Microscopic experiment teaching question and answer method and device, computer equipment and storage medium

Abstract

The invention provides a microscopic experiment teaching question and answer method. The method comprises the following steps: reading and displaying a target slice; generating a limit prompt word according to the teaching target of the user; inputting the limit cue word into the target large model to limit the question and answer range of the target large model; determining a corresponding target screenshot in response to an annotation selection operation of a user on the target slice; and performing question answering according to the target screenshot and the question of the user by using the target large model. According to the method, the defects that a traditional large model is high in computational power requirement and long in reasoning time are overcome, a user can select different slices at will for learning, the teaching process is optimized, and the efficiency and quality of microscopic experiment teaching are comprehensively improved.

Description

Technical Field

[0001] The present application relates to the field of microscopic experiment technology, and in particular to a microscopic experiment teaching question-answering method, device, computer equipment and storage medium. Background Art

[0002] In the field of medical microscopy experiment teaching, the traditional teaching model relies on the combination of fixed sample slices and textbook graphics and texts, which has significant limitations. With the development of artificial intelligence technology, solutions that use large language models to assist teaching have gradually emerged. However, in the current mainstream solutions, although large model technology can assist in question and answer, it often consumes a lot of computing power to process problems during the reasoning stage, resulting in low learning efficiency, and the interaction with users is also relatively rigid, making it difficult to conduct dynamic interactions for samples that change a lot. Summary of the invention

[0003] The purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defects in the prior art that require high computing power and have poor interactivity with users.

[0004] In a first aspect, the present application provides a microscopic experiment teaching question-answering method, comprising:

[0005] Read and display the target slice;

[0006] Generate restriction prompt words according to the user's teaching objectives;

[0007] Inputting restriction prompt words into the target large model to limit the question and answer scope of the target large model;

[0008] In response to a user's annotation selection operation on a target slice, determining a corresponding target screenshot;

[0009] Use the target big model to conduct Q&A based on the target screenshots and user questions.

[0010] In one embodiment, generating a restriction prompt word according to the user's teaching goal includes:

[0011] Determine the experimental areas according to the selected courses, chapters and subjects;

[0012] Generate restriction cue words based on the experimental domain.

[0013] In one embodiment, the restricted prompt words are generated according to the experimental field, including:

[0014] Generate restriction prompts based on the detection target restrictions and experimental fields selected by the user.

[0015] In one of the embodiments, the target large model is fine-tuned based on a pre-trained large model.

[0016] In one embodiment, the process of fine-tuning the pre-trained large model includes:

[0017] The pre-trained large model is fine-tuned using the question-and-answer dataset; the question-and-answer dataset includes teaching questions and answers corresponding to different types of tissues in multiple organ systems, and the teaching questions and answers include prompts for easy-to-make mistakes corresponding to easy-to-make mistakes annotations.

[0018] In one embodiment, in response to a user's annotation selection operation on a target slice, determining a corresponding target screenshot includes:

[0019] Determine the location of historical annotations based on the annotation records corresponding to the target slice;

[0020] Determine whether the annotation selection operation matches the position of any historical annotation;

[0021] If so, the screenshot corresponding to the historical annotation is used as the target screenshot;

[0022] If not, a target screenshot is obtained according to the frame selection area formed on the target slice by the annotation selection operation, and the annotation record is updated according to this annotation selection operation.

[0023] In one embodiment, the process of using the target macro model to conduct question-answering based on the target screenshot and the user's question also includes:

[0024] Determine the corresponding target knowledge graph according to the experimental field; the target knowledge graph includes preset knowledge points corresponding to the experimental field;

[0025] Before inputting the latest question into the target large model, the latest question is subjected to keyword extraction, and the extracted target keywords are matched with each preset knowledge point in the target knowledge graph to determine the target knowledge point;

[0026] For any target knowledge point, the mastery level of the target knowledge point is updated based on the user's question and answer performance.

[0027] In one embodiment, updating the mastery level of a target knowledge point according to the user's question-answering performance includes:

[0028] Determine the mastery level of the target knowledge point this time according to at least one of the user's answer accuracy rate, number of follow-up questions, and annotation modification records;

[0029] The current mastery level is weightedly summed with the historical mastery levels corresponding to the target knowledge point to update the mastery level of the target knowledge point.

[0030] In one embodiment, the process of using the target macro model to conduct question-answering based on the target screenshot and the user's question also includes:

[0031] If the mastery level of any target knowledge point is lower than the first threshold, the knowledge blind spot corresponding to the target knowledge point is determined;

[0032] Guiding prompt words for each knowledge blind spot are added to the latest question; the guiding prompt words are used to instruct the target large model to guide the user to ask questions about the knowledge blind spot when answering.

[0033] In a second aspect, the present application provides a microscopic experiment teaching question-answering device, comprising:

[0034] A reading module is used to read and display the target slice;

[0035] A restriction prompt word generation module is used to generate restriction prompt words according to the user's teaching objectives;

[0036] A scope restriction module, used for inputting restriction prompt words into the target large model to limit the question and answer scope of the target large model;

[0037] A labeling module, configured to determine a corresponding target screenshot in response to a user's labeling selection operation on a target slice;

[0038] The question-and-answer module is used to use the target large model to conduct question-and-answer based on the target screenshots and the user's questions.

[0039] In a third aspect, the present application provides a computer device comprising one or more processors and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the steps of the microscopic experiment teaching question-and-answer method in any of the above-mentioned embodiments are executed.

[0040] In a fourth aspect, the present application provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the microscopic experiment teaching question-and-answer method in any of the above-mentioned embodiments.

[0041] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0042] Based on the microscopic experiment teaching question and answer method in this embodiment, this set of solutions mainly revolves around microscopic experiment teaching question and answer. First, the target slice is read and converted into a digital image to be displayed to the user. Subsequently, a restriction prompt word is generated according to the user's teaching goal, and it is input into the target large model to limit the scope of question and answer. When the user performs a labeling selection operation on the target slice, the system determines the target screenshot. Finally, the target large model with a limited question and answer scope combines the target screenshot and the user's question to conduct question and answer with the user. On the one hand, this solution uses restriction prompt words and target screenshots to accurately constrain the scope of the large model's answer, solving the defects of high computing power requirements and long reasoning time of traditional large models. On the other hand, it uses the large language model's ability to process multimodal data and generalize learning knowledge, so that users can arbitrarily choose different slices for learning, optimize the teaching process, and comprehensively improve the efficiency and quality of microscopic experiment teaching. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0044] Figure 1 A schematic diagram of the process of the microscopic experiment teaching question-answering method provided in the embodiment of the present application;

[0045] Figure 2 A schematic diagram of a process for generating a restriction prompt word in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of a process for determining a target screenshot in an embodiment of the present application;

[0047] Figure 4 A schematic diagram of a process for determining a user's mastery of a target knowledge point in an embodiment of the present application;

[0048] Figure 5 A schematic diagram of a process for improving user learning effect by indicating a target large model in an embodiment of the present application;

[0049] Figure 6 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0051] This application provides a microscopy experiment teaching question and answer method, please refer to Figure 1 , including steps S102 to S110.

[0052] S102, read and display the target slice.

[0053] It can be understood that the target slice is a processed sample slice in the microscopic experiment teaching scenario. It carries information at the microscopic level. It is usually a sample of biological tissue, cell structure or material microstructure, which is cut into thin slices by a slicer, and then stained, fixed and processed so that its internal structure and characteristics can be clearly observed under a microscope. It is the key material for subsequent teaching analysis. This step is the starting point of the entire microscopic experiment teaching question-and-answer method. Reading the target slice is to convert the physical information of the actual microscopic sample into digital information that can be processed by the system, and displaying the target slice is to present this digital information to the user in a visual way, providing an intuitive observation basis for subsequent teaching activities. Compared with the solution in which only partially fixed slices can be used as teaching materials in traditional teaching, the user can select the target slice in this embodiment, so as to meet different teaching needs.

[0054] S104, generating restriction prompt words according to the user's teaching objectives.

[0055] It can be understood that the teaching goal is the learning outcome or knowledge acquisition direction that the user expects to achieve when conducting microscopic experimental teaching, such as mastering the morphological characteristics of cells in a specific course, chapter, or subject, the components of the microstructure, etc. Restrictive prompt words are generated based on the teaching goals. They are used to constrain the keywords or phrases of the target large model's answer range. They can guide the large model to focus on the problem areas that the user is concerned about and avoid answers that deviate from the topic. This step is the key link in connecting the user's teaching needs with the target large model. By analyzing and understanding the user's teaching goals, key information is extracted and converted into restrictive prompt words. These prompt words, as input information, can enable the target large model to generate answers in a targeted manner in the subsequent question-and-answer process, effectively solving the problem that the large model's answers are broad and lack specificity, and improving the accuracy and effectiveness of questions and answers, which is an important guarantee for improving teaching efficiency and quality.

[0056] S106, inputting the restriction prompt words into the target macro model to limit the question and answer scope of the target macro model.

[0057] It can be understood that the target large model here refers to a large language model that has been trained with a large amount of data and has strong language understanding and generation capabilities. In this solution, the target large model is used to perform knowledge retrieval and answer generation based on the input information, provide users with question answers, and the input accepted by the target large model includes at least two modes: text and image. Before the user starts learning, inputting the restricted prompt words into the target large model can change the attention distribution and knowledge retrieval scope of the large model when generating answers. After receiving the restricted prompt words, the large model will filter and match the knowledge stored in itself based on these prompt words, and only generate answers related to the prompt words, thereby effectively limiting the scope of the large model's questions and answers, avoiding it from generating irrelevant or overly broad answers, improving the accuracy and relevance of the answers, and meeting the needs for accurate knowledge acquisition in teaching scenarios. The target large model here can be completely built and trained by the developer himself. However, in order to improve efficiency and reduce costs, it can also be fine-tuned using an open source pre-trained large language model.

[0058] S108: In response to the user's annotation selection operation on the target slice, determine a corresponding target screenshot.

[0059] It can be understood that the annotation selection operation refers to the behavior of the user marking the area or object of interest on the slice image through an interactive device such as a mouse or touch screen when observing the displayed target slice. In addition, the user can also annotate the structural recognition results of the target screenshot through the annotation selection operation, such as when the user determines that a certain tissue structure is an interlobular bile duct. The user's newly created annotations can form a record, and the previous annotations can be retrieved from the historical records when the same slice is studied later. The target screenshot is the image portion containing the content that the user is concerned about, which is captured from the target slice image based on the user's annotation selection operation. It is an important basis for the subsequent large model to conduct question and answer analysis.

[0060] This step is to refine the user's focus from the overall target slice to a specific local area. Through the user's annotation selection operation, the system can determine the location of the microstructure or feature that the user is interested in, and then extract the corresponding target screenshot to provide accurate data support for the subsequent large model based on specific microscopic areas. Specifically, when the user performs an annotation selection operation on the teaching terminal screen, the system records the coordinate information of the user's operation, and based on this coordinate information, the corresponding area is captured from the digital image of the target slice to generate a target screenshot.

[0061] S110, using the target large model to conduct question and answer based on the target screenshot and the user's question.

[0062] It can be understood that user questions refer to questions about microstructure, characteristics, principles, etc. raised by users to the system based on their observation of target slices and their own learning needs. It can also be the user's type discrimination and abnormality recognition of the tissue structure corresponding to the target screenshot. Question and answer refers to the process in which the target large model performs knowledge retrieval, reasoning and answer generation based on the input target screenshot and user questions, and interacts with the user to answer questions. The target large model combines the microscopic image information contained in the target screenshot and the text information of the user's question to perform multimodal information fusion processing. By extracting and understanding image features and analyzing text semantics, the large model retrieves relevant information in its knowledge system, performs reasoning and logical judgment, and finally generates accurate and targeted answers to solve users' doubts in microscopic experimental teaching and achieve the purpose of teaching assistance. Since the target large model has been restricted by the restricted prompt words before starting the question and answer, the knowledge that needs to be paid attention to when reasoning and text generation is greatly reduced, which can reduce the computing power and time required by the model and improve the accuracy of question answering.

[0063] Based on the microscopic experiment teaching question and answer method in this embodiment, this set of solutions mainly revolves around microscopic experiment teaching question and answer. First, the target slice is read and converted into a digital image to be displayed to the user. Subsequently, a restriction prompt word is generated according to the user's teaching goal, and it is input into the target large model to limit the scope of question and answer. When the user performs a labeling selection operation on the target slice, the system determines the target screenshot. Finally, the target large model with a limited question and answer scope combines the target screenshot and the user's question to conduct question and answer with the user. On the one hand, this solution uses restriction prompt words and target screenshots to accurately constrain the scope of the large model's answer, solving the defects of high computing power requirements and long reasoning time of traditional large models. On the other hand, it uses the large language model's ability to process multimodal data and generalize learning knowledge, so that users can arbitrarily choose different slices for learning, optimize the teaching process, and comprehensively improve the efficiency and quality of microscopic experiment teaching.

[0064] In one embodiment, a restriction prompt word is generated according to the user's teaching goal, see Figure 2 , including steps S202 to S204.

[0065] S202, determine the experimental area based on the selected course, chapter and subject.

[0066] It is understandable that when users start experimental teaching, they can choose one from multiple set courses as the goal of this study. Chapters are smaller knowledge modules divided according to the knowledge system within the course. For example, the anatomy chapter elaborates on the knowledge of human anatomical structure; the material performance chapter of the material science course mainly explains the various performance-related contents of materials. In the medical field, science represents different professional directions and business areas. For example, cardiology is mainly aimed at the diagnosis and treatment of heart-related diseases; neurosurgery focuses on surgical treatment of nervous system diseases. Combining these three choices from coarse to fine can determine the specific direction of this study, and thus obtain the experimental field. For example, in the hematology chapter of the medical course, combined with the needs of the hematology department, the experimental field of blood cell morphology and function research is determined.

[0067] The course defines the macro categories of knowledge, the chapters are refined into specific knowledge modules, and the departments further clarify the application scenarios and professional directions in the medical field. Through the comprehensive analysis of information from these three dimensions, it is possible to accurately locate the specific knowledge category and thus determine the corresponding experimental field. This gradual refinement process from macro to micro provides key basic information for the subsequent generation of accurate restriction prompts, ensuring that the generated restriction prompts are closely aligned with the actual teaching needs of users.

[0068] S204, generating restriction prompt words according to the experimental field.

[0069] It can be understood that after determining the experimental field, based on the professional knowledge system, common research problems and experimental focus in the field, restrictive prompt words are generated that can guide the target large model to accurately answer questions. These prompt words serve as a constraint condition to guide the large model to extract information only from the knowledge related to the experimental field during knowledge retrieval and answer generation, avoiding the generation of broad and irrelevant answers, thereby improving the accuracy and pertinence of the large model answering questions and meeting the demand for accurate knowledge in teaching scenarios. Further, in some embodiments, the restrictive prompt words may also include the detection target restriction selected by the user. The detection target restriction is the condition or range set by the user for a specific detection object in a medical experiment or teaching scenario. For example, if the user is concerned about the detection of a specific disease marker, such as alpha-fetoprotein (AFP) for early screening of liver cancer, then "alpha-fetoprotein detection" is a detection target restriction; or the user focuses on identifying a specific structure in the selected department, such as interlobular bile ducts, interlobular arteries, hepatocytes, etc. in the liver department, which are detection target restrictions.

[0070] In one of the embodiments, the target large model is obtained by fine-tuning the pre-trained large model. The pre-trained large model is a model that is pre-trained on large-scale general data, such as training on massive text data, so that it has strong language understanding and generation capabilities. GPT-4, BERT, etc. are typical pre-trained large models, which have shown excellent performance in multiple tasks of natural language processing. Ordinary pre-trained large models do not focus on the medical field. This application can make the target large model better adapt to specific tasks and needs in the medical field by fine-tuning the target large model.

[0071] In one embodiment, the process of fine-tuning the pre-trained large model includes: fine-tuning the pre-trained large model using a question-and-answer dataset. The question-and-answer dataset includes teaching questions and answers corresponding to different types of tissues in multiple organ systems, and the teaching questions and answers include easy-to-error point prompts corresponding to easy-to-error annotations. It can be understood that it includes teaching questions and answers corresponding to different types of tissues in multiple organ systems, and these questions and answers come from real teaching scenarios, medical research literature, and clinical practice cases. For example, according to the organ system classification, the respiratory system, digestive system, urinary system, etc., taking the respiratory system as an example, the different types of tissues included therein can be trachea, lungs, etc. The teaching questions and answers also include easy-to-error point prompts corresponding to easy-to-error annotations. For example, in the teaching questions and answers for the liver, students often misjudge the interlobular bile duct as an interlobular artery because they lack consideration of the surrounding cytoplasm and nucleus. The easy-to-error point prompts include: Please pay attention to the cytoplasm and nucleus in this structure.

[0072] Specifically, when using LORA technology to fine-tune the pre-trained large model, by inserting a low-rank matrix into the model, only the parameters of these newly added low-rank matrices are trained, while most of the parameters of the pre-trained model remain unchanged. In this way, when fine-tuning using a question-and-answer dataset containing teaching questions and answers corresponding to different types of tissues in multiple organ systems and prompts for easy errors, the model can quickly learn professional knowledge and common easy errors in the medical field, so that after receiving restricted prompt words, the target large model can more accurately retrieve and generate answers from knowledge related to the detection target and experimental field, and can also promptly point out easy-to-error items in the question and answer, improving the user's learning effect.

[0073] In one of the embodiments, in response to the user's annotation selection operation on the target slice, determining the corresponding target screenshot includes steps S302 to S308.

[0074] S302: Determine the location of the historical annotation according to the annotation record corresponding to the target slice.

[0075] It can be understood that the annotation record includes the information record of the user's previous annotation operation on the target slice, that is, the historical annotation, which may include information such as the annotation location, annotation time, and annotation content, and is an important reference data for subsequent operations. The annotation record is a retention of the user's past operations. By querying the annotation record corresponding to the target slice, the location information of the previous annotation can be obtained. This step provides a data basis for the subsequent judgment of whether the current annotation selection operation matches the historical annotation. Specifically, the annotation record can be stored in a database, and each record is indexed by the unique identifier of the target slice. When it is necessary to determine the historical annotation location, the corresponding record is queried from the database through the identifier of the target slice to extract the annotation location information therein. For example, create a "annotation record table" in a relational database, including fields such as "slice ID", "annotation location", and "annotation time", and use the SQL query statement "SELECT annotation location FROM annotation record table WHERE slice ID = [target slice ID]" to obtain the historical annotation location.

[0076] S304: Determine whether the annotation selection operation matches the position of any historical annotation.

[0077] It can be understood that the location of the current annotation selection operation is compared with the historical annotation location to determine whether there is overlap or similarity. If it matches, it means that the user may be paying attention to the area that has been annotated before, and the screenshot corresponding to the historical annotation can be used directly to improve the operation efficiency; if it does not match, a new screenshot operation is required. Specifically, a coordinate matching algorithm can be used to compare the start and end coordinates of the annotation selection operation with the coordinate range of the historical annotation. For example, the intersection of two coordinate ranges is calculated. If the intersection is not empty, it is considered that the annotation selection operation matches the historical annotation position. The Shapely library in Python can be used to calculate the intersection of geometric figures to realize coordinate matching judgment.

[0078] S306: If yes, take the screenshot corresponding to the historical annotation as the target screenshot.

[0079] It can be understood that when the annotation selection operation matches the historical annotation position, the screenshot corresponding to the historical annotation is directly used, which avoids repeated screenshot operations, saves system resources and time costs, and improves the efficiency of users obtaining target screenshots.

[0080] S308: If not, a target screenshot is obtained according to the framed area formed by the annotation selection operation on the target slice, and the annotation record is updated according to this annotation selection operation.

[0081] It can be understood that the framed area is a rectangular or other shaped area delineated by the user on the target slice using the mouse or touch operation through the annotation selection operation, which is used to determine the range of the image to be captured. When the annotation selection operation does not match the historical annotation position, it means that the user is paying attention to a new area and needs to generate a new target screenshot based on the framed area. At the same time, the user's latest annotation will also become a historical annotation and be updated to the annotation record for subsequent query and use. Using image capture technology, according to the coordinates of the user's framed area, the corresponding image part is captured from the digital image of the target slice to generate a target screenshot. Intelligent image segmentation technology can also be introduced. When generating a target screenshot based on the framed area, the object of interest in the framed area is automatically segmented and extracted to remove background interference and improve the quality of the target screenshot.

[0082] In one embodiment, in the process of using the target large model to answer questions based on the target screenshot and the user's question, please refer to Figure 4 , also includes steps S402 to S406.

[0083] S402, determining a corresponding target knowledge graph according to the experimental field. The target knowledge graph includes preset knowledge points corresponding to the experimental field.

[0084] It can be understood that the knowledge graph is a structured knowledge representation form based on a semantic network, which graphically displays the knowledge elements (such as concepts, entities, events, etc.) in the experimental field and their relationships. In this embodiment, the target knowledge graph contains preset knowledge points in the corresponding experimental field, and these knowledge points construct an organic knowledge network in the form of nodes and edges. For example, in the knowledge graph in the field of oncology experiments, knowledge points such as "tumor cells", "gene mutations", and "targeted therapy" are used as nodes, and their associations such as "the relationship between gene mutations in tumor cells and targeted therapy" are used as edges to reflect. Different experimental fields have unique knowledge systems and internal connections. By constructing corresponding target knowledge graphs, complex knowledge can be structured. Determining the target knowledge graph according to the experimental field provides a knowledge framework for accurately matching knowledge points and understanding problems in the subsequent question-and-answer process, so that the system can analyze and process within a specific knowledge range, improving the accuracy and professionalism of questions and answers.

[0085] Specifically, knowledge graph construction tools, such as Neo4j and other graph database platforms, can be used to extract information and semantically annotate a large number of experimental field-related literature, teaching materials, clinical data, etc. to build a target knowledge graph. After the experimental field is determined, the system retrieves the corresponding target knowledge graph from the database storing the knowledge graph according to the pre-set mapping relationship. For example, a unique identifier is assigned to each experimental field in the database, and the target knowledge graph is found from the database through the query instruction match.

[0086] S404, extract keywords from the latest question before inputting it into the target large model, and match the extracted target keywords with each preset knowledge point in the target knowledge graph to determine the target knowledge point.

[0087] It can be understood that the latest question refers to the latest question input by the user in the process of question-answering using the target large model. The target keyword is a word extracted from the latest question that can represent the key information of the question. The target knowledge point is the preset knowledge point in the target knowledge graph that successfully matches the target keyword, and it is the key knowledge to answer the user's question. Through keyword extraction, the core of the user's question can be quickly grasped, and the target keyword can be matched with the preset knowledge point in the target knowledge graph, and the knowledge area related to the question can be located in the knowledge graph. In this way, before the question is input into the target large model, the question is preliminarily located and screened, so that the target large model can determine the specific direction of the user's attention when answering the question. Specifically, the keyword extraction algorithm in natural language processing, such as the TextRank algorithm, can be used to extract the target keyword from the latest question. Then, the target keyword is matched with the node (preset knowledge point) in the target knowledge graph using a string matching algorithm or a semantic similarity calculation method. For example, by calculating the cosine similarity between the target keyword and the name of the knowledge graph node, when the similarity exceeds a certain threshold, it is considered that the match is successful and the target knowledge point is determined.

[0088] S406, for any target knowledge point, update the mastery level of the target knowledge point according to the user's question and answer performance.

[0089] It can be understood that the question-answering performance refers to the performance of the user's answers to the questions raised by the target large model, the tasks issued, etc. in the dialogue triggered by the latest question. The user's question-answering performance can be used to determine the user's familiarity and understanding of the target knowledge point, and the mastery level can be used to quantify it. By continuously updating the mastery level of the target knowledge point, the system can track the user's learning progress and knowledge weaknesses, provide a basis for subsequent personalized teaching and targeted tutoring, and achieve dynamic optimization of the teaching process.

[0090] Specifically, the current mastery of the target knowledge point can be determined based on at least one of the user's answer accuracy, the number of follow-up questions, and the annotation modification record. The answer accuracy refers to the accuracy of the user's answer to the target question, which is obtained by comparing the answer content with the standard answer or authoritative knowledge source. The number of follow-up questions refers to the number of times the user further asks related questions after receiving the answer from the target large model. The annotation modification record is a record of the user's modification of the existing annotations in operations related to the target knowledge point, such as image annotation. For example, in medical image analysis, the modification of the tumor area annotation reflects the change in the user's understanding of the relevant knowledge. The answer accuracy directly reflects the fit between the user's answer and the correct knowledge, and can reflect the user's current cognitive accuracy of the knowledge point; the number of follow-up questions indicates the user's satisfaction with the answer and the further demand for knowledge depth. The more follow-up questions, the less the user's mastery of the current knowledge point; the annotation modification record reflects the dynamic changes in the user's understanding of the knowledge point from the actual operation level. By comprehensively considering these factors, the user's current mastery of the target knowledge point can be more comprehensively and accurately evaluated, providing a reliable basis for subsequent updates to the mastery level. After these evaluation factors are quantified, they can be weighted and summed to obtain the user's mastery of the target knowledge point in the current conversation.

[0091] Since the user's mastery of knowledge is a gradual and dynamic process, the user's past mastery of the target knowledge point is recorded in the historical mastery index. After obtaining the current mastery, by weighting the current mastery and the historical mastery, it is possible to comprehensively consider the changes in the user's mastery of knowledge points at different stages and more accurately describe the user's learning trajectory and knowledge mastery level. The more recent mastery reflects the user's current learning status, and giving it a higher weight can timely reflect the user's learning progress; while the historical mastery reflects the user's learning foundation and knowledge accumulation process. The combination of the two can make the updated mastery more stable and reliable, providing a more scientific basis for personalized teaching and tutoring.

[0092] In one of the embodiments, the process of using the target macro model to conduct question-answering based on the target screenshot and the user's question also includes steps S502 to S504.

[0093] S502: If the mastery level of any target knowledge point is lower than a first threshold, determine the knowledge blind spot corresponding to the target knowledge point.

[0094] It can be understood that the first threshold is a pre-set measurement standard used to determine whether the user's mastery of the target knowledge point meets the basic requirements. When the mastery of the target knowledge point is lower than the first threshold, the user corresponding to the knowledge point has not yet understood or has not thoroughly understood the knowledge area. For example, in the knowledge point of "Structure and Function of Heart Valve", if the user's mastery is low, then the specific opening and closing mechanism of the heart valve, the physiological effects of different valve lesions, and other contents may be knowledge blind spots. After the user completes a question and answer session, the system automatically compares the mastery of the target knowledge point with the first threshold, which can be implemented using conditional judgment statements. When determining to directly use the first threshold to determine the knowledge blind spot, the sub-knowledge points related to the target knowledge point but not mastered by the user can also be extracted from the target knowledge graph as the target knowledge blind spot. For different preset knowledge points, a dynamic threshold setting method can be used to adaptively adjust the first threshold according to factors such as the user's learning ability, learning progress, and difficulty of the experimental field. For example, as the user's learning time increases, the size of the first threshold gradually increases the knowledge mastery requirements for the user.

[0095] S504, adding guiding words for each knowledge blind spot to the latest question. The guiding words are used to instruct the target large model to guide the user to ask questions about the knowledge blind spot when answering.

[0096] It can be understood that the guiding prompt words are keywords or phrases specially designed to guide users to pay attention to knowledge blind spots and ask questions. For example, when the knowledge blind spot is "the relationship between the liver lobule structure and the portal area", the guiding prompt words can be "Do you need to understand the relationship between the liver lobule structure and the portal area" and so on. In addition, further divisions can be made for target knowledge points whose mastery level is lower than the first threshold. For example, for target knowledge points whose mastery level is lower than the first threshold but higher than the second threshold, the first level of detail restrictor can be added to the latest question sentence to instruct the large language model to answer the question in the first mode, and the first mode needs to include basic definitions and text descriptions of misunderstandings. The user has a certain degree of mastery of this target knowledge point, and can instruct the large language model to simplify the answer. For target knowledge points whose mastery level is lower than the second threshold, the second level of detail restrictor can be added to the latest question sentence to instruct the large language model to answer the question in the second mode, and the second mode needs to include basic definitions, graphic descriptions of misunderstandings, and typical error cases. The user has a poor grasp of this target knowledge point, and needs to instruct the large language model to answer in a more detailed mode to help students improve their mastery of the target knowledge point as soon as possible.

[0097] The present application provides a microscopic experiment teaching question-and-answer device, comprising a reading module, a restriction prompt word generation module, a range restriction module, a marking module and a question-and-answer module.

[0098] The reading module is used to read and display the target slice. The restriction prompt word generation module is used to generate restriction prompt words according to the user's teaching objectives. The range restriction module is used to input the restriction prompt words into the target macro model to limit the question and answer scope of the target macro model. The annotation module is used to respond to the user's annotation selection operation on the target slice and determine the corresponding target screenshot. The question and answer module is used to use the target macro model to conduct question and answer based on the target screenshot and the user's question.

[0099] For the specific definition of the microscopic experiment teaching question-and-answer device, please refer to the definition of the microscopic experiment teaching question-and-answer method in the above text, which will not be repeated here. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. It should be noted that the division of modules in the embodiment of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0100] The present application provides a computer device, including one or more processors and a memory, wherein the memory stores computer-readable instructions. When the computer-readable instructions are executed by the one or more processors, the steps of the microscopic experiment teaching question-and-answer method in any of the above-mentioned embodiments are executed.

[0101] Indicatively, if Figure 6 As shown, Figure 6 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. Figure 6 The computer device 600 includes a processing component 602, which further includes one or more processors, and a memory resource represented by a memory 601, for storing instructions executable by the processing component 602, such as an application. The application stored in the memory 601 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 602 is configured to execute instructions to perform the microscopic experiment teaching question-answering method of any of the above embodiments.

[0102] The computer device 600 may further include a power supply component 603 configured to perform power management of the computer device 600, a wired or wireless network interface 604 configured to connect the computer device 600 to a network, and an input / output (I / O) interface 605. The computer device 600 may operate based on an operating system stored in the memory 601, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0103] Those skilled in the art will understand that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0104] The present application provides a storage medium storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the microscopic experiment teaching question-and-answer method in any of the above-mentioned embodiments.

[0105] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0106] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

[0107] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A question-answering method for microscopic experiment teaching, characterized in that: include: Read and display the target slice; Generate restriction prompt words according to the user's teaching objectives; Inputting the restriction prompt words into the target macro model to limit the question and answer scope of the target macro model; In response to a user's annotation selection operation on the target slice, determining a corresponding target screenshot; The target large model is used to conduct question and answer based on the target screenshot and the user's question.

2. The microscopic experiment teaching question-answering method according to claim 1 is characterized in that: The step of generating restriction prompt words according to the user's teaching objectives includes: Determine the experimental areas according to the selected courses, chapters and subjects; The restriction prompt words are generated according to the experimental field.

3. The microscopic experiment teaching question-answering method according to claim 2 is characterized in that: The step of generating the restriction prompt word according to the experimental field includes: The restriction prompt words are generated according to the detection target restriction selected by the user and the experimental field.

4. The microscopic experiment teaching question-answering method according to claim 1 is characterized in that: The target large model is obtained by fine-tuning the pre-trained large model.

5. The microscopic experiment teaching question-answering method according to claim 4 is characterized in that: The process of fine-tuning the pre-trained large model includes: The pre-trained large model is fine-tuned using a question-and-answer dataset; the question-and-answer dataset includes teaching questions and answers corresponding to different types of tissues in multiple organ systems, and the teaching questions and answers include error-prone point prompts corresponding to error-prone annotations.

6. The microscopic experiment teaching question-answering method according to claim 1 is characterized in that: The step of determining a corresponding target screenshot in response to a user's annotation selection operation on the target slice includes: Determine the location of the historical annotation according to the annotation record corresponding to the target slice; Determining whether the annotation selection operation matches the position of any of the historical annotations; If yes, taking the screenshot corresponding to the historical annotation as the target screenshot; If not, the target screenshot is obtained according to the framed area formed on the target slice by the annotation selection operation, and the annotation record is updated according to the current annotation selection operation.

7. The microscopic experiment teaching question-answering method according to claim 2 is characterized in that: In the process of using the target macro model to conduct question-answering according to the target screenshot and the user's question, it also includes: Determine a corresponding target knowledge graph according to the experimental field; the target knowledge graph includes preset knowledge points corresponding to the experimental field; Extract keywords from the latest question before inputting it into the target macro model, and match the extracted target keywords with each of the preset knowledge points in the target knowledge graph to determine the target knowledge points; For any of the target knowledge points, the mastery level of the target knowledge point is updated according to the question-answering performance of the user.

8. The microscopic experiment teaching question-answering method according to claim 7 is characterized in that: The updating of the mastery level of the target knowledge point according to the question-answering performance of the user includes: Determine the mastery level of the target knowledge point at this time according to at least one of the user's answer accuracy rate, the number of follow-up questions, and the annotation modification record; The current mastering degree and each historical mastering degree corresponding to the target knowledge point are weightedly summed to update the mastering degree of the target knowledge point.

9. The microscopic experiment teaching question-answering method according to claim 7, characterized in that: In the process of using the target macro model to conduct question-answering according to the target screenshot and the user's question, it also includes: If the mastery level of any of the target knowledge points is lower than a first threshold, determining the knowledge blind spot corresponding to the target knowledge point; Guiding prompt words for each of the knowledge blind spots are added to the latest question; the guiding prompt words are used to instruct the target large model to guide the user to ask questions about the knowledge blind spots when answering.

10. A microscopic experiment teaching question-answering device, characterized in that: include: A reading module is used to read and display the target slice; A restriction prompt word generation module is used to generate restriction prompt words according to the user's teaching objectives; A scope restriction module, used for inputting the restriction prompt words into the target macro model to limit the question and answer scope of the target macro model; A marking module, configured to determine a corresponding target screenshot in response to a user's marking selection operation on the target slice; The question-and-answer module is used to use the target macro model to conduct question-and-answer based on the target screenshot and the user's questions.

11. A computer device, characterized in that: It includes one or more processors and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the microscopic experiment teaching question-and-answer method as described in any one of claims 1 to 9 are executed.

12. A storage medium, characterized in that: The storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the microscopic experiment teaching question-and-answer method as described in any one of claims 1 to 9.

Images