when the score value fed back for the latest reply content of the dialogue reply box is 65, the number of times of retraining of the medical interactive large model is determined to be 200, when the score value fed back for the latest reply content of the dialogue reply box is 70, the number of times of retraining of the medical interactive large model is determined to be 150, when the score value fed back for the latest reply content of the dialogue reply box is 80, the number of times of retraining of the medical interactive large model is determined to be 100, when the score value fed back for the latest reply content of the dialogue reply box is 95, the number of times of retraining of the medical interactive large model is determined to be 50, when the score value fed back for the latest reply content of the dialogue reply box is 98 or above, the number of times of retraining of the medical interactive large model is determined to be 0, and so on. The numerical mapping function is used to represent the numerical mapping relationship between the determined number of supplementary training of the medical interactive large model and the score value of the feedback of the latest reply content of the dialogue reply box, which includes: in the numerical mapping function, the score value of the feedback of the latest reply content of the dialogue reply box is the input value, and the number of supplementary training of the medical interactive large model inversely associated with the score value of the feedback of the latest reply content of the dialogue reply box is the output value of the numerical mapping function. The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first number threshold and the second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average, which includes: in the deep neural network architecture, the single input layer, the plurality of hidden layers and the single output layer are connected in sequence. The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first number threshold and the second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average, which includes: in the deep neural network architecture, the single input layer, the plurality of hidden layers and the single output layer are connected in sequence.

[0031] Embodiment 4

[0032] Figure 5 The internal structure diagram of the medical interactive large model training system based on reinforcement feedback according to Embodiment 4 of the present application is shown.

[0033] As Figure 5 shown, the medical interactive large model training system based on reinforcement feedback includes the following components: The first capture device is used to capture each character of the dialogue input box input corresponding to each part of ASCII code, and to concatenate the first and last parts of ASCII code in the order of input to obtain the input data sequence, and the dialogue input box has a character number limit of the first number threshold; For example, the dialogue input box and the dialogue reply box for performing medical interaction are the interactive subjects of the medical interactive large model, and after the user inputs the inquiry statement through the dialogue input box, the automatic reply of the reply statement in the dialogue reply box is automatically triggered under the operation of the medical interactive large model. a second capturing device configured to capture respective ASCII codes corresponding to respective characters of an automatic reply of a dialogue reply box for performing a medical interaction, and concatenate the respective ASCII codes in a sequence of inputting to obtain an output data sequence, the dialogue reply box having a second number threshold of a number of characters; Specifically, the capturing of the respective ASCII codes corresponding to the respective characters of the automatic reply of the dialogue reply box for performing the medical interaction, and the concatenation of the respective ASCII codes in the sequence of inputting to obtain the output data sequence, the dialogue reply box having the second number threshold of the number of characters includes that a bit number of the ASCII code corresponding to each character is a fixed value. a character searching device connected with the first capturing device and the second capturing device respectively, and configured to search a preset number of medical keywords from the respective characters inputted by the dialogue input box as a plurality of input keywords, and search a preset number of medical keywords from the respective characters of the automatic reply of the dialogue reply box as a plurality of reply keywords; Specifically, the searching of the preset number of medical keywords from the respective characters inputted by the dialogue input box as the plurality of input keywords, and the searching of the preset number of medical keywords from the respective characters of the automatic reply of the dialogue reply box as the plurality of reply keywords includes that the searching of the medical keywords is performed in the sequence of inputting or replying by the dialogue input box and the dialogue reply box. a model establishing device connected with the first capturing device, the second capturing device and the character searching device respectively, and configured to take the first number threshold, the second number threshold, the preset number, the plurality of input keywords, the plurality of reply keywords and the input data sequence as itemized input contents of a medical interactive large model, take the output data sequence as a single output content of the medical interactive large model, and establish the medical interactive large model; Specifically, the taking of the first number threshold, the second number threshold, the preset number, the plurality of input keywords, the plurality of reply keywords and the input data sequence as the itemized input contents of the medical interactive large model, the taking of the output data sequence as the single output content of the medical interactive large model, and the establishing of the medical interactive large model includes that one medical interactive large model corresponds to one interactive subject constituted by the dialogue input box and the dialogue reply box for performing the medical interaction. a reinforcement feedback device connected with the model establishing device, and configured to determine a number of supplementary training of the medical interactive large model according to a score value of a feedback to a latest reply content of the dialogue reply box after one medical interaction of the dialogue input box for performing the medical interaction and the dialogue reply box for performing the medical interaction is completed, to complete reinforcement training of the medical interactive large model, and the determined number of the supplementary training is inversely associated with the score value. Specifically, after the conversation input box for performing the medical interaction and the conversation reply box for performing the medical interaction complete a medical interaction, the number of supplementary training of the medical interactive large model is determined according to the score value of the feedback of the latest reply content of the conversation reply box to complete the reinforcement training of the medical interactive large model, and the determined number of supplementary training is inversely related to the score value, including: when the score value is higher than a set score threshold, the reinforcement training of the medical interactive large model can not be performed, that is, the determined number of supplementary training is zero. The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first number threshold and the second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average. For example, the medical interactive large model is based on a deep neural network architecture, and the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first number threshold and the second number threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average, including: when the average of the first number threshold and the second number threshold is 200, the corresponding number of hidden layers of the deep neural network is 4, when the average of the first number threshold and the second number threshold is 300, the corresponding number of hidden layers of the deep neural network is 6, when the average of the first number threshold and the second number threshold is 400, the corresponding number of hidden layers of the deep neural network is 8, when the average of the first number threshold and the second number threshold is 500, the corresponding number of hidden layers of the deep neural network is 10, and so on. The initial training number of the medical interactive large model is fixed and positively related to the average of the first number threshold and the second number threshold. For example, the initial training number of the medical interactive large model is fixed and positively related to the average of the first number threshold and the second number threshold, including: when the average of the first number threshold and the second number threshold is 200, the initial training number of the medical interactive large model is 500 times, when the average of the first number threshold and the second number threshold is 300, the initial training number of the medical interactive large model is 600 times, when the average of the first number threshold and the second number threshold is 400, the initial training number of the medical interactive large model is 700 times, when the average of the first number threshold and the second number threshold is 500, the initial training number of the medical interactive large model is 800 times, and so on. Wherein, each training of the medical interactive large model adopts the output data sequence corresponding to the certain past medical interaction as the single output content of the medical interactive large model, adopts the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords corresponding to the certain past medical interaction, the plurality of reply keywords corresponding to the certain past medical interaction and the input data sequence corresponding to the certain past medical interaction as the item-by-item input content of the medical interactive large model, completes the training, and each time the training is supplemented, the certain past medical interaction that is not used in the past training is used; Wherein, each character corresponding to each ASCII code of the conversation input box input during the medical interaction is captured, and the ASCII codes are connected at the head and tail in the order of input to obtain the input data sequence, and the character quantity limit of the conversation input box with the first quantity threshold includes: when the number of ASCII codes in the input data sequence is equal to the first quantity threshold and the number of input characters is less than the first quantity threshold, one or more ASCII codes representing zero are supplemented at the tail of the input data sequence to make the input data sequence equal to the first quantity threshold; And wherein, each character corresponding to each ASCII code of the conversation reply box automatic reply during the medical interaction is captured, and the ASCII codes are connected at the head and tail in the order of input to obtain the output data sequence, and the character quantity limit of the conversation reply box with the second quantity threshold includes: when the number of ASCII codes in the output data sequence is equal to the second quantity threshold and the number of output characters is less than the second quantity threshold, one or more ASCII codes representing zero are supplemented at the tail of the output data sequence to make the output data sequence equal to the second quantity threshold.

[0034] Embodiment 5

[0035] Figure 6 The internal structure diagram of the medical interactive large model training system based on reinforcement feedback according to Embodiment 5 of the present application is shown.

[0036] As Figure 6 shown, the medical interactive large model training system based on reinforcement feedback further comprises: A parameter storage device connected with the reinforcement feedback device, used to receive the medical interactive large model after completing the reinforcement training, and realize the model storage of the medical interactive large model after completing the reinforcement training by storing the model parameters of the medical interactive large model after completing the reinforcement training; The parameter storage device connected with the reinforcement feedback device is configured to receive the medical interactive large model after reinforcement training and store the medical interactive large model after reinforcement training by storing the model parameters of the medical interactive large model after reinforcement training. The parameter storage device can be implemented by an MMC storage chip, a TF storage chip or a FLASH flash memory.

[0037] Embodiment 6

[0038] Figure 7 An internal structure diagram of the reinforcement feedback-based medical interactive large model training system according to Embodiment 6 of the present application is shown.

[0039] As shown in Figure 7 The reinforcement feedback-based medical interactive large model training system further includes: The synchronization display device connected with the reinforcement feedback device is configured to receive the score value of the feedback of the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model and synchronously display the score value of the feedback of the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model. The synchronization display device connected with the reinforcement feedback device is configured to receive the score value of the feedback of the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model and synchronously display the score value of the feedback of the latest reply content of the dialogue reply box and the determined number of supplementary training of the medical interactive large model. The synchronization display device can be implemented by an LED display array or an LCD display array.

[0040] In any of Embodiments 4-6 above, the reinforcement feedback-based medical interactive large model training system can further include: The searching of the preset number of medical keywords from the characters input in the dialogue input box as input keywords and the searching of the preset number of medical keywords from the characters automatically replied in the dialogue reply box as reply keywords include: when the number of the searched medical keywords from the characters input in the dialogue input box exceeds the preset number, the first searched preset number of medical keywords are taken as the input keywords; and when the number of the searched medical keywords from the characters automatically replied in the dialogue reply box exceeds the preset number, the first searched preset number of medical keywords are taken as the reply keywords. For example, when the number of medical keywords searched in each character input in the dialogue input box exceeds the preset number, the first searched preset number of medical keywords is taken as the input keywords, and when the number of medical keywords searched in each character automatically replied in the dialogue reply box exceeds the preset number, the first searched preset number of medical keywords is taken as the reply keywords, including that the preset number can be selected as 8; Wherein, the medical keywords include disease name, drug name, operation name, patient symptom and physiological information. Wherein, after the dialogue input box for performing medical interaction and the dialogue reply box for performing medical interaction complete a medical interaction, the number of supplementary training of the medical interactive large model is determined according to the score value of the feedback to the latest reply content of the dialogue reply box to complete the reinforcement training of the medical interactive large model, and the determined number of supplementary training of the medical interactive large model is inversely related to the score value of the feedback to the latest reply content of the dialogue reply box, including that a numerical mapping function is used to represent the numerical mapping relationship between the determined number of supplementary training of the medical interactive large model and the score value of the feedback to the latest reply content of the dialogue reply box. For example, the numerical mapping function is used to represent the numerical mapping relationship between the determined number of supplementary training of the medical interactive large model and the score value of the feedback to the latest reply content of the dialogue reply box, including that when the score value of the feedback to the latest reply content of the dialogue reply box is 60, the number of supplementary training of the medical interactive large model is determined to be 250, when the score value of the feedback to the latest reply content of the dialogue reply box is 65, the number of supplementary training of the medical interactive large model is determined to be 200, when the score value of the feedback to the latest reply content of the dialogue reply box is 70, the number of supplementary training of the medical interactive large model is determined to be 150, when the score value of the feedback to the latest reply content of the dialogue reply box is 80, the number of supplementary training of the medical interactive large model is determined to be 100, when the score value of the feedback to the latest reply content of the dialogue reply box is 95, the number of supplementary training of the medical interactive large model is determined to be 50, when the score value of the feedback to the latest reply content of the dialogue reply box is 98 or more, the number of supplementary training of the medical interactive large model is determined to be 0, and so on. Wherein, the numerical mapping function is used to represent the numerical mapping relationship between the determined number of supplementary training of the medical interactive large model and the score value of the feedback to the latest reply content of the dialogue reply box, including that in the numerical mapping function, the score value of the feedback to the latest reply content of the dialogue reply box is the input value, and the number of supplementary training of the medical interactive large model inversely related to the score value of the feedback to the latest reply content of the dialogue reply box is the output value of the numerical mapping function. The medical interactive large model is based on a deep neural network architecture, the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first quantity threshold and the second quantity threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average, which includes: in the deep neural network architecture, the single input layer, the plurality of hidden layers and the single output layer are connected in sequence. And the medical interactive large model is based on a deep neural network architecture, the deep neural network includes a plurality of hidden layers, a single input layer and a single output layer, and the average of the first quantity threshold and the second quantity threshold is calculated, and the number of hidden layers of the deep neural network is directly proportional to the average, which further includes: the information conversion relationship between the number of hidden layers of the deep neural network and the average is expressed by an information conversion formula, in which the average is the input information of the information conversion formula, and the number of hidden layers of the deep neural network directly proportional to the average is the output information of the information conversion formula.

[0041] In addition, in the medical interactive large model training method and system based on reinforcement feedback according to the present application: The first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence are taken as the item-by-item input content of the medical interactive large model, and the output data sequence is taken as the single-item output content of the medical interactive large model, and the medical interactive large model is established, which includes: the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence are respectively converted into binary values, and then input to the medical interactive large model in parallel; For example, after the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence are respectively converted into binary values, and then input to the medical interactive large model in parallel, which includes: using a numerical value conversion device to convert the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence into binary values respectively; For example, after the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence are respectively converted into binary values, and then input to the medical interactive large model in parallel, which further includes: using a parallel control device connected with the numerical value conversion device to complete the parallel input of the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence converted into binary values respectively to the medical interactive large model; And wherein the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords and the input data sequence are taken as item-by-item input contents of the medical interactive large model, and the output data sequence is taken as a single-item output content of the medical interactive large model, and the establishing the medical interactive large model further comprises: taking the output data sequence converted by the binary number as the single-item output content of the medical interactive large model.

[0042] In addition, the terms “first”, “second”, “third” and the like are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the technical features indicated. Therefore, the features defined with “first”, “second” and the like can explicitly or implicitly include one or more of the features. In the description of the embodiments of the present application, the meaning of “plurality” is two or more, unless otherwise explicitly and specifically limited.

[0043] In the description of the present application, the description of the terms “one embodiment”, “certain embodiments”, “illustrative embodiments”, “example”, “specific example” or “some examples” means that the specific features, structures, materials or characteristics described in connection with the embodiments or examples are included in at least one embodiment or example of the present application.

[0044] In the present application, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0045] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A medical interactive large model training method based on enhanced feedback, characterized in that: The method comprises: Capturing ASCII codes corresponding to respective characters input into a dialogue input box for executing medical interaction, and concatenating the ASCII codes end to end in the order of input to obtain an input data sequence, wherein the dialogue input box has a character quantity limit of a first threshold value; capturing ASCII codes corresponding to respective characters of an automatic reply in a dialogue reply box for executing a medical interaction, and concatenating the ASCII codes end to end in the order of input to obtain an output data sequence, wherein the dialogue reply box has a character quantity limit of a second quantity threshold; Searching a preset number of medical keywords from each character input in the dialogue input box as a plurality of input keywords, and searching a preset number of medical keywords from each character automatically replied in the dialogue reply box as a plurality of reply keywords; The medical interactive big model is established by using the first quantity threshold, the second quantity threshold, the preset quantity, the multiple input keywords, the multiple response keywords, and the input data sequence as the input content of the medical interactive big model item by item, and using the output data sequence as the single output content of the medical interactive big model; After a medical interaction is completed in the dialogue input box for executing medical interaction and the dialogue response box for executing medical interaction, the number of supplementary training times for the medical interactive big model is determined based on the score value of the latest response content feedback in the dialogue response box to complete the intensive training of the medical interactive big model. The determined number of supplementary training times is inversely correlated with the score value.

2. The medical interactive large model training method based on enhanced feedback according to claim 1, characterized in that: The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes multiple hidden layers, a single input layer, and a single output layer, and calculates the mean of the first quantity threshold and the second quantity threshold, wherein the number of hidden layers of the deep neural network is proportional to the mean; The initial training times of the medical interactive large model are fixed and positively correlated with the mean of the first quantity threshold and the second quantity threshold; Each training of the medical interactive big model uses an output data sequence corresponding to a known past medical interaction as a single output content of the medical interactive big model, and uses a first quantity threshold, a second quantity threshold, a preset quantity, multiple input keywords corresponding to a past medical interaction, multiple reply keywords corresponding to a past medical interaction, and an input data sequence corresponding to a past medical interaction as the item-by-item input content of the medical interactive big model to complete this training, and in each supplementary training, a past medical interaction that has not been used in previous training is used; The method captures the ASCII codes corresponding to the characters input into the dialogue input box for executing the medical interaction, and connects the ASCII codes end to end in the order of input to obtain an input data sequence. The dialogue input box has a character quantity restriction with a first quantity threshold, including: when the number of ASCII codes in the input data sequence is equal to the first quantity threshold and the number of characters input is less than the first quantity threshold, padding the input data sequence with one or more ASCII codes representing zeros to make the input data sequence equal to the first quantity threshold; Among them, the ASCII codes corresponding to the various characters of the automatic reply of the dialogue response box for performing medical interaction are captured, and the ASCII codes are connected end to end in the order of input to obtain an output data sequence. The dialogue response box has a character quantity limit with a second quantity threshold, including: when the number of ASCII codes in the output data sequence is equal to the second quantity threshold and the number of characters output is less than the second quantity threshold, one or more ASCII codes representing zeros are added to the end of the output data sequence to make the output data sequence equal to the second quantity threshold.

3. The medical interactive large model training method based on enhanced feedback according to claim 2, characterized in that: After a medical interaction is completed in a dialogue input box for executing medical interaction and a dialogue reply box for executing medical interaction, the number of supplementary training times for the medical interaction big model is determined based on a score value of the latest reply content feedback in the dialogue reply box to complete intensive training of the medical interaction big model, and after the determined number of supplementary training times is inversely correlated with the score value, the method further includes: The medical interactive big model after the intensive training is received, and the model storage of the medical interactive big model after the intensive training is realized by storing various model parameters of the medical interactive big model after the intensive training.

4. The medical interactive large model training method based on enhanced feedback according to claim 2, characterized in that: After a medical interaction is completed in a dialogue input box for executing medical interaction and a dialogue reply box for executing medical interaction, the number of supplementary training times for the medical interaction big model is determined based on a score value of the latest reply content feedback in the dialogue reply box to complete intensive training of the medical interaction big model, and after the determined number of supplementary training times is inversely correlated with the score value, the method further includes: Receive the scoring value for the latest reply content feedback in the dialogue reply box and the determined number of supplementary training times for the medical interactive large model, and synchronously display the scoring value for the latest reply content feedback in the dialogue reply box and the determined number of supplementary training times for the medical interactive large model.

5. The medical interactive large model training method based on enhanced feedback according to any one of claims 2 to 4, characterized in that: Searching a preset number of multiple medical keywords from each character input in the dialogue input box as multiple input keywords, and searching a preset number of multiple medical keywords from each character automatically replied in the dialogue reply box as multiple reply keywords includes: when the number of medical keywords searched from each character input in the dialogue input box exceeds a preset number, using the preset number of multiple medical keywords searched first as the multiple input keywords, and when the number of medical keywords searched from each character automatically replied in the dialogue reply box exceeds a preset number, using the preset number of multiple medical keywords searched first as the multiple reply keywords; Among them, medical keywords include disease names, drug names, surgery names, patient symptoms, and physiological information; wherein, after a medical interaction is completed in a dialogue input box for executing medical interaction and a dialogue reply box for executing medical interaction, the number of supplementary training times for the medical interactive large model is determined according to the score value of the latest reply content feedback of the dialogue reply box to complete the intensive training of the medical interactive large model, and the determined number of supplementary training times and the score value are reversely associated, including: using a numerical mapping function to represent the numerical mapping relationship of the reverse association between the determined number of supplementary training times for the medical interactive large model and the score value of the latest reply content feedback of the dialogue reply box; The numerical mapping relationship between the determined number of supplementary training times for the medical interactive large model and the reverse correlation between the score value of the latest reply content feedback for the dialogue reply box and the score value of the latest reply content feedback for the dialogue reply box is represented by a numerical mapping function, including: in the numerical mapping function, the score value of the latest reply content feedback for the dialogue reply box is an input value, and the number of supplementary training times for the medical interactive large model that is reversely correlated with the score value of the latest reply content feedback for the dialogue reply box is an output value of the numerical mapping function; The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes multiple hidden layers, a single input layer, and a single output layer, and calculating the mean of the first quantity threshold and the second quantity threshold, and the number of hidden layers of the deep neural network is proportional to the mean includes: in the deep neural network architecture, the single input layer, the multiple hidden layers, and the single output layer are connected in sequence; Among them, the medical interactive large model is based on a deep neural network architecture and the deep neural network includes multiple hidden layers, a single input layer and a single output layer, and calculates the mean of the first quantity threshold and the second quantity threshold. The number of hidden layers of the deep neural network is proportional to the mean. It also includes: using an information conversion formula to express the information conversion relationship that the number of hidden layers of the deep neural network is proportional to the mean. In the information conversion formula, the mean is the input information of the information conversion formula, and the number of hidden layers of the deep neural network proportional to the mean is the output information of the information conversion formula.

6. A medical interactive large model training system based on enhanced feedback, characterized in that: The system comprises: a first capturing device for capturing respective ASCII codes corresponding to respective characters input into a dialogue input box for performing medical interaction, and concatenating the respective ASCII codes end to end in the order of input to obtain an input data sequence, wherein the dialogue input box has a character quantity limit of a first threshold value; a second capturing device for capturing respective ASCII codes corresponding to respective characters of an automatic reply in a dialogue reply box for executing a medical interaction, and concatenating the respective ASCII codes end to end in the order of input to obtain an output data sequence, wherein the dialogue reply box has a character quantity limit of a second quantity threshold; a character search device, connected to the first capture device and the second capture device, respectively, for searching a preset number of medical keywords from each character input in the dialogue input box as a plurality of input keywords, and for searching a preset number of medical keywords from each character automatically replied in the dialogue reply box as a plurality of reply keywords; a model building device, connected to the first capture device, the second capture device, and the character search device, respectively, for using the first quantity threshold, the second quantity threshold, the preset quantity, the plurality of input keywords, the plurality of reply keywords, and the input data sequence as item-by-item input content of the medical interactive large model, and using the output data sequence as the single item output content of the medical interactive large model, to build the medical interactive large model; An enhanced feedback device is connected to a model device and is used to determine the number of supplementary training times for the medical interactive large model based on the score value of the latest response content feedback of the dialogue response box after a medical interaction is completed in the dialogue input box for executing the medical interaction and the dialogue response box for executing the medical interaction, so as to complete the enhanced training of the medical interactive large model. The determined number of supplementary training times is inversely correlated with the score value.

7. The medical interactive large model training system based on enhanced feedback according to claim 6, characterized in that: The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes multiple hidden layers, a single input layer, and a single output layer, and calculates the mean of the first quantity threshold and the second quantity threshold, wherein the number of hidden layers of the deep neural network is proportional to the mean; The initial training times of the medical interactive large model are fixed and positively correlated with the mean of the first quantity threshold and the second quantity threshold; Each training of the medical interactive big model uses an output data sequence corresponding to a known past medical interaction as a single output content of the medical interactive big model, and uses a first quantity threshold, a second quantity threshold, a preset quantity, multiple input keywords corresponding to a past medical interaction, multiple reply keywords corresponding to a past medical interaction, and an input data sequence corresponding to a past medical interaction as the item-by-item input content of the medical interactive big model to complete this training, and in each supplementary training, a past medical interaction that has not been used in previous training is used; The method captures the ASCII codes corresponding to the characters input into the dialogue input box for executing the medical interaction, and connects the ASCII codes end to end in the order of input to obtain an input data sequence. The dialogue input box has a character quantity restriction with a first quantity threshold, including: when the number of ASCII codes in the input data sequence is equal to the first quantity threshold and the number of characters input is less than the first quantity threshold, padding the input data sequence with one or more ASCII codes representing zeros to make the input data sequence equal to the first quantity threshold; Among them, the ASCII codes corresponding to the various characters of the automatic reply of the dialogue response box for performing medical interaction are captured, and the ASCII codes are connected end to end in the order of input to obtain an output data sequence. The dialogue response box has a character quantity limit with a second quantity threshold, including: when the number of ASCII codes in the output data sequence is equal to the second quantity threshold and the number of characters output is less than the second quantity threshold, one or more ASCII codes representing zeros are added to the end of the output data sequence to make the output data sequence equal to the second quantity threshold.

8. The medical interactive large model training system based on enhanced feedback according to claim 7, characterized in that: The system further comprises: The parameter storage device is connected to the reinforcement feedback device, and is used to receive the medical interactive large model after the reinforcement training is completed, and realize the model storage of the medical interactive large model after the reinforcement training by storing various model parameters of the medical interactive large model after the reinforcement training.

9. The medical interactive large model training system based on enhanced feedback according to claim 7, characterized in that: The system further comprises: A synchronous display device is connected to the enhanced feedback device, and is used to receive the scoring value of the latest response content feedback of the dialogue response box and the determined number of supplementary training times for the medical interactive large model, and synchronously display the scoring value of the latest response content feedback of the dialogue response box and the determined number of supplementary training times for the medical interactive large model.

10. The medical interactive large model training system based on enhanced feedback according to any one of claims 7 to 9, characterized in that: Searching a preset number of multiple medical keywords from each character input in the dialogue input box as multiple input keywords, and searching a preset number of multiple medical keywords from each character automatically replied in the dialogue reply box as multiple reply keywords includes: when the number of medical keywords searched from each character input in the dialogue input box exceeds a preset number, using the preset number of multiple medical keywords searched first as the multiple input keywords, and when the number of medical keywords searched from each character automatically replied in the dialogue reply box exceeds a preset number, using the preset number of multiple medical keywords searched first as the multiple reply keywords; Among them, medical keywords include disease names, drug names, surgery names, patient symptoms, and physiological information; wherein, after a medical interaction is completed in a dialogue input box for executing medical interaction and a dialogue reply box for executing medical interaction, the number of supplementary training times for the medical interactive large model is determined according to the score value of the latest reply content feedback of the dialogue reply box to complete the intensive training of the medical interactive large model, and the determined number of supplementary training times and the score value are reversely associated, including: using a numerical mapping function to represent the numerical mapping relationship of the reverse association between the determined number of supplementary training times for the medical interactive large model and the score value of the latest reply content feedback of the dialogue reply box; The numerical mapping relationship between the determined number of supplementary training times for the medical interactive large model and the reverse correlation between the score value of the latest reply content feedback for the dialogue reply box and the score value of the latest reply content feedback for the dialogue reply box is represented by a numerical mapping function, including: in the numerical mapping function, the score value of the latest reply content feedback for the dialogue reply box is an input value, and the number of supplementary training times for the medical interactive large model that is reversely correlated with the score value of the latest reply content feedback for the dialogue reply box is an output value of the numerical mapping function; The medical interactive large model is based on a deep neural network architecture, and the deep neural network includes multiple hidden layers, a single input layer, and a single output layer, and calculating the mean of the first quantity threshold and the second quantity threshold, and the number of hidden layers of the deep neural network is proportional to the mean includes: in the deep neural network architecture, the single input layer, the multiple hidden layers, and the single output layer are connected in sequence; Among them, the medical interactive large model is based on a deep neural network architecture and the deep neural network includes multiple hidden layers, a single input layer and a single output layer, and calculates the mean of the first quantity threshold and the second quantity threshold. The number of hidden layers of the deep neural network is proportional to the mean. It also includes: using an information conversion formula to express the information conversion relationship that the number of hidden layers of the deep neural network is proportional to the mean. In the information conversion formula, the mean is the input information of the information conversion formula, and the number of hidden layers of the deep neural network proportional to the mean is the output information of the information conversion formula.

Images