Medical college online learning system based on big data analysis

Abstract

The invention provides a medical college online learning system based on big data analysis, and relates to the technical field of data processing, and the system comprises a knowledge graph construction module which is used for carrying out the preprocessing and multi-dimensional feature extraction of original learning data uploaded in a medical college online learning environment, and generating a dynamically updated medical knowledge graph; the vector conversion module is used for encrypting and decomposing the entity attribute identifier in the medical knowledge graph by utilizing elliptic curve cryptography to generate a multi-dimensional feature vector with three-dimensional space coordinates; and the data extraction module is used for constructing a data distribution model based on the space coordinates of the multi-dimensional feature vectors and extracting local density concentration characteristic values and boundary dispersion characteristic values of the data distribution model. According to the method, the accuracy, the safety and the efficiency of online learning of medical colleges are comprehensively improved by dynamically constructing the medical knowledge graph, encrypting the feature vectors, accurately evaluating the entity association degree and combining meta-learning to generate a differentiated learning path.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an online learning system for medical colleges based on big data analysis. Background Technology

[0002] Traditional online medical learning platforms are increasingly showing shortcomings in adapting to the needs of modern medical education. Currently, many online medical learning systems still primarily consist of static course resource libraries or basic management functions, which have certain limitations. The level of knowledge integration needs to be improved: Multimodal learning resources such as medical textbooks, clinical cases, surgical videos, and imaging data are often stored in a scattered state, which easily forms "data silos" and fails to build a closely related knowledge network. This may affect students' in-depth understanding of the internal logic between knowledge points.

[0003] Personalized recommendations lack accuracy and dynamism: Existing recommendation algorithms rely heavily on students' historical clicks or browsing records, and lack in-depth analysis of the knowledge system structure. There is room for improvement in dynamically adjusting recommended content based on students' learning progress, professional direction (such as surgery or internal medicine), and clinical practice performance. This may lead to homogenization of recommended content, which may affect learning efficiency to some extent. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide an online learning system for medical colleges based on big data analysis, which integrates multimodal medical data and accurately generates personalized learning paths to improve the quality and efficiency of medical education.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an online learning system for medical colleges based on big data analysis, comprising: The knowledge graph construction module is used to preprocess and extract multi-dimensional features from the raw learning data uploaded to the online learning environment of medical colleges and universities, and generate a dynamically updated medical knowledge graph. The vector transformation module is used to use elliptic curve cryptography to encrypt and decompose entity attribute identifiers in the medical knowledge graph, generating multidimensional feature vectors with three-dimensional spatial coordinates. The data extraction module is used to construct a data distribution model based on the spatial coordinates of multidimensional feature vectors, and to extract the local density concentration characteristic value and boundary dispersion characteristic value of the data distribution model. The center point optimization module is used to calculate the initial center point of the multidimensional feature vector, generate center calibration parameters based on the local density concentration characteristic value and the boundary dispersion characteristic value, and dynamically adjust the initial center point through the center calibration parameters to generate the optimized center point; The correlation evaluation module is used to calculate the geometric distance from each entity data point to the optimization center point, sort them by distance, and output the correlation value by combining the medical feature weight matrix and the preset evaluation function; and generate a differentiated learning path recommendation strategy by combining the meta-learning algorithm. The data storage module is used to allocate multidimensional feature vectors to corresponding storage nodes according to the correlation degree value; to encrypt the multidimensional feature vectors using an asymmetric encryption algorithm; to bind the encrypted feature vectors with the correlation degree value, knowledge graph version identifier and timestamp to construct encrypted data blocks and write them to the blockchain.

[0006] In a second aspect, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, enable the one or more processors to implement the system.

[0007] Thirdly, a computer-readable storage medium storing a program that, when executed by a processor, implements the system.

[0008] The above-described solution of the present invention has at least the following beneficial effects: The knowledge graph construction module generates a dynamically updated medical knowledge graph, integrating scattered raw learning data into a tightly connected knowledge network. This helps learners understand the inherent logic between knowledge points and solves the problems of scattered and outdated knowledge in traditional systems. The vector transformation module uses elliptic curve cryptography to encrypt entity attribute identifiers, and the data storage module uses asymmetric encryption combined with blockchain technology. This double encryption, along with the immutability of the blockchain, ensures the privacy, security, and reliability of medical data, reducing the risk of data leakage or tampering.

[0009] By leveraging the collaborative work of data extraction, optimization of centroids, and correlation assessment modules, combined with medical feature weight matrices and meta-learning algorithms, the system can accurately calculate entity correlations and generate differentiated learning paths. This adapts to the learning progress, professional direction, and other needs of different learners, improving learning efficiency. Centroids are optimized based on the characteristic values ​​of multi-dimensional feature vectors, and storage nodes are allocated according to correlation, resulting in a more rational data distribution, facilitating efficient extraction and management, and improving the overall system operating efficiency. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of an online learning system for medical colleges based on big data analysis, provided by an embodiment of the present invention. Detailed Implementation

[0011] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0012] like Figure 1 As shown, an embodiment of the present invention proposes an online learning system for medical colleges based on big data analysis, comprising: The knowledge graph construction module is used to preprocess and extract multi-dimensional features from the raw learning data uploaded to the online learning environment of medical colleges and universities, and generate a dynamically updated medical knowledge graph. The vector transformation module is used to use elliptic curve cryptography to encrypt and decompose entity attribute identifiers in the medical knowledge graph, generating multidimensional feature vectors with three-dimensional spatial coordinates. The data extraction module is used to construct a data distribution model based on the spatial coordinates of multidimensional feature vectors, and to extract the local density concentration characteristic value and boundary dispersion characteristic value of the data distribution model. The center point optimization module is used to calculate the initial center point of the multidimensional feature vector, generate center calibration parameters based on the local density concentration characteristic value and the boundary dispersion characteristic value, and dynamically adjust the initial center point through the center calibration parameters to generate the optimized center point; The correlation evaluation module is used to calculate the geometric distance from each entity data point to the optimization center point, sort them by distance, and output the correlation value by combining the medical feature weight matrix and the preset evaluation function; and generate a differentiated learning path recommendation strategy by combining the meta-learning algorithm. The data storage module is used to allocate multidimensional feature vectors to corresponding storage nodes according to the correlation degree value; to encrypt the multidimensional feature vectors using an asymmetric encryption algorithm; to bind the encrypted feature vectors with the correlation degree value, knowledge graph version identifier and timestamp to construct encrypted data blocks and write them to the blockchain.

[0013] In this embodiment of the invention, by integrating preprocessed multi-dimensional raw learning data and dynamically updating the knowledge graph, the correlation between medical knowledge can be strengthened, helping learners systematically understand the knowledge system and enabling real-time iteration of knowledge. Elliptic curve cryptography is used to generate three-dimensional feature vectors, improving data security while preserving entity attribute features. Extracting local density and boundary dispersion characteristics accurately captures data distribution patterns, providing a reliable basis for center point optimization and improving the accuracy of data feature analysis.

[0014] By dynamically adjusting and generating optimized center points, the impact of data distribution bias is reduced. Combining geometric distance, weight matrices, and meta-learning algorithms to generate differentiated paths accurately matches learners' needs, enabling personalized recommendations and improving learning efficiency. Allocating storage nodes based on relevance and incorporating encryption and blockchain technology optimizes data storage efficiency while ensuring data privacy and immutability, enhancing data reliability.

[0015] In a preferred embodiment of the present invention, preprocessing and multi-dimensional feature extraction of the raw learning data uploaded to the online learning environment of medical colleges to generate a dynamically updated medical knowledge graph may include: In this embodiment of the invention, various types of raw data uploaded from the online learning environment are aggregated, covering multimodal data such as medical textbook texts, clinical case records, surgical teaching videos, students' clinical operation records, and test answer data; duplicate data is removed, such as multiple uploads of the same case; erroneous data is corrected, such as obvious diagnostic errors in case studies or typos in textbook texts; invalid data is filtered, such as files with disordered formats, unrecognizable files, or redundant information unrelated to medical learning. Data of different formats is uniformly converted into a format that the system can process, such as converting video files of different formats into a unified encoded video format, and recognizing scanned copies of paper textbooks as an editable text format; professional terminology in the data is standardized to ensure that the same concept is consistently expressed in the data, such as unifying the Chinese and English names and alternative names of diseases.

[0016] Multi-dimensional feature extraction: For text-based data (such as textbooks, medical records, etc.), the professional concept units involved are identified by performing structured parsing of the text content, clarifying the specific attribute descriptions of each concept unit, and sorting out the inherent relationship information between different concept units; For video data (such as surgical videos), the video frames are analyzed frame by frame to extract key actions in the process, identification information of the tools used, and specific parts of the body being operated on. At the same time, the audio information in the video is converted into analyzable text to extract professional information and related content. For operation record data (such as clinical operation records), the data is analyzed to include information such as the type of patient being treated, the name of the operation performed, the sequence of the operation, and the result achieved. Information related to the degree of conformity between the operation process and established standards is also extracted and analyzed.

[0017] Dynamically updated medical knowledge graph generation: In the initial knowledge graph construction phase, extracted professional concept units are used as network nodes, and the relationships between concept units are used as connecting edges, thus building a basic network knowledge structure. When new learning data is input, new concept units and their relationships are obtained according to the aforementioned preprocessing and feature extraction process. These new contents are then integrated into the existing knowledge network: if a new concept unit is associated with an existing node in the network, a corresponding connection is established; if new relationships are discovered or existing relationships need to be adjusted, the connections in the network are updated, ensuring that the knowledge network reflects the latest state of professional knowledge in a timely manner.

[0018] In another preferred embodiment of the present invention, using elliptic curve cryptography to encrypt and decompose entity attribute identifiers in a medical knowledge graph to generate a multidimensional feature vector with three-dimensional spatial coordinates may include: Treat the entity attribute identifiers in the medical knowledge graph as independent data units, and perform the following operations on each data unit: Select a base point on the elliptic curve, and perform a dot product operation using the value of the entity attribute identifier as a multiple of the private key to generate encrypted coordinate points; Extract the horizontal and vertical coordinates of the encrypted coordinate points as feature components; The feature components are bound to entity attribute identifiers and knowledge graph version identifiers to form a three-dimensional spatial coordinate system, where the X-axis is the entity attribute identifier, the Y-axis is the horizontal coordinate of the feature components, and the Z-axis is the vertical coordinate of the feature components.

[0019] In this embodiment of the invention, the entity attribute identifiers in the medical knowledge graph are sorted out, the type of each identifier is clarified (such as International Classification of Diseases (ICD-10), national drug code, anatomical location number, etc.), and a unique numerical label is assigned to each identifier (for example, the ICD-10 code J18.9 for "pneumonia" corresponds to the numerical value 1001, and the drug code for "penicillin" corresponds to the numerical value 2002, etc.), ensuring that each label is unique within the system, thus serving as an independent data unit; simultaneously, an elliptic curve conforming to medical data encryption specifications is selected (such as the SM2 elliptic curve standard recommended by the State Cryptography Administration), the parameters of which include: The equation of the curve is defined as y² = x³ + ax + b, where a and b are preset constants (a = -3, b = 0x28E9FA9E9D9F5E344D5A9E4BCF6509A7F39789F515AB8F92DDBCBD414D940E93), which are constants explicitly defined in the SM2 standard to ensure the safety and universality of the curve; the finite domain in which the curve lies restricts the values ​​of the abscissa and ordinate to between 0 and p-1 (where p is the modulus of the finite domain, SM2). In the standard, p = 0xFFFFFFFEFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF000000000FFFFFFFFFFFFFFFF; and the order of the curve (i.e., the quantity of points on the curve; in the SM2 standard, the order is 0xFFFFFFFEFFFFFFFFFFFFFFFFFFFFFF7203DF6B21C6052B53BBF40939D5412345, ensuring that the operators on the points of the curve meet the encryption requirements). Select a publicly known base point G on this curve (e.g., a point with coordinates (x0, y0). In the SM2 standard, the coordinates of G are x0=0x32C4AE2C1F1981195F9904466A39C9948FE30BBFF2660BE1715A4589334C74C7BC3736A2F4F6779C59BDCEE36B692153D0A9877CC). 62A474002DF32E52139F0A0=0xBC3736A2F4F6779C59BDCEE36B692153D0A9877CC62A474002DF32E52139F0A0), this point satisfies the characteristic that "it is difficult to derive the original multiple through a finite number of operations", and serves as the common starting point for all encryption operations, allowing all data units within the system to call it uniformly during the encryption process.

[0020] Dot multiplication based on private key multiples: For each entity attribute identifier's numerical label (such as 1001 and 2002 above), it is used as a "private key multiplier" and multiplied by the base point G of the elliptic curve. The specific operation is as follows: The sum of any two points P(x1, y1) and Q(x2, y2) on the curve is point R(x3, y3). Calculation requires specific geometric methods to determine the coordinates of R (e.g., drawing a straight line through P and Q, finding the other intersection point of the line and the curve, and then reflecting it about the x-axis to obtain R). If P and Q are the same point, the sum is calculated using the intersection point of the tangent line at that point and the curve. If the identifier's numerical label is n (e.g., n=1001), then the dot product operation n×G is essentially "adding the base point G to itself n-1 times": the first operation is G+G=2G, the second is 2G+G=3G, and so on, until 1000 additions are obtained, resulting in 1001×G. The final result is an encrypted coordinate point P located on the elliptic curve. n (x) n y n ).

[0021] In this process, due to the discrete logarithm problem of elliptic curves (i.e., given the base point G and point P) n It is difficult to calculate n), even if a third party obtains P. n The coordinates cannot be used to deduce the original numerical label n, thus failing to encrypt the entity attribute identifier. Feature component extraction and coordinate splitting: After completing the dot product operation, from the encrypted coordinate point P n (x) n y n Extract two key values ​​from the data: the x-coordinate of the point in the elliptic curve coordinate system. n and the vertical coordinate y n These two values ​​are the encrypted feature components. x needs to be verified during extraction. n and y n Whether it conforms to the preset finite field range (i.e., ensuring the value is between 0 and p-1, where p is the modulus in the aforementioned SM2 standard). If it exceeds the range, it is adjusted through modulo operation to ensure the validity of the feature components. Due to the characteristics of elliptic curves, different numerical labels n correspond to different encrypted coordinate points P. n The x and y coordinates have an extremely low probability of repetition, therefore x n and y n It can serve as a unique feature to distinguish different entity attribute identifiers.

[0022] 3D spatial coordinate binding and construction: The extracted feature components (x) n y n The knowledge graph version identifier (e.g., "2024v3.2" corresponds to the numerical label (n) of the original entity attribute identifier and the version identifier of the knowledge graph) are associated and bound together. The specific steps are as follows: The X-axis corresponds specifically to the original numerical label n of the entity attribute identifier (e.g., 1001), and the Y-axis corresponds to the x-coordinate of the encrypted coordinate point. n The Z-axis corresponds to the y-coordinate of the encrypted coordinate point. n To ensure traceability of binding relationships, a version identifier mapping value is embedded in the coordinate data (e.g., storing 32 as an additional parameter in the coordinate metadata), making each 3D coordinate (n, x...)... n， y n Each of these can be mapped to a specific version of the knowledge graph; the resulting multidimensional feature vector contains both the original entity's identification information (X-axis) and the encrypted features (Y-axis and Z-axis), and is associated with the knowledge graph's update status through version identification, forming structured data that is "identifiable, encryptable, and traceable".

[0023] By employing dot product operations in elliptic curve cryptography, entity attribute identifiers are transformed into irreversible encrypted coordinates, effectively preventing the original identifiers from being maliciously cracked or leaked, thus protecting the privacy and security of medical data (such as patient-related case identifiers and private entity information). The generated 3D feature vector binds the encrypted feature components to the original identifier and version information, transforming unstructured entity attribute data into quantifiable spatial coordinates. The uniqueness of elliptic curve cryptography ensures that each entity attribute identifier corresponds to a unique feature vector. Combined with knowledge graph version identifiers, the correspondence between feature vectors and knowledge graph update states can be accurately traced, providing support for data verification and version management.

[0024] In a preferred embodiment of the present invention, constructing a data distribution model based on the spatial coordinates of a multidimensional feature vector, and extracting the local density concentration characteristic value and boundary dispersion characteristic value of the data distribution model, may include: Group the multidimensional feature vectors by entity attribute identifier and perform the following operations: Calculate the average value of each set of feature vectors in the YZ plane coordinates to generate the center point of the core density region; Feature vectors whose distance from the center point of the core density region of their respective group exceeds a preset threshold are selected as boundary vectors to be verified. When the boundary vector to be verified is located in the coordinate overlap region of different entity attribute groups, it is marked as a boundary feature vector; The average distance from the feature vector within a group to the center point of the core density region is calculated as the local density concentration characteristic value. The width of the angular distribution range of all boundary feature vectors relative to the center point of their respective core density regions is measured as the boundary dispersion characteristic value.

[0025] In this embodiment of the invention, all multidimensional feature vectors are grouped according to the type of entity attribute identifier (such as disease code, drug ID, etc.). For example, all feature vectors containing "pneumonia" related attribute identifiers are grouped into one group, and feature vectors containing "penicillin" related attribute identifiers are grouped into another group, ensuring that the feature vectors in each group correspond to the same type of entity attribute.

[0026] Calculate the center point of the core density region: For each group, focus on the YZ plane coordinates of the feature vectors (i.e., the encrypted horizontal and vertical coordinates); iterate through the Y-axis and Z-axis values ​​of all feature vectors within the group, calculate the average of the Y-axis coordinates and the average of the Z-axis coordinates respectively, and combine these two averages into a new coordinate point, which is the center point of the core density region of the group. For example, if the Y-axis coordinates of the feature vectors in a group are 5, 7, and 9, and the Z-axis coordinates are 3, 5, and 7, then the average Y-axis value is 7, the average Z-axis value is 5, and the center point of the core density region is (7, 5). This point represents the concentrated distribution center of the feature vectors in the YZ plane within the group.

[0027] Filter the boundary vectors to be verified: A preset distance threshold is set for each group (this threshold is determined based on the characteristic distribution of medical data, such as by statistically analyzing the dispersion of historical data). The straight-line distance from each feature vector within the group to the center point of the core density region is calculated one by one. If the distance of a feature vector exceeds the threshold, it is marked as a boundary vector to be verified. For example, if the threshold is set to 4, and the distance of a feature vector to the center point (7, 5) is 5, then this vector is included in the range to be verified and is initially determined to be possibly located in the edge region of the data distribution.

[0028] Marking boundary feature vectors: For all boundary vectors to be verified, check whether their YZ plane coordinates are located in the overlapping region of coordinates from different entity attribute groups. Specifically, check whether the vector's coordinates simultaneously fall within the periphery of the core density region of other groups (e.g., simultaneously close to the core density regions of the "Pneumonia" group and the "Respiratory Infection" group). If this condition is met, the boundary vector to be verified is formally marked as a boundary feature vector, indicating that it is located at the intersection of different entity attribute data and has the potential characteristic of cross-group association.

[0029] Calculate the local density concentration characteristic value: For each group, iterate through all feature vectors within the group (including non-boundary vectors and boundary vectors to be verified), and calculate the straight-line distance from each vector to the center point of the core density region. Then, sum these distances and divide by the total number of feature vectors in the group. The average value obtained is the local density concentration characteristic value of the group. This value reflects the overall degree of clustering of the data within the group. For example, if the average value is small, it indicates that most feature vectors are closely distributed around the center point, and the data concentration is high.

[0030] Measure the boundary dispersion characteristics: For all marked boundary feature vectors within each group, using the center point of the core density region as the reference point, calculate the orientation angle of each boundary feature vector relative to the center point (i.e., the angle between the ray pointing from the center point to the boundary vector and the positive Y-axis). After collecting all these orientation angles, find the maximum and minimum values; the difference between the two is the width of the orientation angle distribution range, which is the boundary dispersion characteristic value of the group. For example, if the minimum orientation angle of a group of boundary feature vectors is 30° and the maximum is 150°, then the distribution range width is 120°, reflecting the degree of spatial dispersion of the boundary data.

[0031] By employing group analysis and core density region calculation, the system clearly presents the concentrated distribution pattern of data belonging to the same type. Simultaneously, through boundary vector labeling and orientation angle analysis, it accurately captures the dispersion characteristics of the boundary regions between different data categories. Local density concentration characteristics intuitively reflect the compactness of data within a group, helping to quickly identify the core features of entities of the same type. Boundary dispersion characteristics, on the other hand, reflect the marginal relationships between data across groups, providing support for discovering potential connections between different entity attributes. These two characteristics, as quantitative indicators, can be directly used in subsequent steps such as centroid optimization and correlation assessment, ensuring that the system can dynamically adjust its analysis strategy based on the actual distribution characteristics of the data, thereby improving overall processing accuracy.

[0032] In a preferred embodiment of the present invention, calculating the initial center point of the multidimensional feature vector, generating center calibration parameters based on local density concentration characteristics and boundary dispersion characteristics, and dynamically adjusting the initial center point using the center calibration parameters to generate an optimized center point may include: Use the center point of the core density region as the initial center point; use the local density concentration characteristic value to generate the distance adjustment amount; use the boundary dispersion characteristic value to generate the direction adjustment amount; calculate the average direction angle of the boundary feature vector to generate the moving direction reference; Adjust the movement direction reference by the direction adjustment amount to generate the final movement direction; adjust the initial center point coordinates by the distance adjustment amount to generate intermediate calibration points; Displace the intermediate calibration point along the final moving direction, and output the coordinates after displacement as the optimization center point.

[0033] In this embodiment of the invention, the core density region center point calculated above is used as the initial center point. For example, if the core density region center point of a certain set of entity attribute data is (7, 5) (YZ plane coordinates), then this point is the initial center point, serving as the reference starting point for subsequent adjustments.

[0034] Generate distance adjustment amount: If the local density concentration characteristic value is small (indicating a high degree of data concentration within the group), the distance adjustment should be set to a small value (e.g., 10% of the characteristic value) to avoid over-adjustment that could cause deviation from the core area. If the characteristic value is large (indicating dispersed data), the distance adjustment should be set to a larger value (e.g., 30% of the characteristic value) to improve the representativeness of the center point for the dispersed data through a larger adjustment. For example, when the local density concentration characteristic value is 2, the distance adjustment can be set to 0.2; when the characteristic value is 8, the adjustment can be set to 2.4.

[0035] Generate direction adjustment amount: If the boundary dispersion characteristic value is small (indicating that the boundary vector directions are concentrated), the direction adjustment amount should be small (e.g., 5% of the characteristic value) to reduce interference with the movement direction; if the characteristic value is large (indicating that the boundary vector directions are dispersed), the direction adjustment amount should be large (e.g., 20% of the characteristic value) to make the movement direction more consistent with the overall distribution of the boundary data. For example, when the boundary dispersion characteristic value is 30°, the direction adjustment amount can be set to 1.5°; when the characteristic value is 120°, the adjustment amount can be set to 24°.

[0036] Calculate the reference for the direction of movement: Iterate through all boundary feature vectors and calculate the direction angle of each vector relative to the initial center point (i.e., the angle between the ray pointing from the center point to the vector and the positive Y-axis). Then calculate the average of all direction angles, which is the reference for the movement direction. For example, if the direction angles of the three boundary vectors are 30°, 60°, and 90°, the average is 60°, meaning the reference for the movement direction is 60°.

[0037] Determine the final direction of movement: The final movement direction is obtained by adding or subtracting the direction adjustment amount from the reference direction (the addition or subtraction is selected based on the overall distribution trend of the boundary vectors: if most boundary vectors are located to the left of the reference direction, subtract the adjustment amount; if most are located to the right, add the adjustment amount). For example, if the reference direction is 60°, the direction adjustment amount is 24°, and most boundary vectors are to the right of the reference direction, then the final movement direction is 60° + 24° = 84°.

[0038] Generate intermediate calibration points: Based on the angle of the final movement direction, the distance adjustment is decomposed into displacement components on the Y and Z axes (e.g., when the angle is 84°, the Y-axis component is approximately the distance adjustment multiplied by cos84°, and the Z-axis component is approximately the distance adjustment multiplied by sin84°). The intermediate calibration point is obtained by adding the Y-axis component to the initial center point's Y-axis coordinate and the Z-axis component to the Z-axis coordinate. For example, if the initial center point is (7,5), the distance adjustment is 2.4, and the final movement direction of 84° corresponds to a Y-axis component of approximately 0.25 and a Z-axis component of approximately 2.39, then the intermediate calibration point is (7 + 0.25, 5 + 2.39) = (7.25, 7.39).

[0039] Generate optimization center points: Starting from the intermediate calibration point, move the device a certain distance along the final movement direction (84°) (this distance is 50% of the distance adjustment amount, used for fine-tuning). The coordinates after the displacement are the optimization center point. For example, if the intermediate calibration point is (7.25, 7.39), the displacement distance is 1.2 (50% of 2.4), the decomposed Y-axis component is approximately 0.12, and the Z-axis component is approximately 1.19. Then the optimization center point is (7.25 + 0.12, 7.39 + 1.19) = (7.37, 8.58).

[0040] By dynamically adjusting the center point based on local density and boundary dispersion characteristics, the optimized center point reflects both the central tendency of data within a group and the distribution characteristics of boundary data, thus more accurately representing the overall data distribution. Differential settings for distance and direction adjustments can adapt to datasets with varying densities and dispersion states, avoiding biases caused by a "one-size-fits-all" approach. The optimized center point serves as a benchmark for correlation evaluation, improving the accuracy of geometric distance calculations for entity data and consequently enhancing the precision of learning path recommendations.

[0041] In a preferred embodiment of the present invention, based on the optimization center point, the geometric distance from each entity data point to the optimization center point is calculated and sorted by distance. Then, combining the medical feature weight matrix and a preset evaluation function, a correlation degree value is output. A differentiated learning path recommendation strategy is generated using a meta-learning algorithm, which may include: Calculate the geometric distance from each entity data point in the medical knowledge graph to the optimization center point, and sort them in ascending order of distance to generate an ordered entity sequence. Based on the medical type identifier of the entities in the ordered entity sequence, match the corresponding weight value in the medical feature weight matrix to generate a weighted entity set; Input the weighted entity set and entity geometric distance into a preset evaluation function, and output the correlation degree value for each entity data, specifically including: Extract the original geometric distance of each decomposed data in the ordered data sequence, and calculate the distribution interval of all original geometric distances; Extract the original geometric distance of each entity in the ordered entity sequence and calculate the distribution interval of all distances; Each original geometric distance is mapped to a preset standard interval to generate a standardized distance value; Obtain the initial weight values ​​for entity types in the weighted entity set, and dynamically adjust the weights based on the standardized distance values ​​to generate corrected weight values; Combined modified weight values ​​and standardized distance values ​​to generate synthetic parameters, detect the degree of matching between synthetic parameters and medical association rules, and determine the basic association degree level; Adjust the basic relevance level by combining the medical type identifier of the entity, and output the relevance value of each entity data; Evaluation results are generated based on the numerical values ​​of all entity associations, and a final learning path recommendation strategy is generated by combining the meta-learning algorithm, specifically including: Receive the final set of correlation scores and set a first threshold interval and a second threshold interval, wherein the lower limit of the first threshold interval is greater than the upper limit of the second threshold interval; The learning stages are divided according to the threshold range to which the correlation value belongs, and the learning stages include: Entity data with relevance values ​​within the first threshold range constitute the first-stage learning unit; Entity data with correlation scores within the second threshold range constitute the second-stage learning units; Entity data with correlation scores below the lower limit of the second threshold range constitute the third-stage learning unit; A dependency graph of three-stage learning units is established using a meta-learning algorithm to generate a chain-like learning path recommendation strategy, where the first-stage learning unit serves as the starting point of the path; the second-stage learning unit is unlocked after the first-stage learning unit is completed; and the third-stage learning unit is unlocked after the second-stage learning unit is completed.

[0042] In this embodiment of the invention, for each entity data in the medical knowledge graph, the straight-line distance (i.e., geometric distance) from its YZ plane coordinates (encrypted feature components) to the optimization center point is calculated based on the optimization center point. For example, if the coordinates of an entity data are (8, 9) and the optimization center point is (7.37, 8.58), the straight-line length between the two points is calculated using a geometric method as the distance value. After completing the distance calculation for all entity data, the entity data are sorted in ascending order of distance value to form an ordered entity sequence, ensuring that the closest entity is placed at the beginning and the farthest entity is placed at the end.

[0043] Generate a weighted set of entities by matching weight values: The system pre-stores a medical feature weight matrix. This matrix assigns corresponding weight values ​​to different types of entities (such as diseases, drugs, surgical procedures, etc.) based on the relevance and importance of medical knowledge (e.g., a core disease is weighted at 0.8, and an auxiliary drug at 0.5). The system iterates through the ordered entity sequence, extracting the medical type identifier for each entity (e.g., "pneumonia" belongs to the respiratory system disease type). Based on the identifier, the system matches the corresponding weight value in the weight matrix, associating the entity data with the matched weight value to form a weighted entity set. For example, in the ordered sequence, the entity "pneumonia" matches a weight of 0.8, and the entity "penicillin" matches a weight of 0.6; both are included in the set as "entity + weight".

[0044] Calculate the correlation coefficient: Extract the original geometric distances of all entities in the ordered entity sequence, calculate the minimum and maximum values ​​of these distances, and determine the distance distribution range (e.g., minimum distance is 0.5, maximum is 10, the range is 0.5-10). Map the original distance of each entity proportionally to a preset standard range (e.g., 0-1). Obtain the initial weight value for each entity type from the weighted entity set, and dynamically adjust it based on its standardized distance value: if the standardized distance value is small (the entity is close to the optimization center point), increase it appropriately based on the initial weight (e.g., increase by 10%); if the standardized distance value is large, decrease it appropriately (e.g., decrease by 10%). For example, an entity with an initial weight of 0.8, if the standardized distance value is 0.3 (relatively close), the adjusted weight is 0.88; if the standardized distance value is 0.7 (relatively far), the adjusted weight is 0.72.

[0045] The modified weight value and the standardized distance value are combined according to preset rules (such as the proportional relationship between the two) to generate composite parameters. These composite parameters are then compared with medical association rules (such as the association rules between diseases and symptomatic drugs, and the association rules between surgical procedures and instruments). Based on the degree of matching, a basic association level is assigned (e.g., a perfect match is level 5, a partial match is level 3, and a weak match is level 1). For example, the composite parameters of the entities "pneumonia" and "penicillin" conform to the "association rules between bacterial infections and antibiotics," resulting in a basic level of 4. The basic association level is then fine-tuned by combining the entity's medical type identifier: if the entity belongs to a core knowledge point (such as a basic disease mechanism), the level increases by 1 level; if it belongs to an extended knowledge point (such as rare case variants), the level decreases by 1 level. Finally, the level is converted into a specific numerical value (e.g., level 5 corresponds to 90-100, and level 4 corresponds to 70-89), which is the entity's association value.

[0046] Divide the learning process into stages and generate learning paths: The system receives the correlation scores of all entities and sets a first threshold range (e.g., 70-100) and a second threshold range (e.g., 40-69), where the lower limit (70) of the first threshold range is greater than the upper limit (69) of the second threshold range. Entities with correlation scores falling within the first range are assigned to the first-stage learning unit (e.g., "Etiology and Diagnostic Criteria of Pneumonia"), those falling within the second range are assigned to the second-stage learning unit (e.g., "Drug Treatment Plan for Pneumonia"), and those below the lower limit (40) of the second range are assigned to the third-stage learning unit (e.g., "Management of Complications and Rehabilitation Nursing for Pneumonia"). The system analyzes the dependencies between the three-stage learning units using a meta-learning algorithm (e.g., "Understanding the etiology is necessary to comprehend the treatment plan" and "Understanding the treatment plan is necessary to learn how to manage complications") and constructs a dependency graph. Based on this graph, a chain path is generated: starting from the first-stage unit, the system automatically unlocks the second-stage unit after the student completes all first-stage units; after completing the second-stage unit, the third-stage unit is unlocked, forming a progressive learning path of "basic → advanced → extended".

[0047] By calculating geometric distance, dynamically adjusting weights, and matching medical rules, the correlation coefficient reflects both the spatial relationship between entities and optimization centers and the logical connections within medical knowledge, providing a reliable basis for dividing learning units. The stage division based on correlation coefficients and dependency analysis of the meta-learning algorithm ensures that the learning path follows a "from shallow to deep, step-by-step" cognitive pattern, while also adapting to different learners' knowledge levels (e.g., prioritizing core content with high correlation). A chain-like unlocking mechanism encourages learners to complete learning in a logical order, avoiding knowledge fragmentation; the three-stage division clearly defines learning priorities, helping learners allocate their energy effectively and improve learning efficiency.

[0048] In a preferred embodiment of the present invention, multidimensional feature vectors are allocated to corresponding storage nodes based on their correlation scores; the multidimensional feature vectors are encrypted using an asymmetric encryption algorithm; the encrypted feature vectors are bound to their correlation scores, knowledge graph version identifiers, and timestamps to construct encrypted data blocks, which are then written into the blockchain, including: Extract the learning stage sequence from the chained learning path recommendation strategy and allocate storage nodes according to the order of the learning stages. Specifically, the first stage learning unit is allocated to the priority storage node; the second stage learning unit is allocated to the secondary storage node; and the third stage learning unit is allocated to the regular storage node. Asymmetric encryption key pairs are generated for each type of storage node. Entity data is encrypted using the node's public key, and learning stage identifiers, knowledge graph version identifiers, and timestamp data are bound together. Encrypted data blocks are constructed and written to the blockchain in order of learning stage priority.

[0049] In this embodiment of the invention, a clear sequence of learning stages is extracted from the chained learning path recommendation strategy, namely, a list of learning units for the first, second, and third stages (e.g., the first stage includes core knowledge points such as "the basic causes of pneumonia," the second stage includes advanced content such as "drug treatment of pneumonia," and the third stage includes extended content such as "management of rare complications of pneumonia"). Subsequently, corresponding storage nodes are assigned to the multidimensional feature vectors of each stage according to their priority. The feature vectors of the first-stage learning units are assigned to priority storage nodes. These nodes have faster read and write speeds and higher resource priority, ensuring that students can quickly access core content in the early stages of learning. The feature vectors of the second-stage learning units are allocated to secondary storage nodes, which have slightly lower performance than the priority nodes, but can still meet the access requirements of advanced learning. The feature vectors of the third-stage learning units are assigned to regular storage nodes, which is suitable for extended content with low access frequency, thus achieving a reasonable allocation of storage resources.

[0050] Generate asymmetric encryption key pairs and encrypt entity data: Independent asymmetric encryption key pairs are generated for priority storage nodes, secondary storage nodes, and regular storage nodes. Each key pair contains a public key (used for encrypting data and publicly available) and a private key (used for decrypting data and strictly guarded by the system). For the entity data (i.e., multi-dimensional feature vectors) in each storage node, the public key corresponding to that node is used for encryption. During encryption, the system converts the original data of the feature vector (including entity attribute identifiers on the X-axis and encrypted feature components on the Y and Z axes) into encrypted text. Only the private key of that node can decrypt and restore the original data, ensuring the security of data during storage and transmission.

[0051] Bind multi-dimensional information to the encrypted feature vector: Relevance score: This is the relevance score of the entity data in the learning path (e.g., 85 points), used to identify its importance in the learning sequence; Knowledge graph version identifier: such as "2024v3.2", which records the knowledge graph version corresponding to the entity data to ensure that the data and knowledge are updated in sync; Timestamp: An encrypted operation time record accurate to the second (e.g., "2024-07-08 15:30:22"), used to trace the time when the data was generated.

[0052] The encrypted feature vector is integrated with the above three pieces of information into an indivisible encrypted data block. Each data block contains complete entity information, associated attributes, and a time trace. The blockchain system uses a chain structure to store data blocks, and writes follow the priority order of the learning phase: Encrypted data blocks (first-stage learning units) in priority storage nodes are written to the front-end blocks of the blockchain first, ensuring rapid verification and access to core content; The data blocks in the secondary storage nodes (second-stage learning units) are then written into the intermediate blocks; Data blocks (third-stage learning units) in regular storage nodes are written into subsequent blocks.

[0053] Each data block must be verified for its legitimacy through the blockchain's consensus mechanism (such as node verification) when it is written, and it cannot be tampered with once written, while retaining a complete chain index.

[0054] Storage nodes are allocated based on learning stage priority, ensuring core content occupies high-performance storage resources while supplementary content uses conventional resources, balancing access efficiency and cost. Asymmetric encryption ensures that entity data can only be decrypted using a dedicated private key. Combined with the immutability of blockchain, this effectively prevents the leakage or tampering of sensitive medical data (such as case characteristics and encrypted identifiers). By binding knowledge graph version identifiers and timestamps, the generation background and corresponding knowledge version of each data block can be accurately traced, ensuring that learning content is synchronized with the latest medical knowledge. Data is written to the blockchain in the order of learning stages, making the storage structure highly compatible with the learning path. When accessing content, learners can quickly locate the content needed for the current stage through the blockchain index, enhancing the learning experience.

[0055] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0056] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system as described above. All implementations in the above system embodiments are applicable to this embodiment and can achieve the same technical effects.

[0057] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An online learning system for medical colleges based on big data analysis, characterized in that, include: The knowledge graph construction module is used to preprocess and extract multi-dimensional features from the raw learning data uploaded to the online learning environment of medical colleges and universities, and generate a dynamically updated medical knowledge graph. The vector transformation module is used to use elliptic curve cryptography to encrypt and decompose entity attribute identifiers in the medical knowledge graph, generating multidimensional feature vectors with three-dimensional spatial coordinates. The data extraction module is used to construct a data distribution model based on the spatial coordinates of multidimensional feature vectors, and to extract the local density concentration characteristic value and boundary dispersion characteristic value of the data distribution model. The center point optimization module is used to calculate the initial center point of the multidimensional feature vector, generate center calibration parameters based on the local density concentration characteristic value and the boundary dispersion characteristic value, and dynamically adjust the initial center point through the center calibration parameters to generate the optimized center point; The correlation evaluation module is used to calculate the geometric distance of each entity data to the optimization center point based on the optimization center point, sort them by distance, and output the correlation value by combining the medical feature weight matrix and the preset evaluation function; and generate a differentiated learning path recommendation strategy by combining meta-learning algorithm. The data storage module is used to allocate multidimensional feature vectors to corresponding storage nodes based on the correlation degree value; The multidimensional feature vectors are encrypted using an asymmetric encryption algorithm; the encrypted feature vectors are then bound to the correlation degree value, the knowledge graph version identifier, and the timestamp to construct an encrypted data block, which is then written into the blockchain.

2. The online learning system for medical colleges based on big data analysis according to claim 1, characterized in that, Elliptic curve cryptography is used to encrypt and decompose entity attribute identifiers in a medical knowledge graph, generating multidimensional feature vectors with three-dimensional spatial coordinates, including: Treat the entity attribute identifiers in the medical knowledge graph as independent data units, and perform the following operations on each data unit: Select a base point on the elliptic curve, and perform a dot product operation using the value of the entity attribute identifier as a multiple of the private key to generate encrypted coordinate points; Extract the horizontal and vertical coordinates of the encrypted coordinate points as feature components; The feature components are bound to entity attribute identifiers and knowledge graph version identifiers to form a three-dimensional spatial coordinate system, where the X-axis is the entity attribute identifier, the Y-axis is the horizontal coordinate of the feature components, and the Z-axis is the vertical coordinate of the feature components.

3. The online learning system for medical colleges based on big data analysis according to claim 2, characterized in that, A data distribution model is constructed based on the spatial coordinates of multidimensional feature vectors. Local density concentration characteristics and boundary dispersion characteristics of the data distribution model are extracted, including: Group the multidimensional feature vectors by entity attribute identifier and perform the following operations: Calculate the average value of each set of feature vectors in the YZ plane coordinates to generate the center point of the core density region; Feature vectors whose distance from the center point of the core density region of their respective group exceeds a preset threshold are selected as boundary vectors to be verified. When the boundary vector to be verified is located in the coordinate overlap region of different entity attribute groups, it is marked as a boundary feature vector; The average distance from the feature vector within a group to the center point of the core density region is calculated as the local density concentration characteristic value. The width of the angular distribution range of all boundary feature vectors relative to the center point of their respective core density regions is measured as the boundary dispersion characteristic value.

4. The online learning system for medical colleges based on big data analysis according to claim 3, characterized in that, Calculate the initial center point of the multidimensional feature vector, generate center calibration parameters based on local density concentration characteristics and boundary dispersion characteristics, and dynamically adjust the initial center point using the center calibration parameters to generate optimized center points, including: Use the center point of the core density region as the initial center point; use the local density concentration characteristic value to generate the distance adjustment amount; use the boundary dispersion characteristic value to generate the direction adjustment amount; calculate the average direction angle of the boundary feature vector to generate the moving direction reference; Adjust the movement direction reference by the direction adjustment amount to generate the final movement direction; adjust the initial center point coordinates by the distance adjustment amount to generate intermediate calibration points; Displace the intermediate calibration point along the final moving direction, and output the coordinates after displacement as the optimization center point.

5. The online learning system for medical colleges based on big data analysis according to claim 4, characterized in that, Based on the optimization center point, calculate the geometric distance of each entity data to the optimization center point and sort them by distance. Combine the medical feature weight matrix and the preset evaluation function to output the correlation degree value. Combining meta-learning algorithms to generate differentiated learning path recommendation strategies includes: Calculate the geometric distance from each entity data in the medical knowledge graph to the optimization center point, and sort them in ascending order of distance to generate an ordered entity sequence; Based on the medical type identifier of the entities in the ordered entity sequence, match the corresponding weight value in the medical feature weight matrix to generate a weighted entity set; Input the weighted entity set and the geometric distance between entities into a preset evaluation function, and output the correlation degree value of each entity data; Evaluation results are generated based on the numerical values ​​of all entity associations, and the final learning path recommendation strategy is generated by combining meta-learning algorithms.

6. The online learning system for medical colleges based on big data analysis according to claim 5, characterized in that, Input the weighted entity set and entity geometric distance into a preset evaluation function, and output the correlation degree value of each entity data, including: Extract the original geometric distance of each decomposed data in the ordered data sequence, and calculate the distribution interval of all original geometric distances; Extract the original geometric distance of each entity in the ordered entity sequence and calculate the distribution interval of all distances; Each original geometric distance is mapped to a preset standard interval to generate a standardized distance value; Obtain the initial weight values ​​for entity types in the weighted entity set, and dynamically adjust the weights based on the standardized distance values ​​to generate corrected weight values; Combined modified weight values ​​and standardized distance values ​​to generate synthetic parameters, detect the degree of matching between synthetic parameters and medical association rules, and determine the basic association degree level; Adjust the basic relevance level by combining the entity's medical type identifier, and output the relevance value of each entity's data.

7. The online learning system for medical colleges based on big data analysis according to claim 6, characterized in that, Evaluation results are generated based on the numerical values ​​of all entity associations, and a final learning path recommendation strategy is generated by combining the meta-learning algorithm, including: Receive the final set of correlation scores and set a first threshold interval and a second threshold interval, wherein the lower limit of the first threshold interval is greater than the upper limit of the second threshold interval; The learning stages are divided according to the threshold range to which the correlation value belongs, and the learning stages include: Entity data with relevance values ​​within the first threshold range constitute the first-stage learning unit; Entity data with correlation scores within the second threshold range constitute the second-stage learning units; Entity data with correlation scores below the lower limit of the second threshold range constitute the third-stage learning unit; A dependency graph of three-stage learning units is established using a meta-learning algorithm to generate a chain-like learning path recommendation strategy, where the first-stage learning unit serves as the starting point of the path; the second-stage learning unit is unlocked after the first-stage learning unit is completed; and the third-stage learning unit is unlocked after the second-stage learning unit is completed.

8. The online learning system for medical colleges based on big data analysis according to claim 7, characterized in that, The multidimensional feature vectors are assigned to corresponding storage nodes based on the correlation degree values; the multidimensional feature vectors are then encrypted using an asymmetric encryption algorithm. The encrypted feature vector is bound to the correlation value, knowledge graph version identifier, and timestamp to construct an encrypted data block, which is then written to the blockchain, including: Extract the learning stage sequence from the chained learning path recommendation strategy and allocate storage nodes according to the order of the learning stages. Specifically, the first stage learning unit is allocated to the priority storage node; the second stage learning unit is allocated to the secondary storage node; and the third stage learning unit is allocated to the regular storage node. Asymmetric encryption key pairs are generated for each type of storage node. Entity data is encrypted using the node's public key, and learning stage identifiers, knowledge graph version identifiers, and timestamp data are bound together. Encrypted data blocks are constructed and written to the blockchain in order of learning stage priority.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the system as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the system as described in any one of claims 1 to 8.

Images