Cloud service-based innovative entrepreneurship education resource sharing and collaborative education system

Abstract

The application relates to the technical field of informationization of educational resources, in particular to an innovative and entrepreneurial education resource sharing and collaborative education system based on cloud services, which comprises the following steps: collecting original educational resources of multiple colleges and universities and original industrial resources of multiple enterprises through a resource acquisition module, classifying the educational resources by using an improved clustering algorithm, and constructing a classified educational resource library; completing industrial resource association processing by using an improved association rule mining algorithm, and constructing an associated industrial resource library; performing cross-database index mapping on the two types of resource libraries by a cloud server, and generating a total index of collaborative education resources; building a multi-tenant access interface based on the total index, and distinguishing three types of user access permissions; the system receives a user search request and completes resource matching, and generates a personalized recommendation list; the application optimizes cloud resource processing logic, improves the problems of scattered resources and weak association, realizes differentiated resource access of multiple users, and adapts to the use scene of innovative and entrepreneurial collaborative education.

Description

Technical Field

[0001] This invention relates to the field of information technology for educational resources, and in particular to a cloud-based collaborative education system for sharing innovative entrepreneurship educational resources. Background Technology

[0002] Currently, innovation and entrepreneurship education resources and industry practice resources mostly rely on cloud storage platforms for data collection. Different universities and independent enterprises upload various educational materials and industry training materials separately, resulting in scattered data sources and heterogeneous data formats within the cloud resource pool. General clustering algorithms applied to educational resource processing can only perform simple classification based on surface-level labels, failing to adapt to the multi-dimensional and fragmented textual characteristics of innovation and entrepreneurship education resources. Heterogeneous educational resources are mixed together, and similar high-quality teaching resources are stored in a scattered manner, resulting in disordered resource arrangement.

[0003] When general association rule mining algorithms process data from industry practice resources, they can only identify explicit relationships on the surface of resources, making it difficult to capture hidden business linkages between industry resources. Data relationships within industry resources are fragmented, and resources from different industries and training types cannot form a collaborative structure. There is a lack of compatible index mapping mechanisms between cloud databases; educational resource databases and industry resource databases are independent, resulting in poor data interoperability. General cloud platforms have a simplistic access permission system, failing to adapt to the differentiated needs of universities, enterprises, and students; resource matching logic is simplistic, and the relevance of search and push content is low. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system, comprising: The resource acquisition module acquires original innovation and entrepreneurship education resource sets uploaded by multiple universities and original industry practice resource sets uploaded by multiple enterprises. The classification and clustering module performs resource classification processing based on an improved clustering algorithm on the original set of innovation and entrepreneurship education resources to obtain a classified education resource library; The association mining module performs resource association processing based on an improved association rule mining algorithm on the original set of industry practice resources to obtain an association industry resource library; The index mapping module uploads the categorized educational resource database and the associated industry resource database to the cloud server for cross-database index mapping processing to generate a general index of collaborative education resources. The multi-tenant interface module constructs a multi-tenant access interface based on the overall index of collaborative education resources. The multi-tenant access interface is used to provide differentiated resource access permissions to university users, enterprise users, and student users. The recommendation matching module receives resource retrieval requests from target users through the multi-tenant access interface, performs resource matching processing based on the overall index of collaborative education resources, and generates a personalized resource recommendation list for the target user.

[0006] As a further aspect of the present invention, the step of performing resource classification processing based on an improved clustering algorithm on the original set of innovation and entrepreneurship education resources to obtain a classified education resource library specifically includes: The improved clustering algorithm initializes cluster centers based on the joint distribution density of resource type labels and subject labels; Scan the metadata of each resource file in the original set of innovation and entrepreneurship education resources, and extract the resource type tag set and subject tag set of each resource file; Calculate the frequency distribution of each resource type tag in the resource type tag set appearing in the original innovation and entrepreneurship education resource set, and calculate the frequency distribution of each subject tag in the subject tag set appearing in the original innovation and entrepreneurship education resource set. A set of candidate cluster center seeds is generated based on the Cartesian product of the top several resource type labels with the highest frequency in the resource type frequency distribution and the top several subject labels with the highest frequency in the subject frequency distribution. For each candidate cluster center seed in the candidate cluster center seed set, the joint distribution density of the resource type label and the subject label corresponding to the candidate cluster center seed in the original innovation and entrepreneurship education resource set is calculated, and the candidate cluster center seed with the largest joint distribution density is taken as the first initial cluster center. After deleting the resource files corresponding to the candidate cluster center seeds that have been selected as cluster centers, repeat the operation of calculating the joint distribution density and selecting the candidate cluster center seeds corresponding to the maximum joint distribution density until the number of initial cluster centers selected reaches the preset cluster number threshold. Starting from all the selected initial cluster centers, calculate the comprehensive similarity between each resource file in the original set of innovation and entrepreneurship education resources and each initial cluster center, and classify each resource file into the cluster of the initial cluster center with the highest comprehensive similarity, thus obtaining multiple resource clusters; Perform an intra-cluster center update operation on the resource files within each resource cluster, and use the updated intra-cluster center as the new cluster center. Repeat the classification operation and the intra-cluster center update operation until each resource cluster no longer changes. The resulting multiple resource clusters are named according to the common tags of the resource files within each cluster, and the multiple resource clusters and their naming results are stored as the classified educational resource library.

[0007] As a further aspect of the present invention, the comprehensive similarity is calculated by weighting the resource type label similarity, subject label similarity, and resource file vector space model cosine similarity.

[0008] As a further aspect of the present invention, the step of performing resource association processing based on an improved association rule mining algorithm on the original industrial practice resource set to obtain an associated industrial resource library specifically includes: The improved association rule mining algorithm calculates the support of itemsets based on a weighted average of resource usage frequency and resource update time. Analyze the resource usage log of each industry practice resource file in the original industry practice resource set, and extract the access frequency sequence and last update timestamp of each industry practice resource file in multiple historical time periods from the resource usage log; The average access frequency of each industry practice resource file is calculated based on the access frequency sequence, and the time decay coefficient of each industry practice resource file is calculated based on the last update timestamp. The time decay coefficient is inversely proportional to the time interval from the current moment to the last update timestamp. Multiply the average access frequency of each industry practice resource file by the corresponding time decay coefficient to obtain the dynamic weighted support base of each industry practice resource file; Scan all industry practice resource files in the original industry practice resource set, count the number of times any two industry practice resource files are accessed in the same user session, and divide the number of times they are accessed by the total number of user sessions to obtain the original co-occurrence support of the two industry practice resource files. Multiply the original co-occurrence support by the geometric mean of the dynamic weighted support base of the two industry practice resource documents to obtain the improved co-occurrence support of the two industry practice resource documents; All industry practice resource files whose co-occurrence support is greater than the preset minimum support threshold are paired and filtered out to form a frequent itemset candidate set; For each industry practice resource file in the frequent itemset candidate set, pairwise combinations are made. The ratio of the improved co-occurrence support of the combination to the dynamic weighted support cardinality of the first industry practice resource file in the combination is calculated to obtain the unidirectional association strength from the first industry practice resource file to the second industry practice resource file. The ratio of the improved co-occurrence support of the combination to the dynamic weighted support cardinality of the second industry practice resource file in the combination is also calculated to obtain the reverse unidirectional association strength. Industry practice resource files whose unidirectional association strength and reverse unidirectional association strength are both greater than the preset minimum confidence threshold are paired and marked as bidirectional strong association pairs; Using all bidirectional strong association pairs as edges and each industry practice resource file as a node, construct an industry practice resource association graph, and store the industry practice resource association graph as the associated industry resource library.

[0009] As a further aspect of the present invention, the time decay coefficient is calculated using an exponential decay function, and the time constant of the exponential decay function is dynamically determined by the technology update cycle of the industry field to which the industry practice resource file belongs.

[0010] As a further aspect of the present invention, the step of uploading the categorized educational resource database and the associated industry resource database to a cloud server for cross-database index mapping processing to generate a comprehensive index of collaborative education resources specifically includes: Assign a first storage domain identifier to each resource cluster in the classified educational resource library, and assign a first file identifier to each resource file within each resource cluster. Combine the first storage domain identifier and the first file identifier to generate a classified library index key for each resource file. Assign a second storage domain identifier to each industry practice resource file in the associated industry resource library, and keep the original file identifier of each industry practice resource file as the second file identifier. Combine the second storage domain identifier and the second file identifier to generate the association library index key for each industry practice resource file. Extract the resource title field and resource summary field of each resource file in the classified educational resource library, perform word segmentation on the resource title field and the resource summary field, and generate the first keyword set of each resource file; Extract the industry domain field and skill requirement field from each industry practice resource file in the associated industry resource library, perform word segmentation on the industry domain field and the skill requirement field, and generate a second keyword set for each industry practice resource file; Calculate the keyword overlap between the first keyword set and the second keyword set. When the keyword overlap is greater than a preset overlap threshold, record the mapping relationship between the classification library index key of the resource file and the association library index key of the industry practice resource file as a cross-library link entry. All cross-database link entries are aggregated to generate a cross-database mapping table, and all category database index keys of the categorized educational resource database are merged with all associated database index keys of the associated industry resource database to generate a total index key space. Each index key in the total index key space is associated with the corresponding resource file metadata and the cross-database link entries related to that index key in the cross-database mapping table, and the total index of collaborative education resources is generated.

[0011] As a further aspect of the present invention, the keyword overlap is the ratio of the number of elements in the intersection of the first keyword set and the second keyword set to the number of elements in the union.

[0012] As a further aspect of the present invention, a multi-tenant access interface is constructed based on the overall index of collaborative education resources. This multi-tenant access interface is used to provide differentiated resource access permissions to university users, enterprise users, and student users. Specifically, this includes: Read the first set of storage domain identifiers corresponding to the classified education resource database and the second set of storage domain identifiers corresponding to the associated industry resource database from the overall index of collaborative education resources. Configure a first permission matrix for university user types. The first permission matrix allows access to all resource files in the first storage domain identifier set and resource files associated with cross-database link entries in the second storage domain identifier set. Configure a second permission matrix for enterprise user types, which allows access to all resource files in the second storage domain identifier set and resource files in the first storage domain identifier set associated with cross-database link entries; Configure a third permission matrix for student user types. The third permission matrix allows access to resource files belonging to the preset innovation and entrepreneurship education level in the first storage domain identifier set, and to industry practice resource files belonging to the preset entry level in the second storage domain identifier set. The first permission matrix, the second permission matrix, and the third permission matrix are compiled into executable access control lists, and the access control lists are deployed in the interface gateway layer of the cloud server. A Uniform Resource Locator (URL) routing table is established at the interface gateway layer. The URL routing table is used to map resource identifiers in external access requests to the corresponding index keys in the overall index of collaborative education resources. The access control list and the uniform resource locator routing table are jointly encapsulated to generate the multi-tenant access interface.

[0013] As a further aspect of the present invention, the steps of receiving a resource retrieval request from a target user through the multi-tenant access interface, performing resource matching processing based on the collaborative education resource index, and generating a personalized resource recommendation list for the target user specifically include: The multi-tenant access interface receives resource retrieval requests sent by target users and parses the user type identifier, the university identifier to which the user belongs, and the sequence of search keywords entered by the user from the resource retrieval requests. The corresponding access control list is retrieved from the multi-tenant access interface based on the user type identifier. The access control list is used to perform permission filtering on the search keyword sequence, deleting keyword items in the search keyword sequence that exceed the access permissions of the target user, and generating a permission-filtered search keyword sequence. Each keyword in the search keyword sequence after permission filtering is fuzzily matched with the total index key space in the total index of collaborative education resources, and the text similarity between each keyword and the resource file metadata corresponding to each index key is calculated. For each resource file, the text similarity of all keywords corresponding to that resource file is weighted and summed to obtain the preliminary matching score of that resource file. The weight coefficient used in the weighted summation is determined according to the order of appearance of each keyword in the search keyword sequence after the permission filtering. The earlier the keyword appears, the larger the weight coefficient it corresponds to. Based on the user's university identifier, the set of disciplinary advantage tags of the target user's university is read from a preset university characteristic resource preference database; Resource files whose preliminary matching scores are greater than a preset matching threshold are selected to form a candidate resource set. For each resource file in the candidate resource set, the overlap between the subject tag of the resource file and the subject advantage tag set is calculated. The overlap between the tag and the preliminary matching score of the resource file is multiplied to obtain the final matching score of the resource file. The resource files in the candidate resource set are sorted in descending order of the final matching score. The top few resource files are selected to generate the personalized resource recommendation list, and the personalized resource recommendation list is returned to the target user through the multi-tenant access interface.

[0014] As a further aspect of the present invention, the text similarity is the semantic similarity between keywords and the corresponding metadata of index keys, calculated based on a pre-trained word vector model.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The improved clustering algorithm optimizes resource feature extraction, modifies clustering decision constraints, and adapts to the data attributes of multi-source heterogeneous innovation and entrepreneurship education resources. The algorithm masks invalid features in resource data, reducing the interference of redundant data on the resource classification process and broadening the classification dimensions of educational resources. Based on the content and application attributes of the resources themselves, it completes hierarchical division, aggregates scattered educational resources, unifies the storage layout standards of heterogeneous educational resources, standardizes the data storage structure within the resource repository, and reduces the data dispersion of similar resources.

[0016] The improved association rule mining algorithm adjusts the resource association calculation rules and corrects the criteria for frequent itemset selection, completing the data mining of implicit associations in industrial practice resources. The algorithm associates and binds industrial resources from different industries and training types, establishing fixed data association links and improving the data organization structure of industrial resources. Cross-database index mapping eliminates data isolation between two types of resource databases, opening up data interaction channels. Multi-tenant interfaces implement hierarchical access permissions, defining the data access scope for different user groups. The resource matching process relies on index data to complete logical operations, aligning with user search behavior characteristics and standardizing the data filtering logic for resource recommendations. Attached Figure Description

[0017] Figure 1 This is a sequence diagram of the cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system described in this invention; Figure 2 A flowchart for resource classification processing using an improved clustering algorithm; Figure 3 A flowchart for building a multi-tenant access interface. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] See Figure 1 This invention provides a cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system, specifically including: The resource acquisition module periodically synchronizes original innovation and entrepreneurship education resource sets and original industry practice resource sets from various university and enterprise systems connected to the alliance. These resource sets are uploaded to a cloud-based temporary storage area via a secure transmission channel in the form of structured data packets. The classification and clustering module utilizes cloud computing resources to perform resource classification processing on the original innovation and entrepreneurship education resource sets in the temporary storage area using an improved clustering algorithm, generating a stable classified education resource library through multiple iterative calculations. Simultaneously, the association mining module performs resource association processing on the original industry practice resource sets using an improved association rule mining algorithm, analyzing implicit connections between resources and constructing a related industry resource library. The index mapping module pushes the classified education resource library and the related industry resource library to the index service node of the cloud server, performing cross-library index mapping processing to establish a unified overall index for collaborative education resources. This overall index maintains the mapping relationship between resource logical addresses and physical storage locations. The multi-tenant interface module defines resource access control policies based on the overall index for collaborative education resources and constructs a multi-tenant access interface. This interface can dynamically load differentiated resource access permissions based on the user identity token of the request source. The recommendation matching module listens for resource retrieval requests from the multi-tenant access interface, calls the metadata and relationships in the collaborative education resource index in real time, performs resource matching processing, and generates and returns a personalized resource recommendation list for the target user.

[0020] In one embodiment of the present invention, see [reference] Figure 2When performing resource classification processing based on an improved clustering algorithm on the original set of innovation and entrepreneurship education resources, the system processor loads the improved clustering algorithm program, which initializes cluster centers based on the joint distribution density of resource type labels and subject labels. The system scans the metadata header of each resource file in the original set of innovation and entrepreneurship education resources, extracting the resource type label set and subject label set for each resource file. The resource type labels include, but are not limited to, courseware, videos, cases, and question banks, and the subject labels cover first-level and second-level subject codes. The system calculates the frequency distribution of each resource type label in the resource type label set and the frequency distribution of each subject label in the subject label set. Based on the Cartesian product of the top few most frequent resource type labels in the resource type frequency distribution and the top few most frequent subject labels in the subject frequency distribution, the system generates a set of candidate cluster center seeds, such as a combination of the high-frequency type "video" and the high-frequency subjects "Computer Science and Technology" and "Business Administration". For each candidate cluster center seed in the candidate cluster center seed set, the system calculates the joint distribution density of the resource type label and subject label corresponding to that candidate cluster center seed in the original innovation and entrepreneurship education resource set. The candidate cluster center seed with the highest joint distribution density is selected as the first initial cluster center. After deleting the resource files corresponding to the candidate cluster center seeds that have been selected as cluster centers, the system repeats the operation of calculating the joint distribution density and selecting the candidate cluster center seed corresponding to the highest joint distribution density until the number of selected initial cluster centers reaches the preset cluster number threshold K. Starting from all the selected initial cluster centers, the system calculates the comprehensive similarity between each resource file in the original innovation and entrepreneurship education resource set and each initial cluster center, and classifies each resource file into the cluster of the initial cluster center with the highest comprehensive similarity, resulting in multiple resource clusters. The overall similarity is calculated by weighting resource type label similarity, subject label similarity, and resource file vector space model cosine similarity. Resource type label similarity is obtained by comparing the Jaccard coefficient of the label set; subject label similarity is calculated by the path distance of the subject classification coding tree; and resource file vector space model cosine similarity is calculated using TF-IDF feature vectors. The system performs an intra-cluster center update operation on the resource files within each resource cluster, using the updated intra-cluster center as the new cluster center. This process is repeated until the resource clusters no longer change. The system names the resulting resource clusters according to the common labels of the resource files within them and stores the multiple resource clusters and their naming results in a categorized educational resource repository.

[0021] In its implementation, a cloud platform of a regional university innovation and entrepreneurship education alliance deployed a cloud-based collaborative education system for sharing innovation and entrepreneurship education resources. The original set of innovation and entrepreneurship education resources included 12,000 resource files from Universities A, B, and C. Resource type tags covered four categories: courseware, videos, case studies, and question banks. Subject tags covered four first-level disciplines: Computer Science and Technology, Business Administration, Mechanical Engineering, and Electronic Information. An improved clustering algorithm initialized cluster centers based on the joint distribution density of resource type tags and subject tags. The system scanned the metadata header of each resource file in the original set of innovation and entrepreneurship education resources, extracting the resource type tag set and subject tag set for each resource file. The resource type tag set recorded the format attributes of the resource file, and the subject tag set recorded the subject classification code to which the resource file belonged.

[0022] In some embodiments, the system calculates the frequency distribution of each resource type label in the resource type label set appearing in the original innovation and entrepreneurship education resource set, and calculates the frequency distribution of each subject label in the subject label set appearing in the original innovation and entrepreneurship education resource set. The resource type frequency distribution shows that courseware appears 4200 times, videos appear 3800 times, cases appear 2500 times, and question banks appear 1500 times. The subject frequency distribution shows that computer science and technology appears 3500 times, business administration appears 3200 times, mechanical engineering appears 2800 times, and electronic information appears 2500 times. The system generates a set of candidate cluster center seeds based on the Cartesian product of the top few resource type labels with the highest frequency in the resource type frequency distribution and the top few subject labels with the highest frequency in the subject frequency distribution. For example, if the top two resource type labels (courseware, video) and the top two subject labels (computer science and technology, business administration) with the highest frequency are selected, the Cartesian product will generate four candidate cluster center seeds: (courseware, computer science and technology), (courseware, business administration), (video, computer science and technology), and (video, business administration).

[0023] Optionally, for each candidate cluster center seed in the candidate cluster center seed set, the system calculates the joint distribution density of the resource type label and subject label corresponding to that candidate cluster center seed in the original innovation and entrepreneurship education resource set. The formula for calculating the joint distribution density is: ; in: Indicates the seed of the candidate cluster center The joint distribution density, Indicates the seed of the candidate cluster center The frequency of occurrence of the corresponding resource type tags in the original set of innovation and entrepreneurship education resources. Indicates the seed of the candidate cluster center The frequency of the corresponding subject tags appearing in the original collection of innovation and entrepreneurship education resources. Indicates the seed of the candidate cluster center The number of corresponding resource files. The system selects the candidate cluster center seed with the highest joint distribution density as the first initial cluster center. For example, among the four candidate cluster center seeds, the joint distribution density of (Courseware, Computer Science and Technology) is (4200×3500) / 1200=12250, (Courseware, Business Administration) is (4200×3200) / 1100≈12218, (Video, Computer Science and Technology) is (3800×3500) / 1000=13300, and (Video, Business Administration) is (3800×3200) / 900≈13511. Therefore, (Video, Business Administration) is selected as the first initial cluster center.

[0024] In some embodiments, after deleting the resource files corresponding to the candidate cluster center seeds selected as cluster centers, the system repeatedly performs the operation of calculating the joint distribution density and selecting the candidate cluster center seeds corresponding to the largest joint distribution density until the number of selected initial cluster centers reaches a preset cluster number threshold K=8. Starting from all selected initial cluster centers, the system calculates the comprehensive similarity between each resource file in the original innovation and entrepreneurship education resource set and each initial cluster center, classifying each resource file into the cluster containing the initial cluster center with the highest comprehensive similarity, thus obtaining multiple resource clusters. The comprehensive similarity is calculated by weighting resource type label similarity, subject label similarity, and resource file vector space model cosine similarity. Resource type label similarity is obtained by comparing the Jaccard coefficient of the label set, subject label similarity is calculated by the path distance of the subject classification coding tree, and resource file vector space model cosine similarity is calculated by TF-IDF feature vectors.

[0025] Optionally, the system performs an intra-cluster center update operation on the resource files within each resource cluster, using the updated intra-cluster center as the new cluster center. This process of classification and intra-cluster center update is repeated until the resource clusters no longer change. The system then names the resulting resource clusters according to the common tags of their resource files. For example, if the resource files within a cluster are mostly video files and their subject tags are concentrated in Business Administration, the cluster would be named "Business Administration Innovation and Entrepreneurship Video Resource Cluster." The system then stores these resource clusters and their naming results in a categorized educational resource database.

[0026] In one embodiment of the present invention, when performing resource association processing based on an improved association rule mining algorithm on the original industrial practice resource set, the system processor loads the improved association rule mining algorithm program, which calculates the itemset support based on a weighted average of resource usage frequency and resource update time. The system parses the resource usage logs of each industrial practice resource file in the original industrial practice resource set, extracting the access frequency sequence and last update timestamp of each industrial practice resource file over multiple historical time periods. The system calculates the average access frequency of each industrial practice resource file based on the access frequency sequence, and calculates the time decay coefficient of each industrial practice resource file based on the last update timestamp. The time decay coefficient is calculated using an exponential decay function, and the time constant of the exponential decay function is dynamically determined by the technology update cycle of the industry field to which the industrial practice resource file belongs. For example, the time constant for the artificial intelligence field is set to 6 months, and the time constant for the mechanical manufacturing field is set to 18 months. The system multiplies the average access frequency of each industrial practice resource file by the corresponding time decay coefficient to obtain the dynamic weighted support cardinality of each industrial practice resource file. The system scans all industry practice resource files in the original industry practice resource set, counts the number of times any two industry practice resource files are accessed in the same user session, and divides the number of co-occurrences by the total number of user sessions to obtain the original co-occurrence support of the two industry practice resource files. The system multiplies the original co-occurrence support by the geometric mean of the dynamic weighted support cardinality of the two industry practice resource files to obtain the improved co-occurrence support of the two industry practice resource files. The system pairs and filters all industry practice resource files whose improved co-occurrence support is greater than a preset minimum support threshold to form a frequent itemset candidate set. For each pair of industry practice resource files in the frequent itemset candidate set, the system calculates the ratio of the improved co-occurrence support of the pair to the dynamic weighted support cardinality of the first industry practice resource file in the pair to obtain the unidirectional association strength from the first industry practice resource file to the second industry practice resource file, and calculates the ratio of the improved co-occurrence support of the pair to the dynamic weighted support cardinality of the second industry practice resource file in the pair to obtain the reverse unidirectional association strength. The system pairs industry practice resource files whose unidirectional association strength and reverse unidirectional association strength are both greater than a preset minimum confidence threshold, marking them as bidirectional strong association pairs. The system constructs an industry practice resource association graph using all bidirectional strong association pairs as edges and each industry practice resource file as a node, and stores this graph as an associated industry resource library.

[0027] In its implementation, a cloud platform of a regional intelligent manufacturing industry alliance deployed a cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system. The original industry practice resource set contains 5,000 industry practice resource files from companies D, E, and F. The resource types cover three categories: technical documents, training projects, and job descriptions, and the industry fields include artificial intelligence, mechanical manufacturing, and electronic information. The improved association rule mining algorithm calculates the support of itemsets based on a weighted average of resource usage frequency and resource update time. The system analyzes the resource usage logs of each industry practice resource file in the original industry practice resource set, extracting the access frequency sequence and the last update timestamp for each industry practice resource file over multiple historical time periods. The access frequency sequence records the monthly access volume over the past 12 months, and the last update timestamp records the time of the most recent modification of the resource file.

[0028] In some embodiments, the system calculates the average access frequency of each industry practice resource file based on the access frequency sequence, and calculates the time decay coefficient of each industry practice resource file based on the last update timestamp. The time decay coefficient is calculated using an exponential decay function, and the time constant of the exponential decay function is dynamically determined by the technology update cycle of the industry field to which the industry practice resource file belongs. The technology update cycle is 6 months in the field of artificial intelligence, 18 months in the field of mechanical manufacturing, and 12 months in the field of electronic information. The formula for calculating the time decay coefficient is as follows: ; in: This represents the time decay coefficient of industry practice resource documents. This represents the time interval between the current moment and the last updated timestamp. This indicates the technology update cycle of the industry sector to which the industry practice resource document belongs. The system multiplies the average access frequency of each industry practice resource document by the corresponding time decay coefficient to obtain the dynamic weighted support cardinality for each document. For example, a technical document in the field of artificial intelligence that is accessed an average of 200 times per month and last updated 3 months ago has a dynamic weighted support cardinality of [missing value]. .

[0029] In some embodiments, the system scans all industry practice resource files in the original industry practice resource set, counts the number of times any two industry practice resource files are accessed in the same user session, and divides the number of co-occurrences by the total number of user sessions to obtain the original co-occurrence support of the two industry practice resource files. The system multiplies the original co-occurrence support by the geometric mean of the dynamic weighted support cardinality of the two industry practice resource files to obtain the improved co-occurrence support of the two industry practice resource files. The system pairs and filters all industry practice resource files whose improved co-occurrence support is greater than a preset minimum support threshold of 0.05 to form a frequent itemset candidate set.

[0030] Optionally, for each pair of industry practice resource files in the frequent itemset candidate set, the system calculates the ratio of the improved co-occurrence support of the pair to the dynamic weighted support cardinality of the first industry practice resource file in the pair, obtaining the unidirectional association strength from the first industry practice resource file to the second industry practice resource file. The system also calculates the ratio of the improved co-occurrence support of the pair to the dynamic weighted support cardinality of the second industry practice resource file in the pair, obtaining the reverse unidirectional association strength. The system marks pairwise combinations of industry practice resource files where both the unidirectional association strength and the reverse unidirectional association strength are greater than the preset minimum confidence threshold of 0.6 as bidirectionally strong association pairs.

[0031] In one embodiment of the present invention, when uploading the categorized educational resource library and the associated industry resource library to a cloud server for cross-library index mapping, the system assigns a first storage domain identifier to each resource cluster in the categorized educational resource library and assigns a first file identifier to each resource file within each resource cluster. The first storage domain identifier and the first file identifier are combined to generate the categorized library index key for each resource file. The system assigns a second storage domain identifier to each industry practice resource file in the associated industry resource library and retains the original file identifier of each industry practice resource file as the second file identifier. The second storage domain identifier and the second file identifier are combined to generate the associated library index key for each industry practice resource file. The system extracts the resource title field and resource summary field of each resource file in the categorized educational resource library, performs Chinese word segmentation on the resource title field and resource summary field, removes stop words, and generates a first keyword set for each resource file. The system extracts the industry field field and skill requirement field of each industry practice resource file in the associated industry resource library, performs word segmentation on the industry field field and skill requirement field, and generates a second keyword set for each industry practice resource file. The system calculates the keyword overlap between the first keyword set and the second keyword set. The keyword overlap is the ratio of the number of elements in the intersection of the first and second keyword sets to the number of elements in their union. When the keyword overlap exceeds a preset overlap threshold of 0.35, the mapping relationship between the category library index key of the resource file and the association library index key of the industry practice resource file is recorded as a cross-library link entry. The system aggregates all cross-library link entries to generate a cross-library mapping table and merges all category library index keys of the categorized education resource library with all association library index keys of the associated industry resource library to generate a total index key space. The system associates and stores each index key in the total index key space with the corresponding resource file metadata and the cross-library link entries related to that index key in the cross-library mapping table, generating a total index of collaborative education resources.

[0032] In its implementation, the cloud platform of a regional university innovation and entrepreneurship education alliance has completed the construction of a categorized educational resource database and a related industry resource database. The categorized educational resource database contains 12,000 resource files across 8 resource clusters, while the related industry resource database contains 5,000 nodes in an industry practice resource association graph. The system assigns a first storage domain identifier to each resource cluster in the categorized educational resource database. The first storage domain identifier uses the format "EDU_" prefix followed by the cluster number, for example, "EDU_001" corresponds to the "Business Administration Innovation and Entrepreneurship Video Resource Cluster". Furthermore, each resource file within each resource cluster is assigned a first file identifier, which uses the first 8 bits of the MD5 hash value uploaded when the resource was uploaded. The first storage domain identifier and the first file identifier are combined to generate the categorized database index key for each resource file, for example, "EDU_001_A3F5C8D2". The system assigns a second storage domain identifier to each industry practice resource file in the associated industry resource library. The second storage domain identifier adopts the format of "IND_" prefix plus industry field code, such as "IND_002" corresponding to the artificial intelligence field. The original file identifier of each industry practice resource file is kept as the second file identifier. The original file identifier is the UUID when uploaded by the enterprise. The second storage domain identifier and the second file identifier are combined to generate the associated library index key of each industry practice resource file, such as "IND_002_8A7B9C6D".

[0033] In some embodiments, the system extracts the resource title and resource summary fields from each resource file in the categorized educational resource library. It then performs Chinese word segmentation on the resource title and resource summary fields, removing stop words to generate a first keyword set for each resource file. For example, a resource file titled "Lean Startup Methodology" and with a summary titled "Introducing the core concepts and practical steps of Lean Startup" would result in a first keyword set {"Lean Startup", "Methodology", "Core Concepts", "Practical Steps"} after word segmentation. Similarly, the system extracts the industry domain and skill requirement fields from each industry practice resource file in the associated industry resource library. It then performs word segmentation on these fields to generate a second keyword set for each industry practice resource file. For example, an industry practice resource file with the industry domain "Artificial Intelligence" and skill requirements "Python Programming, Machine Learning Algorithms" would result in a second keyword set {"Artificial Intelligence", "Python Programming", "Machine Learning Algorithms"} after word segmentation.

[0034] Optionally, the system calculates the keyword overlap between the first keyword set and the second keyword set. The keyword overlap is the ratio of the number of elements in the intersection of the first keyword set and the number of elements in the union of the second keyword set. The formula for calculating the keyword overlap is: ; in: Indicates keyword overlap. Represents the set of first keywords. Represents the set of second keywords. This represents the number of elements in the intersection of the first keyword set and the second keyword set. This represents the number of elements in the union of the first keyword set and the second keyword set. When the keyword overlap is greater than the preset overlap threshold of 0.35, the system records the mapping relationship between the category library index key of the resource file and the association library index key of the industry practice resource file as a cross-library link entry. For example, the intersection of the first keyword set {"Artificial Intelligence", "Entrepreneurial Cases"} of a resource file and the second keyword set {"Artificial Intelligence", "Technological Applications"} of an industry practice resource file is {"Artificial Intelligence"}, and the union is {"Artificial Intelligence", "Entrepreneurial Cases", "Technological Applications"}. The keyword overlap is 1 / 3≈0.33, which is less than 0.35, so no cross-library link entry is generated. Another intersection is {"Artificial Intelligence", "Machine Learning"}, and the union is {"Artificial Intelligence", "Machine Learning", "Entrepreneurial Cases", "Technological Applications"}. The keyword overlap is 2 / 4=0.5, which is greater than 0.35, so a cross-library link entry is generated.

[0035] In some embodiments, the system aggregates all cross-database link entries to generate a cross-database mapping table, and merges all category library index keys of the categorized educational resource library with all associated library index keys of the associated industry resource library to generate a total index key space. The system associates and stores each index key in the total index key space with the corresponding resource file metadata and the cross-database link entries related to that index key in the cross-database mapping table, generating a total index of collaborative education resources. Optionally, the cross-database mapping table uses a key-value pair structure, where the key is the category library index key and the value is a list of corresponding associated library index keys. For example, "EDU_003_B2D4F7A1" corresponds to {"IND_005_9C8A3B2D", "IND_007_6E4F1A9C"}. It can be understood that the keyword overlap calculation method can effectively measure the semantic relevance between educational resources and industry resources, providing a quantitative basis for cross-database linking. The construction of the total index key space enables unified addressing of educational resources and industry resources, simplifying the complexity of subsequent resource retrieval and matching.

[0036] In one embodiment of the present invention, when constructing a multi-tenant access interface based on the overall index of collaborative education resources, refer to... Figure 3The system reads the first set of storage domain identifiers corresponding to the categorized educational resource database and the second set of storage domain identifiers corresponding to the associated industry resource database from the overall index of collaborative education resources. The system configures a first permission matrix for university users, allowing access to all resource files in the first set of storage domain identifiers and resource files associated with cross-database link entries in the second set of storage domain identifiers. The system configures a second permission matrix for enterprise users, allowing access to all resource files in the second set of storage domain identifiers and resource files associated with cross-database link entries in the first set of storage domain identifiers. The system configures a third permission matrix for student users, allowing access to resource files belonging to the preset innovation and entrepreneurship education level in the first set of storage domain identifiers and industry practice resource files belonging to the preset introductory level in the second set of storage domain identifiers. The system compiles the first, second, and third permission matrices into executable access control lists and deploys these lists at the interface gateway layer of the cloud server. The system establishes a Uniform Resource Locator (URL) routing table at the interface gateway layer, which maps resource identifiers in external access requests to corresponding index keys in the overall index of collaborative education resources. The system encapsulates the access control list and the Uniform Resource Locator (URL) routing table together to generate a multi-tenant access interface.

[0037] In its implementation, the cloud platform of a regional university innovation and entrepreneurship education alliance has completed the construction of a comprehensive index of collaborative education resources. This index includes a set of identifiers for the first storage domain (EDU_001 to EDU_008) corresponding to categorized educational resource databases and a set of identifiers for the second storage domain (IND_001 to IND_005) corresponding to related industry resource databases. The system reads the first and second storage domain identifier sets from the comprehensive index of collaborative education resources to prepare for configuring differentiated access permissions for multi-tenant users.

[0038] In some embodiments, the system configures a first permission matrix for university users. This first permission matrix explicitly allows university users to access all resource files in a first set of storage domain identifiers, as well as resource files associated with cross-database link entries in a second set of storage domain identifiers. For example, a university user can apply to access all courseware and video resources under EDU_001 to EDU_008, and can also view enterprise training projects under IND_002 associated with their university's courses. The system configures a second permission matrix for enterprise users. This second permission matrix explicitly allows enterprise users to access all resource files in a second set of storage domain identifiers, as well as resource files associated with cross-database link entries in the first set of storage domain identifiers. For example, an enterprise user can manage all technical documents and job descriptions under IND_001 to IND_005, and can also access student entrepreneurship cases under EDU_005 related to their own business.

[0039] Optionally, the system configures a third permission matrix for student user types. This third permission matrix explicitly allows student users to access resource files belonging to the preset innovation and entrepreneurship education levels (beginner and intermediate) in the first storage domain identifier set, and industry practice resource files belonging to the preset introductory level in the second storage domain identifier set. For example, students can browse introductory entrepreneurship guidance videos under EDU_003 and basic job skills descriptions under IND_004. The system compiles the first, second, and third permission matrices into executable access control lists, which store permission status using binary bit-based tagging and are deployed at the interface gateway layer of the cloud server.

[0040] In some embodiments, the system establishes a Uniform Resource Locator (URI) routing table at the interface gateway layer. This URI routing table stores the mapping relationship between resource identifiers in external access requests and index keys in the overall index of collaborative education resources. For example, it maps the request path " / api / resource / edu / 003 / B2D4F7A1" to the category library index key "EDU_003_B2D4F7A1". The system jointly encapsulates the access control list and the URI routing table to generate a multi-tenant access interface. This multi-tenant access interface provides a RESTful API endpoint and supports JSON-formatted identity token authentication.

[0041] Optionally, the formula for calculating the permission matching degree is: ; in: Indicates the degree of permission matching. Indicates the first The weight of each access control list item. Indicates the first The matching results of each access control list item (1 for a match, 0 for no match). This represents the total number of access control list entries. This formula is used to verify the consistency between user requests and the permission matrix, ensuring that only requests with the required permissions are forwarded to the resource storage layer.

[0042] It is understandable that the differentiated configurations of the first, second, and third permission matrices can meet the resource access needs of three types of users: universities, enterprises, and students, avoiding the risk of unauthorized access. It is also understandable that the establishment of the Uniform Resource Locator (URL) routing table standardizes resource request paths and simplifies the request forwarding logic of the interface gateway.

[0043] In one embodiment of the present invention, when receiving a resource retrieval request from a target user through a multi-tenant access interface, the system receives the resource retrieval request sent by the target user through the multi-tenant access interface, and parses the target user's user type identifier, the user's affiliated university identifier, and the user's input search keyword sequence from the resource retrieval request. Based on the user type identifier, the system retrieves the corresponding access control list from the multi-tenant access interface, and uses the access control list to perform permission filtering on the search keyword sequence, deleting keywords that exceed the target user's access permissions, thus generating a permission-filtered search keyword sequence. The system performs fuzzy matching between each keyword in the permission-filtered search keyword sequence and the overall index key space in the collaborative education resource index, calculating the text similarity between each keyword and the metadata of the resource file corresponding to each index key. The text similarity is the semantic similarity between the keyword and the metadata corresponding to the index key, calculated based on a pre-trained word vector model trained using a Skip-gram architecture. For each resource file, the system performs a weighted sum of the text similarity scores of all keywords corresponding to that resource file to obtain a preliminary matching score. The weighting coefficients used in the weighted summation are determined based on the order of each keyword's appearance in the search keyword sequence after permission filtering; keywords appearing earlier in the sequence have higher weighting coefficients. Based on the user's university identifier, the system retrieves the set of subject advantage tags for the target user's university from a pre-defined database of university-specific resource preferences. The system filters resource files with preliminary matching scores greater than a pre-defined matching threshold to form a candidate resource set. For each resource file in the candidate resource set, the system calculates the overlap between the resource file's subject tags and the set of subject advantage tags, multiplies this overlap by the resource file's preliminary matching score, and obtains the final matching score for that resource file. The system sorts the resource files in the candidate resource set according to their final matching scores from highest to lowest, selects the top few files to generate a personalized resource recommendation list, and returns this personalized resource recommendation list to the target user through a multi-tenant access interface.

[0044] In practice, during the operation of the cloud platform of a certain regional university innovation and entrepreneurship education alliance, student users submitted resource retrieval requests through the multi-tenant access interface. The retrieval request included the user type identifier, the identifier of the user's university, and the sequence of search keywords entered by the user.

[0045] In some embodiments, the system retrieves the corresponding access control list from the multi-tenant access interface based on the user type identifier. The access control list specifies that student users can only access resource files belonging to the preset innovation and entrepreneurship education level in the first storage domain identifier set, and industry practice resource files belonging to the preset introductory level in the second storage domain identifier set. The system uses the access control list to perform permission filtering operations on the search keyword sequence, deleting keyword items in the search keyword sequence that exceed the target user's access permissions. For example, if the search keyword sequence contains "advanced algorithm" (which belongs to non-introductory level industry resource keywords), it is deleted, generating a permission-filtered search keyword sequence ("artificial intelligence entrepreneurship case").

[0046] Optionally, the system performs fuzzy matching between each keyword in the permission-filtered search keyword sequence and the total index key space in the collaborative education resource index, and calculates the text similarity between each keyword and the resource file metadata corresponding to each index key. The text similarity is the semantic similarity between the keyword and the metadata corresponding to the index key, calculated based on a pre-trained word vector model. The pre-trained word vector model is trained using a Skip-gram architecture, and the training corpus contains all text metadata from the classified education resource database and the related industry resource database.

[0047] In some embodiments, for each resource file, the system performs a weighted sum of the text similarity scores of all keywords corresponding to that resource file to obtain a preliminary matching score for that resource file. The formula for calculating the preliminary matching score is as follows: ; in: This indicates the initial matching score of the resource file. Indicates the first The weight coefficient of each keyword (the weight coefficient is determined according to the position order of the keyword in the search keyword sequence after permission filtering. The earlier the keyword is, the greater the weight coefficient. For example, the first keyword has a weight coefficient of 0.5, the second has 0.3, and the third has 0.2). Indicates the first The similarity between each keyword and the text of the resource file. This indicates the length of the search keyword sequence after permission filtering.

[0048] Optionally, the system reads the set of subject advantage tags of the target user's university from a preset university characteristic resource preference database based on the user's university identifier. For example, the set of subject advantage tags of XX University is {"Computer Science and Technology", "Artificial Intelligence", "Entrepreneurship Management"}. The system filters out resource files with a preliminary matching score greater than the preset matching threshold (0.6) to form a candidate resource set. For each resource file in the candidate resource set, the system calculates the tag overlap degree between the subject tag of the resource file and the set of subject advantage tags. The tag overlap degree is the number of intersection elements between the subject tag of the resource file and the set of subject advantage tags divided by the total number of subject tags of the resource file. For example, if the subject tag of a resource file is {"Artificial Intelligence", "Entrepreneurship Management"}, the number of intersection elements is 2, the total number of subject tags is 2, and the tag overlap degree is 1.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system, characterized in that: include: The resource acquisition module acquires original innovation and entrepreneurship education resource sets uploaded by multiple universities and original industry practice resource sets uploaded by multiple enterprises. The classification and clustering module performs resource classification processing based on an improved clustering algorithm on the original set of innovation and entrepreneurship education resources to obtain a classified education resource library, specifically including: The improved clustering algorithm initializes cluster centers based on the joint distribution density of resource type labels and subject labels; Scan the metadata of each resource file in the original set of innovation and entrepreneurship education resources, and extract the resource type tag set and subject tag set of each resource file; Calculate the frequency distribution of each resource type tag in the resource type tag set appearing in the original innovation and entrepreneurship education resource set, and calculate the frequency distribution of each subject tag in the subject tag set appearing in the original innovation and entrepreneurship education resource set. A set of candidate cluster center seeds is generated based on the Cartesian product of the top several resource type labels with the highest frequency in the resource type frequency distribution and the top several subject labels with the highest frequency in the subject frequency distribution. For each candidate cluster center seed in the candidate cluster center seed set, the joint distribution density of the resource type label and subject label corresponding to the candidate cluster center seed in the original innovation and entrepreneurship education resource set is calculated, and the candidate cluster center seed with the largest joint distribution density is taken as the first initial cluster center. After deleting the resource files corresponding to the candidate cluster center seeds that have been selected as cluster centers, repeat the operation of calculating the joint distribution density and selecting the candidate cluster center seeds corresponding to the maximum joint distribution density until the number of initial cluster centers selected reaches the preset cluster number threshold. Starting from all the selected initial cluster centers, calculate the comprehensive similarity between each resource file in the original set of innovation and entrepreneurship education resources and each initial cluster center, and classify each resource file into the cluster of the initial cluster center with the highest comprehensive similarity, thus obtaining multiple resource clusters; Perform an intra-cluster center update operation on the resource files within each resource cluster, and use the updated intra-cluster center as the new cluster center. Repeat the classification operation and the intra-cluster center update operation until each resource cluster no longer changes. The resulting multiple resource clusters are named according to the common tags of the resource files within each cluster, and the multiple resource clusters and their naming results are stored as the classified educational resource library. The association mining module performs resource association processing based on an improved association rule mining algorithm on the original set of industry practice resources to obtain an associated industry resource library, specifically including: The improved association rule mining algorithm calculates the support of itemsets based on a weighted average of resource usage frequency and resource update time. Analyze the resource usage log of each industry practice resource file in the original industry practice resource set, and extract the access frequency sequence and last update timestamp of each industry practice resource file in multiple historical time periods from the resource usage log; The average access frequency of each industry practice resource file is calculated based on the access frequency sequence, and the time decay coefficient of each industry practice resource file is calculated based on the last update timestamp. The time decay coefficient is inversely proportional to the time interval from the current moment to the last update timestamp. Multiply the average access frequency of each industry practice resource file by the corresponding time decay coefficient to obtain the dynamic weighted support base of each industry practice resource file; Scan all industry practice resource files in the original industry practice resource set, count the number of times any two industry practice resource files are accessed in the same user session, and divide the number of times they are accessed by the total number of user sessions to obtain the original co-occurrence support of the two industry practice resource files. Multiply the original co-occurrence support by the geometric mean of the dynamic weighted support base of the two industry practice resource documents to obtain the improved co-occurrence support of the two industry practice resource documents; All industry practice resource files whose co-occurrence support is greater than the preset minimum support threshold are paired and filtered out to form a frequent itemset candidate set; For each industry practice resource file in the frequent itemset candidate set, pairwise combinations are made. The ratio of the improved co-occurrence support of the combination to the dynamic weighted support cardinality of the first industry practice resource file in the combination is calculated to obtain the unidirectional association strength from the first industry practice resource file to the second industry practice resource file. The ratio of the improved co-occurrence support of the combination to the dynamic weighted support cardinality of the second industry practice resource file in the combination is also calculated to obtain the reverse unidirectional association strength. Industry practice resource files whose unidirectional association strength and reverse unidirectional association strength are both greater than the preset minimum confidence threshold are paired and marked as bidirectional strong association pairs; Using all bidirectional strong association pairs as edges and each industry practice resource file as a node, construct an industry practice resource association graph, and store the industry practice resource association graph as the associated industry resource library; The index mapping module uploads the categorized educational resource database and the associated industry resource database to the cloud server for cross-database index mapping processing to generate a general index of collaborative education resources. The multi-tenant interface module constructs a multi-tenant access interface based on the overall index of collaborative education resources. The multi-tenant access interface is used to provide differentiated resource access permissions to university users, enterprise users, and student users. The recommendation matching module receives resource retrieval requests from target users through the multi-tenant access interface, performs resource matching processing based on the overall index of collaborative education resources, and generates a personalized resource recommendation list for the target user.

2. The cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system according to claim 1, characterized in that, The overall similarity is calculated by weighting the resource type label similarity, subject label similarity, and resource file vector space model cosine similarity.

3. The cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system according to claim 1, characterized in that, The time decay coefficient is calculated using an exponential decay function, and the time constant of the exponential decay function is dynamically determined by the technology update cycle of the industry field to which the industry practice resource file belongs.

4. The cloud-based innovation and entrepreneurship education resource sharing and collaborative education system according to claim 1, characterized in that, The steps of uploading the categorized educational resource database and the associated industry resource database to a cloud server for cross-database index mapping to generate a comprehensive index of collaborative education resources specifically include: Assign a first storage domain identifier to each resource cluster in the classified educational resource library, and assign a first file identifier to each resource file within each resource cluster. Combine the first storage domain identifier and the first file identifier to generate a classified library index key for each resource file. Assign a second storage domain identifier to each industry practice resource file in the associated industry resource library, and keep the original file identifier of each industry practice resource file as the second file identifier. Combine the second storage domain identifier and the second file identifier to generate the association library index key for each industry practice resource file. Extract the resource title field and resource summary field of each resource file in the classified educational resource library, perform word segmentation on the resource title field and the resource summary field, and generate the first keyword set of each resource file; Extract the industry domain field and skill requirement field from each industry practice resource file in the associated industry resource library, perform word segmentation on the industry domain field and the skill requirement field, and generate a second keyword set for each industry practice resource file; Calculate the keyword overlap between the first keyword set and the second keyword set. When the keyword overlap is greater than a preset overlap threshold, record the mapping relationship between the classification library index key of the resource file and the association library index key of the industry practice resource file as a cross-library link entry. All cross-database link entries are aggregated to generate a cross-database mapping table, and all category database index keys of the categorized educational resource database are merged with all associated database index keys of the associated industry resource database to generate a total index key space. Each index key in the total index key space is associated with the corresponding resource file metadata and the cross-database link entries related to that index key in the cross-database mapping table, and the total index of collaborative education resources is generated.

5. The cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system according to claim 4, characterized in that, The keyword overlap is the ratio of the number of elements in the intersection of the first keyword set and the second keyword set to the number of elements in the union.

6. The cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system according to claim 5, characterized in that, Based on the overall index of collaborative education resources, a multi-tenant access interface is constructed. This multi-tenant access interface is used to provide differentiated resource access permissions to university users, enterprise users, and student users. Specifically, the steps include: Read the first set of storage domain identifiers corresponding to the classified education resource database and the second set of storage domain identifiers corresponding to the associated industry resource database from the overall index of collaborative education resources. Configure a first permission matrix for university user types. The first permission matrix allows access to all resource files in the first storage domain identifier set and resource files associated with cross-database link entries in the second storage domain identifier set. Configure a second permission matrix for enterprise user types, which allows access to all resource files in the second storage domain identifier set and resource files in the first storage domain identifier set associated with cross-database link entries; Configure a third permission matrix for student user types. The third permission matrix allows access to resource files belonging to the preset innovation and entrepreneurship education level in the first storage domain identifier set, and to industry practice resource files belonging to the preset entry level in the second storage domain identifier set. The first permission matrix, the second permission matrix, and the third permission matrix are compiled into executable access control lists, and the access control lists are deployed in the interface gateway layer of the cloud server. A Uniform Resource Locator (URL) routing table is established at the interface gateway layer. The URL routing table is used to map resource identifiers in external access requests to the corresponding index keys in the overall index of collaborative education resources. The access control list and the uniform resource locator routing table are jointly encapsulated to generate the multi-tenant access interface.

7. The cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system according to claim 6, characterized in that, The steps of receiving resource retrieval requests from target users through the multi-tenant access interface, performing resource matching processing based on the collaborative education resource index, and generating a personalized resource recommendation list for the target users specifically include: The multi-tenant access interface receives resource retrieval requests sent by target users and parses the user type identifier, the university identifier to which the user belongs, and the sequence of search keywords entered by the user from the resource retrieval requests. The corresponding access control list is retrieved from the multi-tenant access interface based on the user type identifier. The access control list is used to perform permission filtering on the search keyword sequence, deleting keyword items in the search keyword sequence that exceed the access permissions of the target user, and generating a permission-filtered search keyword sequence. Each keyword in the search keyword sequence after permission filtering is fuzzily matched with the total index key space in the total index of collaborative education resources, and the text similarity between each keyword and the resource file metadata corresponding to each index key is calculated. For each resource file, the text similarity of all keywords corresponding to that resource file is weighted and summed to obtain the preliminary matching score of that resource file. The weight coefficient used in the weighted summation is determined according to the order of appearance of each keyword in the search keyword sequence after the permission filtering. The earlier the keyword appears, the larger the weight coefficient it corresponds to. Based on the user's university identifier, the set of disciplinary advantage tags of the target user's university is read from a preset university characteristic resource preference database; Resource files whose preliminary matching scores are greater than a preset matching threshold are selected to form a candidate resource set. For each resource file in the candidate resource set, the overlap between the subject tag of the resource file and the subject advantage tag set is calculated. The overlap between the tag and the preliminary matching score of the resource file is multiplied to obtain the final matching score of the resource file. The resource files in the candidate resource set are sorted in descending order of the final matching score. The top few resource files are selected to generate the personalized resource recommendation list, and the personalized resource recommendation list is returned to the target user through the multi-tenant access interface.

8. The cloud-based innovation and entrepreneurship education resource sharing and collaborative talent cultivation system according to claim 7, characterized in that, The text similarity is the semantic similarity between keywords and the corresponding metadata of index keys, calculated based on a pre-trained word vector model.

Images