Medical training content generation and pushing method based on knowledge graph

Abstract

The application discloses a medical training content generation and pushing method based on a knowledge graph, relates to the technical field of medical training content generation, and comprises the following steps: establishing a knowledge graph used for medical training, collecting explicit feature data and implicit feature data of trainees, performing trainee similarity analysis, screening similar trainees of the trainees, obtaining trainee similarity data, performing knowledge blind area analysis according to the implicit feature data of the trainees, obtaining the knowledge blind area of the trainees, obtaining trainee knowledge blind area data, producing personalized training content for each trainee based on the trainee knowledge blind area data and the trainee similarity data, and performing pushing; the application is used for solving the problem that the existing medical training content generation technology cannot analyze the potential knowledge blind area of training personnel according to the relevant data of the training personnel in a training system and cannot customize the training content of the training personnel when generating training content based on a knowledge graph.

Description

Technical Field

[0001] This invention relates to the field of medical training content generation technology, specifically a method for generating and pushing medical training content based on knowledge graphs. Background Technology

[0002] Medical training content generation technology refers to a technology system that automatically or semi-automatically generates diverse training content based on an authoritative medical knowledge system, with the core objective of meeting the needs of medical professionals in medical education and vocational training scenarios, including knowledge learning, skills practice, clinical thinking cultivation, and assessment.

[0003] Existing medical training content generation technologies, when generating training content based on knowledge graphs, often produce uniform and generalized training content for all trainees according to fixed training templates or objectives. However, medical trainees vary greatly in their knowledge base, skill level, and professional context. To cover all possible knowledge points, uniform and generalized training content often piles up a large amount of information that trainees have already mastered or that is irrelevant to them. It fails to consider the trainees' cognitive level and knowledge gaps, leading to comprehension difficulties due to a lack of prior knowledge. Furthermore, for trainees who have already mastered the relevant knowledge, if the training content remains focused on their existing knowledge and does not address their knowledge gaps, they will be unable to fill these gaps through training. Conversely, if trainees' knowledge gaps are identified through their answers to questions, and the generated training content is then tailored accordingly... While adjustments can compensate for trainees' knowledge gaps to some extent, the limited number of questions and the limited knowledge points covered mean that these methods can only identify knowledge gaps within the covered knowledge points, not potential knowledge gaps in areas not covered by the questions. For example, patent application CN112102676A describes a method and apparatus for generating training content. This approach adjusts the training content based on user responses, which, while addressing some knowledge gaps, cannot target potential knowledge gaps. Therefore, existing medical training content generation technologies, when generating training content based on knowledge graphs, cannot analyze trainees' potential knowledge gaps based on relevant data from the training system, thus failing to personalize the training content for individual trainees. Summary of the Invention

[0004] This invention aims to at least partially address one of the technical problems in existing technologies. It involves establishing a knowledge graph for medical training and collecting explicit and implicit characteristic data of trainees. Trainee similarity analysis is performed to identify similar trainees, yielding trainee similarity data. Based on the trainees' implicit characteristic data, knowledge blind spot analysis is conducted to identify their knowledge blind spots, yielding trainee knowledge blind spot data. Based on the trainee knowledge blind spot data and trainee similarity data, personalized training content is generated for each trainee and pushed to them. This addresses the problem that existing medical training content generation technologies, when generating training content based on knowledge graphs, cannot analyze trainees' potential knowledge blind spots based on relevant data within the training system, thus failing to personalize training content for individual trainees.

[0005] To achieve the above objectives, this application provides a method for generating and pushing medical training content based on knowledge graphs, including the following steps:

[0006] Establish a knowledge graph for medical training and collect explicit and implicit characteristic data of trainees;

[0007] Based on the explicit and implicit characteristic data of the students, a student similarity analysis is performed to filter out similar students and obtain student similarity data.

[0008] Based on the implicit characteristic data of the students, we can analyze the knowledge blind spots of the students and obtain the knowledge blind spot data of the students.

[0009] Based on data on trainees' knowledge gaps and similarity data, personalized training content is generated for each trainee and then pushed to them.

[0010] Furthermore, establishing a knowledge graph for medical training and collecting explicit and implicit characteristic data of trainees includes the following sub-steps:

[0011] The system used for medical training is denoted as the medical training system; a corresponding knowledge graph is established based on the training content of medical training and denoted as the training knowledge graph; the content and process of medical training are divided into multiple chapters, which are denoted as training chapter 1 to training chapter n1 in sequence, and any training chapter is denoted as training chapter i, where n1 is the total number of training chapters, and i∈[1,n].

[0012] Trainees who have completed medical training will be recorded as past trainees, and any past trainee will be recorded as the first past trainee; trainees who are currently undergoing medical training will be recorded as current trainees, and any current trainee will be recorded as the first current trainee.

[0013] Furthermore, establishing a knowledge graph for medical training and collecting explicit and implicit characteristic data of trainees includes the following sub-steps:

[0014] For the first trainee in the current period, based on the training chapter i that has been completed, collect the correct answer rate of the first trainee in the current period in the corresponding training chapter i; and collect the learning time of the first trainee in the current period in training chapter i to obtain the explicit characteristics of the first trainee in training chapter i.

[0015] When the trainees in the first training session are training in training chapter i, collect the keywords related to medical training that the trainees in the first training session search in the medical training system, record them as search keywords, and obtain the number of searches for search keywords; and obtain the page dwell time of the trainees in the first training session on the relevant pages after the search, record it as the search time for search keywords. After completion, the implicit characteristics of the trainees in the first training session in training chapter i are obtained.

[0016] Repeatedly collect the explicit and implicit features of the training chapters 1 to i completed by the first current trainee, and record them as the training feature data of the first current trainee; and collect the explicit and implicit features of the training chapters 1 to i of the first previous trainees, and record them as the training feature data of the first previous trainees; and repeatedly collect the training feature data of all current trainees and previous trainees.

[0017] Furthermore, based on the explicit and implicit characteristic data of the trainees, a trainee similarity analysis is performed to screen similar trainees and obtain trainee similarity data, including the following sub-steps:

[0018] Based on the training characteristic data of the first batch of trainees, the correct answer rate and learning time of the training chapters 1 to i completed by the first batch of trainees are arranged according to the order of the corresponding training chapters, and recorded in order as the first correct answer rate sequence and the first time sequence.

[0019] Calculate the ratio of the data at the corresponding position of the first accuracy sequence to the data at the first time sequence, and record it as the efficiency index. This will give you the learning efficiency sequence of the first current student and the learning efficiency sequence of the first previous student.

[0020] Let any efficiency index in the learning efficiency sequence of the first current student be denoted as DAj, where j represents the position number, ji∈[1,i], and let the efficiency index of the same position number in the learning efficiency sequence of the first previous students be denoted as WAj.

[0021] Calculate the efficiency deviation rate PAj between DAj and WAj, where PAj = 1 - |(DAj - WAj) / max(DAj, WAj)|; Repeatedly calculate the efficiency deviation rate of the efficiency index of all the same position in the learning efficiency sequence of the first current student and the first previous student, and arrange them in ascending order of position number, and record them as the efficiency deviation sequence.

[0022] For PAj, let QAj be the weighted deviation rate corresponding to PAj, where QAj = [j / (i*(i+1) / 2)]*PAj. Repeatedly obtain the weighted deviation rate corresponding to all efficiency deviation rates in the efficiency deviation sequence and calculate the average value, which is denoted as the first similarity index between the first current student and the first previous student.

[0023] Calculate the Pearson correlation coefficient between the learning efficiency sequence of the first current student and the learning efficiency sequence of the first previous student, and denote it as the second similarity index between the first current student and the first previous student.

[0024] Furthermore, the process of performing student similarity analysis based on students' explicit and implicit characteristic data, and filtering out similar students to obtain student similarity data, also includes the following sub-steps:

[0025] Repeatedly obtain the first similarity index and the second similarity index between the first current student and all previous students, and arrange all the first similarity indices in descending order and all the second similarity indices in ascending order. Record them as the first similarity sequence and the second similarity sequence respectively.

[0026] Obtain the position numbers of the first similarity index and the second similarity index of the first previous student in the corresponding first similarity sequence and second similarity sequence, respectively, and sum them up to record the similarity score of the first current student and the first previous student.

[0027] Repeatedly obtain the similarity scores of the first current student and all previous students, and arrange them in descending order to form a similarity score sequence. Record the previous students corresponding to the first k1% of the similarity score sequence as the first similar students of the first current student, where k1% is a set percentage.

[0028] Furthermore, the process of performing student similarity analysis based on students' explicit and implicit characteristic data, and filtering out similar students to obtain student similarity data, also includes the following sub-steps:

[0029] Obtain all the query keywords of the current trainees in training chapters 1 to i, and record them as the keyword set. Record the query keywords in the keyword set as keyword 1 to keyword n2, where n2 represents the total number of query keywords in the keyword set. Record any keyword as keyword m, where m represents the sequence number.

[0030] For the first batch of trainees, based on the implicit feature data of training chapters 1 to i of the first batch of trainees, the total number of queries corresponding to keyword m is counted and recorded as the total number of queries for keyword m corresponding to the first batch of trainees CXm.

[0031] Repeatedly obtain the total number of queries for all query keywords in the keyword set corresponding to the first student in the current period, and establish the query feature vector TM corresponding to the first student in the current period, where TM={CX1, CX2, ..., CXn2};

[0032] Repeatedly obtain the query feature vector of the first similar student, and calculate the cosine similarity between the query feature vector of the first current student and the query feature vector of each first similar student. Arrange them in descending order and denote them as the cosine similarity sequence. The first similar students corresponding to the first k2% of the cosine similarity sequence and greater than k3 are denoteed as the second similar students of the first current student, where k2% is the set proportion and k3 is the set threshold.

[0033] Furthermore, based on the trainees' implicit characteristic data, a knowledge blind spot analysis is performed to obtain the trainees' knowledge blind spots. The process of obtaining trainees' knowledge blind spot data includes the following sub-steps:

[0034] For any keyword m in the keyword set, based on the implicit feature data of the first current student and the corresponding second similar student, obtain the total number of queries for keyword m in training chapter j, which is recorded as the total number of queries for keyword m in training chapter j. Repeat this process to obtain the total number of queries for keyword m from training chapter 1 to training chapter n1, and sort them in the corresponding order, which is recorded as the initial total number sequence for keyword m.

[0035] The first query count greater than k4 in the initial total count sequence of keyword m is taken as the starting point. The total number of queries from the starting point to the next point is denoted as the query total count sequence of keyword m, where k4 is the set threshold.

[0036] Based on the query count sequence of keyword m, the total number of queries in the query count sequence is denoted as V0, and any query count is denoted as CRv, where v represents the position number. The weighted total QRv corresponding to CRv is calculated, where QRv = [v / (V0 * (V0 + 1) / 2)] * CRv. The weighted total corresponding to all query counts in the query count sequence is obtained repeatedly and summed, and denoted as the representative total of keyword m.

[0037] Repeatedly obtain the representative total number of all keywords in the keyword set, and arrange them in descending order, denoted as the representative total number sequence. Obtain the k5th and k6th percentiles of the representative total number, and denote them as the first threshold and second threshold respectively, where k5 and k6 are the set percentiles. <k6;

[0038] Obtain the trend of the total number of queries for keyword m, including upward trend, downward trend, and no obvious trend;

[0039] If the total number of queries for keyword m shows an upward trend, and the total number of representative queries for keyword m is greater than the second threshold, then keyword m is marked as a suspected keyword.

[0040] If the total number of queries for keyword m shows a downward trend, and the total number of representative queries for keyword m is greater than the first threshold, then keyword m is marked as a suspected keyword.

[0041] If the total number of queries for keyword m shows no obvious trend, and the total number of representative queries for keyword m is greater than the first threshold, then keyword m is marked as a suspected keyword.

[0042] Repeatedly retrieve all suspected keywords from the keyword collection and record them as the suspected keyword collection for the first student in the current period.

[0043] Furthermore, based on the trainees' implicit characteristic data, a knowledge blind spot analysis is performed to obtain the trainees' knowledge blind spots. Obtaining the trainees' knowledge blind spot data also includes the following sub-steps:

[0044] For any suspected keyword in the suspected word set, it is recorded as the first suspected word; based on the implicit feature data of the first current student and the corresponding second similar student, the total query time of the first suspected word in training chapter j is obtained and recorded as the total query time of the first suspected word in training chapter j. The total query time of the first suspected word from training chapter 1 to training chapter n1 is repeatedly obtained and sorted in the corresponding order. The part corresponding to the total query count sequence of the first suspected word is recorded as the total time sequence.

[0045] Calculate the ratio of the total duration sequence of the first suspected word to the data at the corresponding position in the total number of queries sequence, and record it as the single query time. After completion, a single time sequence is obtained. Record any single query time in the single time sequence as the first single time DTv. Calculate the weighted time QTv corresponding to DTv, where QTv=[v / (V0*(V0+1) / 2)]*DTv. Repeatedly obtain the weighted time corresponding to all single query times in the total duration sequence and sum them up, and record it as the average single time of keyword m.

[0046] Repeatedly obtain the average single time of all suspected keywords and arrange them in descending order, which is recorded as the average single time sequence. Record the suspected keywords corresponding to the average single time of the first k7% of the average single time sequence as the current blind spot keywords of the first student in the current period, where k7 is the set ratio.

[0047] Furthermore, based on the trainees' implicit characteristic data, a knowledge blind spot analysis is performed to obtain the trainees' knowledge blind spots. Obtaining the trainees' knowledge blind spot data also includes the following sub-steps:

[0048] Based on the latent feature data of the second similar trainees in training chapter (i+1), obtain the query keywords of all second similar trainees in training chapter (i+1), and record them as predicted keywords. Obtain the total number of queries and the total query duration for each predicted keyword in training chapter (i+1), and arrange them in ascending order, and record them as the prediction total number sequence and prediction duration sequence respectively.

[0049] Let any one of the predicted keywords be the first predicted keyword. Get the position number of the total number of queries and the total query time of the first predicted keyword in the corresponding prediction total number sequence and prediction time sequence, and sum them up. This sum is recorded as the blind spot score of the first predicted keyword.

[0050] Repeatedly obtain the blind spot scores of all predicted keywords, arrange them in descending order, and record the predicted keywords corresponding to the top k7% of blind spot scores as the predicted blind spot keywords of the first student in the current period.

[0051] The current blind spot keywords and predicted blind spot keywords of the first batch of students are recorded as the student knowledge blind spot data of the first batch of students.

[0052] Furthermore, based on trainees' knowledge gaps and similarity data, personalized training content is generated for each trainee and pushed out, including the following sub-steps:

[0053] Based on the current blind spot keywords and predicted blind spot keywords of the first batch of trainees, knowledge information related to the current blind spot keywords and predicted blind spot keywords is obtained from the training knowledge graph, and corresponding training content is generated and recorded as the knowledge blind spot content of the first batch of trainees.

[0054] When the first trainee in the current period is receiving training on the training chapter (i+1), personalized recommendations are made to the first trainee based on their knowledge gaps, and this personalized recommendation is repeated for all trainees in the current period.

[0055] The beneficial effects of this invention are as follows: This invention establishes a knowledge graph for medical training and collects explicit and implicit feature data of trainees; it performs trainee similarity analysis based on the explicit and implicit feature data to screen similar trainees and obtain trainee similarity data; it performs knowledge blind spot analysis based on the implicit feature data of trainees to obtain trainee knowledge blind spot data; and it generates personalized training content for each trainee based on the trainee knowledge blind spot data and pushes it to them. This solves the problem of analyzing trainees' potential knowledge blind spots based on relevant data in the training system and customizing training content for trainees when generating training content based on knowledge graphs, thereby filling the trainees' knowledge blind spots and improving the quality and efficiency of medical education.

[0056] This invention calculates an efficiency index based on answer accuracy and learning time, then calculates the average weighted deviation rate, and combines this with the Pearson correlation coefficient to screen for the most similar trainees. The advantage lies in the fact that the average weighted deviation rate quantifies the static differences in efficiency levels, while the Pearson correlation coefficient captures the dynamic trend consistency of efficiency changes. The combination of these two indicators, compared to a single indicator, is more effective in identifying trainees with truly similar knowledge levels and cognitive abilities, improving the accuracy of subsequent knowledge gap analysis. By calculating the representative total number of query keywords and filtering trainees' knowledge gap keywords based on trends and average single-session time, the invention can accurately identify trainees' knowledge gaps. Furthermore, by analyzing the query keywords of similar trainees, it can determine potential future knowledge gaps, thereby generating corresponding training content to fill these gaps and prevent future ones, ultimately improving the quality and efficiency of medical education. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the steps of the method of the present invention;

[0058] Figure 2 This is a flowchart of the first similar student screening process of the present invention;

[0059] Figure 3 This is a flowchart of the second similar student screening process of the present invention;

[0060] Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1, please refer to Figure 1 As shown, this application provides a method for generating and pushing medical training content based on knowledge graphs, including the following steps:

[0063] Step S1 involves establishing a knowledge graph for medical training and collecting explicit and implicit feature data of trainees. Step S1 includes the following sub-steps:

[0064] Step S101: The system used for medical training is denoted as the medical training system; a corresponding knowledge graph is established based on the training content of the medical training, denoted as the training knowledge graph; the multiple chapters of the medical training content process are divided and sequentially denoted as training chapter 1 to training chapter n1, and any training chapter is denoted as training chapter i, where n1 is the total number of training chapters, i∈[1,n]; the knowledge graph structures the concepts, entities, attributes and interrelationships of medical training, which facilitates semantic retrieval, reasoning and automatic assembly of training segments; for the division of the content process, splitting the entire training into multiple chapters can align the trainees' behavior and performance according to the time sequence, which is convenient for subsequent calculations;

[0065] Step S102: Record trainees who have completed medical training as past trainees, and record any past trainee as the first past trainee; record trainees who are currently undergoing medical training as current trainees, and record any current trainee as the first current trainee; past trainees are historical samples for subsequent processing, which can be used to find historical trajectories similar to those of current trainees, and thereby find and predict potential knowledge gaps. For past trainees from a long time ago, if the training content of medical training has changed significantly, it cannot be used as a reference.

[0066] Step S103: For the first trainee in the current period, based on the training chapter i that has been completed, collect the correct answer rate of the first trainee in the current period for training chapter i; and collect the learning time of the first trainee in the current period for training chapter i to obtain the explicit characteristics of the first trainee in the current period for training chapter i; for training chapters that have been completed, the data is complete and can be processed later; for training chapters that are currently being trained, no further processing is performed because the data is incomplete; the explicit characteristics are a direct quantification of mastery level and learning input. The correct answer rate directly reflects the degree to which the knowledge points of the chapter have been mastered; the learning time reflects the input spent to achieve the correct answer rate.

[0067] Step S104: When the first trainees are training on training chapter i, collect the keywords related to medical training that the first trainees search in the medical training system, record them as search keywords, and obtain the number of searches for the search keywords; and obtain the page dwell time of the first trainees on the relevant pages after the search, record it as the search time of the search keywords. After completion, the implicit characteristics of the first trainees in training chapter i are obtained; the implicit characteristics are behavioral signals of the trainees' knowledge blind spots and weak knowledge points; if a user actively searches for a certain keyword and stays on the relevant page, it indicates that the knowledge point has caused him confusion, which is often more likely to expose the knowledge blind spots not covered by the questions than the answer results;

[0068] Step S105: Repeatedly collect the explicit and implicit features of the training chapters 1 to i completed by the first current trainee, and record them as the training feature data of the first current trainee; and collect the explicit and implicit features of the training chapters 1 to i of the first previous trainees, and record them as the training feature data of the first previous trainees; and repeatedly collect the training feature data of all current trainees and previous trainees.

[0069] In practical implementation, knowledge graphs are a structured knowledge representation and storage framework. Through entity-relationship-entity or entity-attribute-value triples, they organize scattered information into a network of knowledge, intuitively describing things in the real world and the relationships between them. Medical training knowledge graphs take serving medical education and training as their core goal, transforming scattered and disordered medical knowledge into a structured and relational semantic network.

[0070] Step S2 involves performing student similarity analysis based on the students' explicit and implicit feature data, filtering out similar students, and obtaining student similarity data. Step S2 includes the following sub-steps:

[0071] For step S201, please refer to... Figure 2 As shown, based on the training characteristic data of the first trainees, the correct answer rate and learning time of the training chapters 1 to i completed by the trainees in the first period are arranged according to the order of the corresponding training chapters, and recorded in order as the first correct answer rate sequence and the first time sequence.

[0072] Step S202: Calculate the ratio of the data at the corresponding position of the first accuracy rate sequence to the data at the corresponding position of the first time sequence, and record it as the efficiency index to obtain the learning efficiency sequence of the first current student and the learning efficiency sequence of the first previous student; Efficiency index = answer accuracy rate / learning time; The efficiency index sequence quantifies the learning quality per unit time by the ratio of answer accuracy rate / learning time, and avoids misjudging the similarity between students.

[0073] Step S203: Denote any efficiency index in the learning efficiency sequence of the first current student as DAj, where j represents the position number, ji∈[1,i], and denote the efficiency index of the same position number in the learning efficiency sequence of the first previous students as WAj.

[0074] Step S204: Calculate the efficiency deviation rate PAj between DAj and WAj, where PAj = 1 - |(DAj - WAj) / max(DAj, WAj)|; Repeatedly calculate the efficiency deviation rate of the efficiency index at the same position in the learning efficiency sequence of the first current student and the first previous student, and arrange them in ascending order of position number, denoted as the efficiency deviation sequence; The efficiency deviation rate quantifies the efficiency difference between current and previous students in a single training chapter, avoiding the overall sequence comparison from masking significant differences in local links; For example, if the mean of the overall efficiency deviation rates of the two are close, but the deviation rate of a certain training chapter is extremely high, it indicates that the previous student differs greatly from the current student in a certain training chapter and should not be classified as a similar student.

[0075] Step S205: For PAj, denoted as the weighted deviation rate QAj corresponding to PAj, where QAj = [j / (i*(i+1) / 2)]*PAj. Repeatedly obtain the weighted deviation rate corresponding to all efficiency deviation rates in the efficiency deviation sequence and calculate the average value, which is denoted as the first similarity index between the first current trainee and the first past trainee. The closer to the training chapter currently completed by the current trainee, i.e., the larger the training chapter number, the higher the weight; the further back the training chapter, the lower the weight. The learning process of medical training is continuous and time-sensitive, and the trainee's learning status will change dynamically as the training chapter progresses. The learning efficiency of recent chapters can better reflect the current true level. The weighted average deviation rate highlights the impact of recent efficiency differences by tilting the weights towards recent chapters, and avoids interference from distant data in similarity judgment.

[0076] Step S206: Calculate the Pearson correlation coefficient between the learning efficiency sequence of the first current learner and the learning efficiency sequence of the first previous learner, denoted as the second similarity index between the first current learner and the first previous learner; the average weighted deviation rate, i.e. the first similarity index, only reflects the static difference in learning efficiency levels, but cannot capture the consistency of the dynamic trend of learning efficiency changes; to make up for the shortcomings of the first similarity index in emphasizing static differences and neglecting dynamic trends, it can identify learners with slightly different efficiency levels but consistent change rates, such as learners in medical training who master knowledge points slowly at first and then quickly, and are more similar to previous learners who also mastered knowledge points slowly at first and then quickly, avoiding the mistaken exclusion of potential similar learners due to slightly larger absolute differences.

[0077] Step S207: Repeatedly obtain the first similarity index and the second similarity index between the first current student and all previous students, and arrange all the first similarity indices in descending order and arrange all the second similarity indices in ascending order, and record them as the first similarity sequence and the second similarity sequence respectively.

[0078] Step S208: Obtain the position numbers of the first similarity index and the second similarity index of the first previous student in the corresponding first similarity sequence and second similarity sequence, respectively, and sum them up to record as the similarity score of the first current student and the first previous student; make a judgment by combining the first similarity index and the second similarity index to avoid the one-sidedness of a single parameter, and ensure that similar students not only have similar current learning efficiency, but also have the same trend of learning efficiency change, so as to ensure the relevance and accuracy of subsequent content generation;

[0079] Step S209: Repeatedly obtain the similarity scores of the first current student and all previous students, and arrange them in descending order to form a similarity score sequence. Record the previous students corresponding to the first k1% of the similarity score sequence as the first similar students of the first current student. Here, k1% is a set ratio. In this embodiment, k1%=30%, which can be flexibly set according to the actual application scenario.

[0080] For step S210, please refer to... Figure 3 As shown, obtain all the query keywords of all trainees in the current training chapters 1 to i, and record them as the keyword set, i.e. the query keyword category set. Record the query keywords in the keyword set as keyword 1 to keyword n2, where n2 represents the total number of query keywords in the keyword set. Record any keyword as keyword m, where m represents the sequence number.

[0081] Step S211: For the first current period trainees, based on the implicit feature data of training chapter 1 to training chapter i of the first current period trainees, count the total number of queries corresponding to keyword m, and record it as the total number of queries CXm corresponding to keyword m of the first current period trainees;

[0082] Step S212: Repeatedly obtain the total number of queries for all query keywords in the keyword set corresponding to the first student in the current period, and establish the query feature vector TM corresponding to the first student in the current period, where TM={CX1, CX2, ..., CXn2}; TM is used to describe the search intensity of the student for the keywords, that is, the frequency of the keywords they pay attention to; it reflects the distribution of the student's focus and confusion points.

[0083] Step S213: Repeatedly obtain the query feature vector of the first similar student, and calculate the cosine similarity between the query feature vector of the first current student and the query feature vector of each first similar student. Arrange them in descending order and record them as the cosine similarity sequence. Record the first similar students corresponding to the first k2% of the cosine similarity sequence that are greater than k3 as the second similar students of the first current student. Here, k2% is the set ratio and k3 is the set threshold. In this embodiment, k2%=30% and k3=0.4, which can be flexibly set according to the actual application scenario.

[0084] In the specific implementation process, cosine similarity is used to quantify the similarity of knowledge needs among students. It can effectively measure the similarity of vector directions, that is, whether the types of keywords they focus on are consistent. The first similar students selected based on efficiency and sequence morphology are candidate students who are similar in terms of behavioral efficiency. Cosine similarity is used to further screen the first similar students to ensure that the final similar students are also similar in terms of their focus and confusion.

[0085] Step S3 involves analyzing the knowledge blind spots based on the trainees' implicit characteristic data to obtain the trainees' knowledge blind spots data. Step S3 includes the following sub-steps:

[0086] Step S301: For any keyword m in the keyword set, based on the implicit feature data of the first current student and the corresponding second similar student, obtain the total number of queries for keyword m in training chapter j, that is, the total number of queries for all the first current students and the corresponding second similar students, and record it as the total number of queries for keyword m in training chapter j. Repeat the process of obtaining the total number of queries for keyword m from training chapter 1 to training chapter n1, and sort them in the corresponding order, and record them as the initial total number sequence of keyword m.

[0087] Step S302: Take the first query total greater than k4 in the initial total number sequence of keyword m as the starting point, and record the total number of queries from the starting point and thereafter as the query total sequence of keyword m. Here, k4 is the set threshold. In this embodiment, k4=0. It can be set flexibly because some keywords will not appear in the initial training chapter, but in the middle training chapter. Therefore, the total number of queries in the chapters before the first appearance should be 0, which has no reference value and can be discarded. In order to avoid misjudgment caused by data fluctuations, k4 can also be flexibly set to a number greater than 0.

[0088] Step S303: Based on the query total sequence of keyword m, the total number of queries in the query total sequence is denoted as V0, and any query total is denoted as CRv, where v represents the position number. The weighted total QRv corresponding to CRv is calculated, where QRv = [v / (V0*(V0+1) / 2)]*CRv. The weighted total corresponding to all query totals in the query total sequence is repeatedly obtained and summed, and denoted as the representative total of keyword m. The representative total is weighted according to the chapter order to amplify the query keywords that have recently appeared, helping to identify keywords whose importance has recently increased, representing the students' query needs for keywords.

[0089] Step S304: Repeatedly obtain the total number of representatives of all keywords in the keyword collection, sort them in descending order, denote it as the representative total number sequence, obtain the k5 percentile and the k6 percentile of the representative total number, and denote them as the first threshold and the second threshold in sequence, where k5 and k6 are set percentiles, and k5 < k6; in this embodiment, k5 = 40, k6 = 60, and the first threshold is greater than the second threshold;

[0090] Step S305: Obtain the change trend of the query total number sequence of keyword m, and the change trend includes an upward trend, a downward trend, and no obvious trend; when the representative total number of the query total number sequence shows a fluctuating state, or the representative total number of the query total number sequence remains basically unchanged, it can be classified as no obvious trend;

[0091] Step S306: If the query total number sequence of keyword m shows an upward trend and the representative total number of keyword m is greater than the second threshold, then mark keyword m as a suspected keyword; the query demand for the keyword gradually increases as the training progresses, which may mean that the trainees are becoming more and more confused about this knowledge point and need to query repeatedly; so it only needs to be greater than the smaller second threshold;

[0092] Step S307: If the query total number sequence of keyword m shows a downward trend and the representative total number of keyword m is greater than the first threshold, then mark keyword m as a suspected keyword; the overall query demand for the keyword decreases, and being greater than the first threshold indicates that there are still high-frequency queries recently, that is, it means that this knowledge point has not been fully mastered;

[0093] Step S308: If the query total number sequence of keyword m shows no obvious trend and the representative total number of keyword m is greater than the first threshold, then mark keyword m as a suspected keyword; the query demand for the keyword remains stable in each link, that is, the trainees continue to have questions about this knowledge point, but the degree of doubt does not change significantly, which may be a knowledge point that most trainees have not mastered for a long time;

[0094] Step S309: Repeatedly obtain all suspected keywords in the keyword collection, and denote it as the suspected word collection of the first current trainees; that is, the possible knowledge blind spots and weak knowledge points of the first current trainees.

[0095] Step S310: Denote any suspected keyword in the suspected word collection as the first suspected word; according to the implicit feature data of the first current trainees and the corresponding second similar trainees, obtain the total query time of the first suspected word corresponding to the training chapter j, and denote it as the total query duration of the first suspected word in the training chapter j. Repeatedly obtain the total query duration of the first suspected word in training chapter 1 - training chapter n1, sort them in the corresponding order, and denote the part corresponding to the query total number sequence of the first suspected word as the total duration sequence;

[0096] Step S311: Calculate the ratio of the total duration sequence of the first suspected word to the data at the corresponding position in the total number of queries sequence, denoted as the single query time. That is, the ratio of the total query duration of the first suspected word to the total number of queries in the same training chapter. Single query time = total query duration / total number of queries. After completion, a single time sequence is obtained. Any single query time in the single time sequence is denoted as the first single time DTv. Calculate the weighted time QTv corresponding to DTv, where QTv = [v / (V0*(V0+1) / 2)]*DTv. Repeatedly obtain the weighted time corresponding to all single query times in the total duration sequence and sum them up, denoted as the average single time of keyword m. The weighted time further emphasizes the recent query time. A large average single time indicates that the cognitive load of the keyword is large and difficult to understand, representing the real difficulty and also indicating the knowledge blind spots of trainees.

[0097] Step S312: Repeatedly obtain the average single time of all suspected keywords and arrange them in descending order, which is recorded as the average single time sequence. Record the suspected keywords corresponding to the average single time of the first k7% of the average single time sequence as the current blind spot keywords of the first student in the current period, where k7 is the set ratio; in this embodiment, k7=35%, which can be flexibly set.

[0098] Step S313: Based on the latent feature data of the second similar trainees in training chapter (i+1), obtain the query keywords of all second similar trainees in training chapter (i+1), and record them as predicted keywords. Obtain the total number of queries and the total query duration for each predicted keyword in training chapter (i+1), and arrange them in ascending order, recording them as the predicted total number sequence and the predicted duration sequence respectively. Training chapter (i+1) is the training chapter that the first trainee is currently attending or about to attend.

[0099] Step S314: Record any predicted keyword as the first predicted word, obtain the position number of the total number of queries and the total query time of the first predicted word in the corresponding prediction total number sequence and prediction time sequence, and sum them up to record the blind spot score of the first predicted word; the larger the blind spot score, the greater the possibility that the first predicted word is a knowledge blind spot.

[0100] Step S315: Repeatedly obtain the blind spot scores of all predicted keywords, arrange them in descending order, and record the predicted keywords corresponding to the top k7% of blind spot scores as the predicted blind spot keywords of the first student in the current period.

[0101] Step S316: Record the current blind spot keywords and predicted blind spot keywords of the first student in the current period as the student knowledge blind spot data of the first student in the current period.

[0102] In the specific implementation process, the current blind spot keywords represent knowledge points that students and similar students have shown high attention to and high learning difficulty in the completed chapters, reflecting the students' knowledge blind spots and weak knowledge points, which need to be filled in subsequent training; the predicted blind spot keywords are knowledge points that similar students have shown high attention to and high learning difficulty in the training process, that is, the students' potential knowledge blind spots and weak knowledge points in the future, which need to be focused on in subsequent training to avoid the emergence of knowledge blind spots.

[0103] Step S4 involves generating personalized training content for each student based on their knowledge gaps and similarity data, and then pushing this content to them. Step S4 includes the following sub-steps:

[0104] Step S401: Based on the current blind spot keywords and predicted blind spot keywords of the first trainee, obtain relevant knowledge information about the current blind spot keywords and predicted blind spot keywords from the training knowledge graph, and generate corresponding training content, which is recorded as the knowledge blind spot content of the first trainee.

[0105] Step S402: When the first student in the current period is training on the training chapter (i+1), personalized push is made to the first student in the current period based on the knowledge gaps of the first student in the current period, and personalized push is repeated for all students in the current period.

[0106] In the specific implementation process, the knowledge information related to keywords includes the core definition, essential connotation, attributes, characteristic parameters, and associated knowledge network of keywords.

[0107] Example 2, please refer to Figure 4 As shown, Figure 4 A schematic diagram of an electronic device is provided, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer-readable instructions, and the processor can call these instructions. When the processor executes a computer-readable instruction, it performs steps such as those in a knowledge graph-based medical training content generation and delivery method to achieve the following functions: establishing a knowledge graph for medical training and collecting explicit and implicit feature data of trainees; performing trainee similarity analysis based on the explicit and implicit feature data, filtering similar trainees, and obtaining trainee similarity data; performing knowledge blind spot analysis based on the trainees' implicit feature data to obtain trainees' knowledge blind spots, and obtaining trainee knowledge blind spot data; and generating and delivering personalized training content for each trainee based on the trainee knowledge blind spot data and trainee similarity data.

[0108] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0109] Example 3: This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs steps such as those in the knowledge graph-based medical training content generation and delivery method to achieve the following functions: establishing a knowledge graph for medical training and collecting explicit and implicit feature data of trainees; performing trainee similarity analysis based on the explicit and implicit feature data of trainees, filtering similar trainees to obtain trainee similarity data; performing knowledge blind spot analysis based on the implicit feature data of trainees to obtain trainees' knowledge blind spots to obtain trainee knowledge blind spot data; and generating personalized training content for each trainee based on the trainee knowledge blind spot data and trainee similarity data, and then delivering it.

[0110] Based on the above description of the embodiments, the embodiments of the present invention can be provided as methods, systems, or computer program products. Based on this understanding, the above technical solutions, in essence or in terms of their contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments.

[0111] In the embodiments provided in this application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces. The indirect coupling or communication connection between systems, modules, and units may be electrical, mechanical, or other forms.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1.A method for generating and pushing medical training content based on a knowledge graph, characterized in that, The method comprises the following steps: establishing a knowledge graph for medical training, collecting explicit feature data and implicit feature data of trainees; conducting trainee similarity analysis on the explicit feature data and the implicit feature data of the trainees, screening similar trainees of the trainees, and obtaining trainee similarity data; conducting knowledge blind area analysis on the implicit feature data of the trainees, obtaining the knowledge blind area of the trainees, and obtaining trainee knowledge blind area data; based on the trainee knowledge blind area data and the trainee similarity data, producing personalized training content for each trainee and pushing the personalized training content; conducting trainee similarity analysis on the explicit feature data and the implicit feature data of the trainees, screening similar trainees of the trainees, and obtaining trainee similarity data comprises the following sub-steps: arranging the corresponding answer accuracy and learning time of the first-period trainee in the training chapters 1-i that have been completed in the corresponding training chapters in order, and recording the first accuracy sequence and the first time sequence in order; calculating the ratio of the data at the corresponding positions of the first accuracy sequence and the first time sequence, recording the ratio as an efficiency index, obtaining the learning efficiency sequence of the first-period trainee, and obtaining the learning efficiency sequence of the first past-period trainee; recording any one efficiency index in the learning efficiency sequence of the first-period trainee as DAj, where j represents the position number, j∈[1, i], and recording the efficiency index with the same position number in the learning efficiency sequence of the first past-period trainee as WAj; calculating the efficiency deviation rate PAj of DAj and WAj, where PAj=1-|(DAj-WAj) / max(DAj,WAj)|; repeating the calculation of the efficiency deviation rates of all efficiency indexes with the same position number in the learning efficiency sequences of the first-period trainee and the first past-period trainee, arranging the efficiency deviation rates in order from small to large according to the position number, and recording the efficiency deviation sequence as PAj; for PAj, recording the corresponding weighted deviation rate QAj, where QAj=[j / (i*(i+1) / 2)]*PAj, repeating the calculation of the weighted deviation rates corresponding to all efficiency deviation rates in the efficiency deviation sequence, and averaging the weighted deviation rates to obtain the first similarity index of the first-period trainee and the first past-period trainee; calculating the Pearson correlation coefficient of the learning efficiency sequence of the first-period trainee and the learning efficiency sequence of the first past-period trainee, and recording the Pearson correlation coefficient as the second similarity index of the first-period trainee and the first past-period trainee. 2.The knowledge graph-based medical training content generation and push method according to claim 1, characterized in that, The method comprises the following steps: The system used for medical training is recorded as a medical training system; a corresponding knowledge graph is established according to the training content of the medical training, and the corresponding knowledge graph is recorded as a training knowledge graph; and the content process of the medical training is divided into a plurality of chapters, which are recorded in order as training chapter 1-training chapter n1, and any one training chapter is recorded as training chapter i, where n1 is the total number of training chapters, i∈[1, n]; the trainees who have completed the medical training are recorded as past-period trainees, and any one past-period trainee is recorded as a first past-period trainee; the trainees who are currently undergoing the medical training are recorded as current-period trainees, and any one current-period trainee is recorded as a first current-period trainee. 3.The knowledge graph-based medical training content generation and push method according to claim 2, characterized in that, Establishing a knowledge graph for medical training, and collecting explicit feature data and implicit feature data of trainees includes the following sub-steps: For the first current trainee, collecting the correct answer rate of the first current trainee in the training chapter i corresponding to the training chapter i completed by the first current trainee; and collecting the learning time of the first current trainee in the training chapter i, to obtain the explicit features of the first current trainee in the training chapter i; When the first current trainee is training in the training chapter i, collect the keywords related to medical training that the first current trainee queries in the medical training system, denoted as query keywords, and obtain the query times of the query keywords; And get the page dwell time of the relevant page after the first current trainee queries, denoted as query time of query keywords, and complete the implicit features of the first current trainee in the training chapter i; Repeat the collection of explicit features and implicit features of the first current trainee in the training chapter 1 to the training chapter i completed by the first current trainee, denoted as training feature data of the first current trainee; and collect the explicit features and implicit features of the first past trainee in the training chapter 1 to the training chapter i, denoted as training feature data of the first past trainee; Repeat the collection of training feature data of all current trainees and past trainees. 4.The knowledge graph-based medical training content generation and push method according to claim 3, characterized in that, According to the explicit feature data and the implicit feature data of the trainees, the similarity of the trainees is analyzed, and similar trainees of the trainees are screened, and the trainee similarity data further includes the following sub-steps: Repeat the collection of the first similarity index and the second similarity index of the first current trainee and all past trainees, and arrange all the first similarity indexes in descending order, and arrange all the second similarity indexes in ascending order, and sequentially record them as the first similarity sequence and the second similarity sequence respectively; Respectively obtain the position sequence numbers of the corresponding first similarity index and second similarity index of the first past trainee in the corresponding first similarity sequence and second similarity sequence, and sum them up, denoted as similarity score of the first current trainee and the first past trainee; Repeat the collection of the similarity score of the first current trainee and all past trainees, and arrange them in descending order, denoted as similarity score sequence, and the past trainees corresponding to the first similarity trainees of the first current trainee are recorded as the first similarity trainees of the first current trainee, wherein k1% is a set proportion. 5.The knowledge graph-based medical training content generation and pushing method according to claim 4, characterized in that, According to the explicit feature data and the implicit feature data of the trainees, the similarity of the trainees is analyzed, and similar trainees of the trainees are screened, and the trainee similarity data further includes the following sub-steps: Obtain the query keywords of all current trainees in the training chapter 1 to the training chapter i, denoted as keyword set, and sequentially record the query keywords in the keyword set as keyword 1 to keyword n2, wherein n2 represents the total number of query keywords in the keyword set, and any one keyword is denoted as keyword m, wherein m represents the serial number; For the first current trainee, according to the implicit feature data of the first current trainee in the training chapter 1 to the training chapter i, the total sum of the query times corresponding to the keyword m is calculated, denoted as the total query times CXm of the keyword m corresponding to the first current trainee; Repeating the acquisition of the total number of queries of all query keywords in the keyword set corresponding to the first-period student, and establishing the query feature vector TM corresponding to the first-period student, wherein TM={CX1, CX2, …, CXn2}; Repeating the acquisition of the query feature vector of the first similar student, and calculating the cosine similarity between the query feature vector of the first-period student and the query feature vector of each first similar student, and arranging them in descending order, denoted as a cosine similarity sequence, and recording the first similar student corresponding to the cosine similarity greater than k3 in the first k2% of the cosine similarity sequence as the second similar student of the first-period student, wherein k2% is a set proportion, and k3 is a set threshold. 6.The knowledge graph-based medical training content generation and pushing method according to claim 5, characterized in that, According to the implicit feature data of the student, the knowledge blind area of the student is analyzed to obtain the student knowledge blind area data, including the following sub-steps: For any keyword m in the keyword set, according to the implicit feature data of the first-period student and the corresponding second similar student, the total number of queries of keyword m in training chapter j is obtained, denoted as the total number of queries of keyword m in training chapter j, and the total number of queries of keyword m in training chapter 1-training chapter n1 is repeatedly obtained, and arranged in corresponding order, denoted as the initial total number sequence of keyword m; The first query total number greater than k4 in the initial total number sequence of keyword m is taken as the starting point, and the starting point and the query total numbers thereafter are recorded as the query total number sequence of keyword m, wherein k4 is a set threshold, According to the query total number sequence of keyword m, the total number of query total numbers in the query total number sequence is recorded as V0, and any query total number is recorded as CRv, wherein v represents the position sequence number, the weighted total number QRv corresponding to CRv is calculated, wherein QRv=[v / (V0*(V0+1) / 2)]*CRv, the weighted total number corresponding to all query total numbers in the query total number sequence is repeatedly obtained, and summed up, denoted as the representative total number of keyword m; The representative total numbers of all keywords in the keyword set are repeatedly obtained and arranged in descending order, denoted as a representative total number sequence, the k5th percentile and the k6th percentile of the representative total number are obtained, and denoted as the first threshold and the second threshold in order, respectively, wherein k5 and k6 are set percentiles, and k5<k6; The trend of the query total number sequence of keyword m is obtained, including the upward trend, the downward trend, and the no obvious trend; If the query total number sequence of keyword m shows an upward trend, and the representative total number of keyword m is greater than the second threshold, keyword m is marked as a suspected keyword; If the query total number sequence of keyword m shows a downward trend, and the representative total number of keyword m is greater than the first threshold, keyword m is marked as a suspected keyword; If the query total number sequence of keyword m shows no obvious trend, and the representative total number of keyword m is greater than the first threshold, keyword m is marked as a suspected keyword; All suspected keywords in the keyword set are repeatedly obtained, denoted as the suspected keyword set of the first-period student. 7.The knowledge graph-based medical training content generation and push method according to claim 6, characterized in that, According to the implicit characteristic data of the trainee, knowledge blind area analysis is performed to obtain the knowledge blind area of the trainee, and the trainee knowledge blind area data further includes the following sub-steps: For any one suspected key word in the suspected key word set, the suspected key word is recorded as a first suspected word; according to the implicit characteristic data of the first trainee and the corresponding second similar trainee, the total sum of the query time of the first suspected word corresponding to the training chapter j is obtained, which is recorded as the total query time of the first suspected word in the training chapter j, the total query time of the first suspected word in the training chapter 1 to the training chapter n1 is repeatedly obtained, and the corresponding order is sorted, and the part corresponding to the total query number sequence of the first suspected word is recorded as the total time sequence; The ratio of the data at the corresponding position of the total time sequence and the total query number sequence of the first suspected word is calculated, which is recorded as the single query time, and after completion, the single time sequence is obtained, any one single query time in the single time sequence is recorded as the first single time DTv, and the weighted time QTv corresponding to DTv is calculated, wherein QTv = [v / (V0 * (V0 + 1) / 2)] * DTv, the weighted time corresponding to all single query times in the total time sequence is repeatedly obtained, and the sum is calculated, which is recorded as the average single time of the key word m; The average single time of all suspected key words is repeatedly obtained, and arranged in descending order, recorded as the average single sequence, and the suspected key word corresponding to the average single time of the first k7% in the average single sequence is recorded as the current blind area key word of the first trainee, wherein k7 is a set proportion. 8.The knowledge graph-based medical training content generation and push method according to claim 7, characterized in that, According to the implicit characteristic data of the trainee, knowledge blind area analysis is performed to obtain the knowledge blind area of the trainee, and the trainee knowledge blind area data further includes the following sub-steps: According to the implicit characteristic data of the second similar trainee in the training chapter (i+1), all query key words of the second similar trainee in the training chapter (i+1) are obtained, which are recorded as predicted key words, and the total number of queries and the total time of queries of each predicted key word in the training chapter (i+1) are obtained, and arranged in ascending order, respectively, recorded as the predicted total number sequence and the predicted time sequence; Any one predicted key word is recorded as a first predicted word, the position sequence number of the total number of queries and the total time of queries of the first predicted word in the corresponding predicted total number sequence and predicted time sequence is obtained, and the sum is calculated, which is recorded as the blind area score of the first predicted word; The blind area score of all predicted key words is repeatedly obtained, and arranged in descending order, and the predicted key word corresponding to the first k7% blind area score is recorded as the predicted blind area key word of the first trainee; The current blind area key word and the predicted blind area key word of the first trainee are recorded as the trainee knowledge blind area data of the first trainee. 9.The knowledge graph-based medical training content generation and push method according to claim 8, characterized in that, Based on the trainee knowledge blind area data and the trainee similar data, individualized training content is produced for each trainee, and the pushing includes the following sub-steps: According to the current blind area key word and the predicted blind area key word of the first trainee, the knowledge information related to the current blind area key word and the predicted blind area key word is obtained according to the training knowledge graph, and the corresponding training content is generated, which is recorded as the knowledge blind area content of the first trainee. When the first current student carries out the training chapter (i+1) training, the first current student is pushed according to the knowledge blind area content of the first current student, and the personalized pushing of all current students is repeated.

Images