Proprietary education fusion management method based on big data analysis

Abstract

The invention discloses a production and education fusion management method based on big data analysis, which realizes student portrait characterization and resource content semantic characterization by obtaining and deeply mining student multi-dimensional portrait data and recommendable resource content data, and realizes the multi-scale semantic progressive interaction technology. And the adaptation degree between the portrait and the recommendation resources is effectively improved. Through the mode, accurate recommendation oriented to individual differences can be realized, the growth track change of students can be dynamically responded, and a scientific decision basis is provided for colleges and enterprises.

Description

Technical Field

[0001] The present application relates to the field of intelligent management, and more specifically, to a production-education integration management method based on big data analysis. Background Art

[0002] With the rapid development of the social economy and the continuous upgrading of the industrial structure, industry-university integration has become an important way to promote deep collaboration between higher education and industry, and to align talent development with industrial needs. Establishing a scientific and efficient industry-university integration management plan will not only help improve students' practical skills and employment competitiveness, but also enhance universities' ability to serve local economic and social development and achieve the optimal allocation of educational and industrial resources. Therefore, establishing a comprehensive industry-university integration management system is of great significance for achieving the organic connection between the education chain, talent chain, industrial chain, and innovation chain.

[0003] Currently, most existing industry-education integration management solutions rely on manual experience or simple information technology, typically matching resources through manual recommendations, fixed course settings, or analysis of small amounts of data. These methods have many shortcomings in practical applications: on the one hand, it is difficult to fully tap into the differences among individual students in multiple dimensions such as knowledge structure, interests, specialties, and career orientations, resulting in severe homogeneity in recommended content and an inability to meet students' personalized growth needs; on the other hand, the ability to integrate resources from the enterprise and education sides is limited, and the lack of a dynamic data-driven mechanism makes industry-education integration inefficient. In addition, due to the lack of intelligent analysis methods, current management systems have difficulty perceiving changes in student status in real time and are unable to accurately predict their development paths, thus affecting the quality of talent training and the effectiveness of industry-education collaboration.

[0004] Therefore, we look forward to an optimized management method of industry-education integration based on big data analysis. Summary of the Invention

[0005] In order to solve the above technical problems, this application is proposed. The embodiment of this application provides an industry-education integration management method based on big data analysis. By acquiring and deeply mining student multi-dimensional portrait data and recommendable resource content data, the student portrait characterization and resource content semantic characterization are realized, and the multi-scale semantic progressive interaction technology is used to effectively improve the adaptability between portraits and recommended resources. In this way, not only can accurate recommendations based on individual differences be achieved, but also changes in students' growth trajectories can be dynamically responded to, providing scientific decision-making basis for colleges and enterprises.

[0006] According to one aspect of the present application, a method for managing industry-education integration based on big data analysis is provided, which includes:

[0007] Obtain multi-dimensional portrait data of target students;

[0008] Get a collection of recommended resource content data;

[0009] Performing student portrait characterization on the multi-dimensional portrait data of the target student object to obtain the current state portrait vector of the target student;

[0010] Performing resource content characterization on each piece of recommendable resource content data in the set of recommendable resource content data to obtain a set of recommendable resource content semantic feature vectors;

[0011] Inputting each of the recommended resource content semantic feature vectors and the target student's current status portrait vector in the set of recommended resource content semantic feature vectors into a portrait-recommended resource matching analysis network to obtain a set of comprehensive fitness scores;

[0012] The recommended resource content data corresponding to the top K maximum comprehensive fitness scores in the set of comprehensive fitness scores are output as a recommended resource content recommendation list.

[0013] Compared with existing technologies, this application provides a production-education integration management method based on big data analysis. By acquiring and deeply mining multi-dimensional student portrait data and recommended resource content data, it achieves the characterization of student portraits and the semantic characterization of resource content. It also uses multi-scale semantic progressive interaction technology to effectively improve the adaptability between portraits and recommended resources. In this way, not only can accurate recommendations based on individual differences be achieved, but it can also dynamically respond to changes in students' growth trajectories, providing scientific decision-making basis for universities and enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 Flowchart of a method for managing industry-education integration based on big data analysis according to an embodiment of the present application;

[0016] Figure 2 Schematic diagram of data flow of the industry-education integration management method based on big data analysis according to an embodiment of the present application;

[0017] Figure 3 This is a flowchart of sub-step S5 of the industry-education integration management method based on big data analysis according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0019] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0020] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0021] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0022] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0023] In the technical solution of this application, a production-education integration management method based on big data analysis is proposed. Figure 1 Flowchart of the industry-education integration management method based on big data analysis according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the industry-education integration management method based on big data analysis according to the embodiment of the present application. Figure 1 and Figure 2As shown, the industry-education integration management method based on big data analysis according to the embodiment of the present application includes the following steps: S1, obtaining student multi-dimensional portrait data of the target student object; S2, obtaining a set of recommendable resource content data; S3, performing student portrait characterization on the student multi-dimensional portrait data of the target student object to obtain the target student current state portrait vector; S4, performing resource content characterization on each recommendable resource content data in the set of recommendable resource content data to obtain a set of recommendable resource content semantic feature vectors; S5, inputting each recommendable resource content semantic feature vector in the set of recommendable resource content semantic feature vectors and the target student current state portrait vector into the portrait-recommendation resource matching analysis network to obtain a set of comprehensive fitness scores; S6, outputting the recommendable resource content data corresponding to the top K largest comprehensive fitness scores in the set of comprehensive fitness scores as a recommendable resource content recommendation list.

[0024] In particular, the S1 obtains the student multi-dimensional portrait data of the target student object. Among them, the student multi-dimensional portrait data includes student portrait structured data and student portrait unstructured data; student portrait structured data includes academic performance, courses taken and majors; student portrait unstructured data includes project reports, course assignment summaries, interest self-statements, and career goal self-statements. Specifically, structured data such as academic performance, courses taken and majors can accurately reflect students' knowledge base and professional development paths, and provide a quantitative basis for analyzing their academic level and learning progress. Unstructured data such as project reports and course assignment summaries can demonstrate students' innovative ability and problem-solving ability in practical applications; interest self-statements and career goal self-statements reveal students' internal motivation, interest preferences and future development intentions. This information is crucial for accurately identifying students' potential and growth direction. In the technical solution of this application, by obtaining multi-dimensional portrait data of target students, a detailed data foundation can be provided for subsequent data characterization processing and intelligent matching networks, enabling the system to model student status based on more comprehensive information, thereby improving the pertinence and adaptability of resource recommendations. This not only optimizes the allocation of educational resources, but also provides the industry with talents that better meet the needs, thereby promoting collaborative innovation between industry and education and high-quality development.

[0025] In particular, the S2 obtains a collection of recommendable resource content data. The collection of recommendable resource content data includes structured data of recommendable resource content and unstructured data of recommendable resource content; the structured data of recommendable resource content includes prerequisite requirements for courses, a list of technology stacks required for projects, and hard requirements for positions (educational background, years of work experience); the unstructured data of recommendable resource content includes course outlines, project descriptions, and job responsibilities. Specifically, structured data such as prerequisite requirements for courses, a list of technology stacks required for projects, and hard requirements for positions (including academic background and years of work experience, etc.) can clearly define the basic thresholds and conditions for each resource, and provide a rigorous quantitative basis for resource screening and matching, which is directly related to whether the recommendation system can accurately identify the degree of fit between resources and students' abilities and backgrounds. Unstructured data such as course outlines, project descriptions, and job responsibilities elaborate on the knowledge content, practical requirements, and actual work content of the positions, fully reflecting the industry's multi-dimensional needs for talent skills and responsibilities. These rich semantic information helps to deeply understand resource characteristics and improve the accuracy of semantic matching for recommendations. In the technical solution of this application, by acquiring a collection of recommendable resource content data, a detailed and multi-level data foundation can be provided for subsequent resource content characterization processing and matching with student profiles. This effectively improves the comprehensiveness and expressiveness of resource descriptions in the industry-education integration management system, making the recommendation results more in line with the actual needs of the industry. This not only enhances the speed and accuracy of educational resources' response to industry needs, but also promotes a high degree of matching between student abilities and industry job requirements, thereby improving the quality and efficiency of talent training and laying a solid foundation for the sustainable and healthy development of industry-education integration.

[0026] In particular, the S3 performs student portrait characterization on the student multi-dimensional portrait data of the target student object to obtain the target student current state portrait vector. It should be understood that the original student multi-dimensional portrait data contains a large amount of structured and unstructured information. These information types are diverse and the forms of expression are complex. It is difficult to perform effective comparison and matching directly. Therefore, it is necessary to convert them into a unified vector form with semantic expression capabilities to support subsequent intelligent analysis and recommendation. Therefore, in the technical solution of the present application, first, the student portrait structured data is passed through a student state portrait encoder based on the BERT model to capture the inherent semantic features of quantitative and ordered information such as academic performance, courses taken, and major directions, as well as its specific reflection of the student's academic level and professional development status, to obtain the first target student current state portrait vector. This approach enables deep semantic understanding of structured data, transforming simple numerical and categorical information into vector representations with contextual semantics, thereby improving the model's perception of students' academic status. Next, each unstructured data element within the student profile unstructured data is structuredly encoded to obtain a set of student profile embedding vectors. This set of student profile embedding vectors is then passed through a BERT-based student status profile encoder to extract the deep semantic features of the unstructured data and the relevance of the student status, enhancing the accurate portrayal of the student's potential, interests, and development direction, and obtaining a second target student current status profile vector. Finally, to achieve the comprehensive utilization of multimodal and multi-source information and construct a comprehensive and detailed student current status profile vector that includes both quantitative academic and major information and rich expressions of interests and career intentions, the first target student current status profile vector and the second target student current status profile vector are fused to obtain the target student current status profile vector. This approach significantly improves the overall expressiveness and adaptability of the profile, enabling the subsequent matching analysis network to achieve efficient resource matching and personalized recommendations based on more accurate, multi-dimensional, and dynamically changing information.

[0027] In particular, the S4 performs resource content characterization on each of the recommendable resource content data in the set of recommendable resource content data to obtain a set of recommendable resource content semantic feature vectors. It should be understood that the recommendable resource content data itself contains rich and diverse information types, including both clearly quantified structured data, such as prerequisite requirements for courses, a list of technology stacks required for projects, and rigid requirements for positions, and also covers deep semantics and descriptive unstructured data, such as course outlines, project descriptions, and job responsibilities. The form and content of these data vary significantly, and it is difficult to construct a unified representation that can reflect the connotation and characteristics of the resources by direct use. Therefore, in the technical solution of the present application, similarly, each of the recommendable resource content data in the set of recommendable resource content data is characterized, and the structured and unstructured data are semantically encoded respectively, and then merged into a unified recommendable resource content semantic feature vector. Specifically, the structured data in the recommended resource content is first encoded. Semantic encoders based on deep learning models such as BERT can effectively capture the inherent logical relationships and contextual semantics contained in course prerequisites, technology stack lists, and job requirements, rendering these quantitative indicators no longer isolated arrays or labels, but rather high-dimensional vector representations with rich semantic expression. At the same time, natural language processing techniques are used to convert unstructured textual information such as course syllabi, project descriptions, and job responsibilities into sets of embedded vectors. These vectors can reflect the details of the knowledge points involved in the resource, the practical requirements, and the multi-dimensional information of the job responsibilities. Subsequently, an encoder based on the BERT model is used to further extract features from these unstructured embedded vectors, deepening the semantic understanding of the unstructured content. Finally, after encoding the structured and unstructured data separately, the two are fused to form a unified semantic feature vector for the recommended resource content. This fusion process not only integrates the quantitative threshold and qualitative description of the resource, but also enhances the vector's expressive power at the semantic level, facilitating subsequent multi-scale and multi-level semantic interaction with the student's current status profile vector. This multimodal semantic fusion can effectively make up for the limitations of single data type expression and achieve a comprehensive characterization of resource characteristics.

[0028] In particular, in said S5, each of the semantic feature vectors of the recommended resource content and the target student's current status portrait vector is input into the portrait-recommended resource matching analysis network to obtain a set of comprehensive fitness scores. In a specific example of the present application, Figure 3 As shown, the S5 includes: S51, performing multi-scale semantic progressive interaction on the semantic feature vector of the recommended resource content and the target student current status portrait vector to obtain the student portrait-recommended resource semantic collaborative coding vector; S52, performing feature decoding on the student portrait-recommended resource semantic collaborative coding vector to obtain a comprehensive fitness score.

[0029] Specifically, the S51 performs multi-scale semantic progressive interaction on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain a student portrait-recommended resource semantic collaborative coding vector. It should be understood that the student's multi-dimensional portrait data (such as academic performance, project reports, and career goal self-descriptions) and resource content data (such as course outlines and job responsibilities) contain structured and unstructured information respectively, and the two are heterogeneous in semantic representation. Structured data (such as grades and technology stacks) reflect explicit ability labels, and unstructured data (such as self-description texts and project descriptions) contain potential interests and demand tendencies. Traditional methods based on manual experience or shallow data analysis are difficult to effectively analyze the complex relationship between the student's multi-dimensional portrait and the resource content semantics, resulting in homogeneous recommendation results and lack of dynamic adaptability. Therefore, in order to construct a dynamic matching model with hierarchical perception capabilities and realize deep semantic collaboration between the education chain and the industrial chain, in the technical solution of this application, a multi-scale semantic progressive interaction is performed on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain a student portrait-recommended resource semantic collaborative coding vector. Specifically, through multi-scale semantic progressive interaction, the system gradually deconstructs the multidimensional associations between individual student characteristics and resource elements: at the low level, it precisely integrates academic performance with course prerequisites; at the mid-level, it establishes a structured correspondence between project experience and technology stack requirements; and at the deep level, it explores the semantic consistency between self-described interests and job responsibilities. This progressive interaction mechanism overcomes the limitations of single-level analysis by integrating semantic information of varying granularity in stages. This enables the system to identify both the explicit fit between students' current skills and the course syllabus and the implicit fit between their potential interests and job responsibilities, ultimately forming a collaborative encoding vector that covers both short-term capability matching and long-term development orientation. The student profile-recommended resource semantic collaborative encoding vector system generated through multi-scale semantic fusion transcends the reliance of traditional recommendation methods on explicit features and establishes a three-dimensional matching model that encompasses students' knowledge structure, practical abilities, and development aspirations with industry resource and technology requirements, job characteristics, and development potential. This not only improves the instant adaptability of resource recommendations such as courses and projects, but also predicts the synergistic relationship between students' growth paths and the evolution trend of industry needs through deep semantic associations, providing a decision-making basis for two-way dynamic adjustments in the industry-education integration ecosystem, and truly realizing the deep coupling of the education chain and the industrial chain.

[0030] Specifically, multi-level latent feature extraction is first performed on the semantic feature vectors of the recommended resource content and the target student's current status profile vector to obtain the mid-level latent feature encoding vectors of the recommended resource, the mid-level latent feature encoding vectors of the student profile, the deep-level latent feature encoding vectors of the recommended resource, and the deep-level latent feature encoding vectors of the student profile. It should be understood that although the feature vectors of the student profile and resource content have been initially encoded, the information they contain still contains potential correlations at multiple levels and dimensions. The student's current status profile vector includes both explicit structured data such as academic performance and implicit unstructured features such as self-described interests; the resource content semantic feature vector also includes hard indicators such as the technology stack list and soft requirements such as job descriptions. Traditional single-layer feature extraction has difficulty effectively deconstructing the nonlinear relationships between these complex features.

[0031] Therefore, in order to build a multi-level feature representation system and provide an interactive basis for granular adaptation for subsequent cross-modal fusion, in the technical solution of this application, multi-level implicit feature extraction is performed on the semantic feature vector of the recommended resource content and the current state portrait vector of the target student to obtain the recommended resource middle-level implicit feature coding vector, the student portrait middle-level implicit feature coding vector, the recommended resource deep-level implicit feature coding vector and the student portrait deep-level implicit feature coding vector. That is, through the hierarchical nonlinear transformation of the deep neural network, the student portrait vector is deconstructed into middle-level implicit features (such as knowledge structure modularization ability, project experience transferability) and deep-level implicit features (such as career development potential, innovative thinking tendency); the resource content vector is parsed into middle-level implicit features (such as technology stack combination mode, course connection logic) and deep-level implicit features (such as job core value demands, industry innovation trend fit). This hierarchical decoupling enables the system to distinguish between explicit ability labels and implicit quality features, avoiding mismatching caused by shallow feature interactions. Specifically, mid-level implicit feature encoding focuses on modular mapping of structured capabilities (e.g., deconstructing a student's courses and project technology stack requirements into composable skill units), while deep-level implicit feature encoding aims to uncover semantic consistency in unstructured text (e.g., extracting the potential correlation between value orientation and job culture attributes from self-statements of career goals). This hierarchical feature extraction mechanism enables subsequent cross-level interactions to ensure both accurate matching of foundational capabilities and forward-looking predictions of development potential, providing a multi-granular decision-making basis for dynamic adaptation of industry-education resources.

[0032] In a specific example of the present application, the following feature extraction formula is used to perform multi-level implicit feature extraction on the semantic feature vector of the recommended resource content and the target student current status portrait vector to obtain the recommended resource middle-level implicit feature coding vector, the student portrait middle-level implicit feature coding vector, the recommended resource deep-level implicit feature coding vector, and the student portrait deep-level implicit feature coding vector; wherein, the feature extraction formula is:

[0033] m1=ReLU(W mid v1+b mid )

[0034] m2=ReLU(W mid v2+b mid )

[0035] d1=Sigmoid(W deep m1+b deep )

[0036] d2=Sigmoid(W deep m2+b deep )

[0037] Among them, v1 is the semantic feature vector of the recommended resource content, v2 is the current status portrait vector of the target student, and W mid represents the weight matrix of mid-level feature extraction, b mid represents the bias term for mid-level feature extraction, ReLU(·) represents the ReLU function, m1 is the mid-level implicit feature encoding vector of the recommended resource, m2 is the mid-level implicit feature encoding vector of the student portrait, W deep represents the weight matrix for deep feature extraction, b deep represents the bias term for deep feature extraction, d1 is the deep implicit feature coding vector of the recommended resource, and d2 is the deep implicit feature coding vector of the student portrait.

[0038] Next, low-level feature fusion is performed on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain the student portrait-recommended resource low-level fusion semantic feature encoding vector. It should be understood that although the student's current status portrait vector and the resource semantic feature vector have been encoded to form a vectorized representation, their original feature space still retains irreplaceable fine-grained information. The structured data in the student portrait (such as academic performance, courses taken) and the structured data of the resource content (such as course prerequisites, technology stack list) are naturally quantitatively comparable in the numerical dimension, and the initial encoding of unstructured data (such as course assignment summaries, project descriptions) also contains word-level semantic associations. If high-level abstraction of the deep network is directly performed, the direct correspondence between explicit ability labels and resource hard conditions may be lost prematurely, resulting in the weakening of the basic matching logic. Low-level feature fusion ensures accurate mapping of explicit conditions by preserving the fine granularity of the original feature space. For example, it can directly link a student's course score with the score threshold of the resource prerequisite requirement, or achieve word-level matching of key technical terms in a project report with the skill description in the job responsibilities.

[0039] Therefore, in order to build a deterministic framework for basic matching and provide reliable underlying support for subsequent mid- and high-level semantic interactions, in the technical solution of this application, low-level feature fusion is performed on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain a student portrait-recommended resource low-level fusion semantic feature encoding vector. The generated student portrait-recommended resource low-level fusion semantic feature encoding vector contains both the numerical correspondence of structured data (such as the linear matching of credit points and course requirements) and the local semantic association of unstructured text (such as the co-occurrence frequency of technical terms), providing an information base with both accuracy and richness for subsequent cross-level feature interactions, enabling the recommendation system to carry out deeper potential mining and trend prediction under the premise of ensuring the accuracy of basic matching.

[0040] In a specific example of the present application, the following feature fusion formula is used to perform low-level feature fusion on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain a student portrait-recommended resource low-level fusion semantic feature encoding vector; wherein, the feature fusion formula is:

[0041]

[0042] in, represents positional addition, MLP(·) represents multi-layer perceptron, and f low A low-level semantic feature encoding vector is generated for the student portrait-recommended resource.

[0043] Next, mid-level feature fusion is performed on the mid-level latent feature encoding vectors of the recommended resources and the mid-level latent feature encoding vectors of the student profile to obtain the mid-level fused semantic feature encoding vector of the student profile and recommended resources. It can be understood that the mid-level latent features of the student profile and resource content have been extracted into more structured representations through the deep network. While the mid-level latent feature encodings of the student profile (such as course modularity and project experience transferability) and the mid-level latent feature encodings of the resources (such as technology stack combination patterns and course connection logic) have been freed from the detailed noise of the original data, they still retain decomposable component-level semantic units. Traditional recommendation systems often focus only on explicit condition matching or shallow semantic associations, making it difficult to capture the structural complementary relationship between student skill sets and resource requirements. For example, a student's experience in multiple discrete projects may be abstracted in the mid-level feature space as "cross-domain collaboration capabilities," while the technology stack list in an enterprise project requirement may be encoded as "multi-technology integration implementation framework." The potential compatibility between the two needs to be revealed through structural interaction.

[0044] Therefore, in order to establish a cross-source structural association network and achieve compositional matching between capability units and demand elements, in the technical solution of this application, the mid-level implicit feature coding vectors of the recommended resources and the mid-level implicit feature coding vectors of the student portraits are fused at the mid-level to obtain the mid-level fusion semantic feature coding vectors of the student portraits and recommended resources. Specifically, through mid-level feature fusion, the system can dynamically map the modular capabilities in the student portraits (such as data analysis foundation, engineering practice ability) with the structural elements in the resource requirements (such as project stage technology dependencies, job skill matrix). For example, the "data processing process optimization experience" implied in the student project report is associated with the "business system iteration support requirements" encoded in the job responsibilities at the component level to identify the transferable value of students in a specific technology ecosystem. This fusion mechanism not only focuses on the correspondence of static skill labels, but also analyzes the topological adaptation relationship between capability combinations and demand structures, such as determining whether the student's existing knowledge modules constitute a sufficient subset of the technology system required for a certain R&D position. By capturing the compositional relationship at the mid-level abstraction level, the system can identify the non-explicit association between students' implicit capabilities and potential resource requirements. For example, from the interaction of mid-level features, we can discover the structural similarities between the systematic thinking tendencies demonstrated in students' course design and the architectural design capabilities required in corporate innovation projects, and then recommend high-potential positions that may be overlooked by traditional matching rules, thereby improving the structured decision-making ability of the recommendation system.

[0045] In a specific example of the present application, the mid-level implicit feature coding vector of the recommended resource and the mid-level implicit feature coding vector of the student portrait are fused at the mid-level using the following feature fusion formula to obtain the mid-level fused semantic feature coding vector of the student portrait-recommended resource; wherein the feature fusion formula is:

[0046] f mid =Attn(m1,m2)+m1⊙m2

[0047] Among them, ⊙ represents the position point multiplication, Attn(·) represents the attention mechanism, softmax(·) represents the softmax function, L represents the feature dimension of the vector, and f mid A vector encoding a hierarchical fusion semantic feature for the student portrait-recommendation resource.

[0048] Then, deep feature fusion is performed on the deep implicit feature encoding vectors of the recommended resources and the deep implicit feature encoding vectors of the student portraits to obtain the deep-level fusion semantic feature encoding vectors of the student portraits and recommended resources. It should be understood that the deep implicit features of the student portraits and resource content have been refined into a highly abstract semantic space through the neural network. Although the deep implicit feature encodings of students (such as career value orientation and innovative thinking patterns) and resources (such as corporate strategic talent demands and project innovation trends) are stripped of specific skill labels, they carry the core semantics of individual development trajectories and resource evolution directions. Traditional recommendation systems often stop at explicit condition matching and fail to capture the implicit resonance between students' growth potential and the strategic value of industry resources.

[0049] Therefore, in order to build a high-order semantic collaborative network and realize the deep coupling of the education chain and the industrial chain in the innovation dimension, in the technical solution of this application, the deep implicit feature coding vector of the recommended resources and the deep implicit feature coding vector of the student portrait are deeply fused to obtain the student portrait-recommended resource deep-level fusion semantic feature coding vector. Through deep feature fusion, the system can align students' potential ability traits (such as interdisciplinary integration tendencies, complex problem-solving patterns) with resource strategic demands (such as disruptive technology research and development needs, industry ecological reconstruction goals) in a unified semantic field. For example, the deep coding of the "system thinking paradigm" displayed in students' course assignments and the deep coding of the "technological paradigm change needs" implicit in corporate innovation projects are interacted in the semantic field to identify the potential synergistic value of students' cognitive patterns and corporate technology routes. By capturing the global collaborative pattern in the high-density semantic space, the system can foresee the long-term fit between students' growth trajectories and resource evolution paths. This deep semantic collaboration not only improves the accuracy of personalized recommendations, but also injects an innovation-driven core into the integration of industry and education by exploring the deep resonance between students' characteristics and industry trends, thus forming a new ecosystem of two-way empowerment between the supply of educational resources and industrial transformation and upgrading.

[0050] In a specific example of the present application, the deep implicit feature coding vector of the recommended resource and the deep implicit feature coding vector of the student portrait are deeply fused using the following feature fusion formula to obtain a student portrait-recommended resource deep-level fusion semantic feature coding vector; wherein the feature fusion formula is:

[0051] f deep =layerNorm(W o h+d1+d2)

[0052] h=GeLU[(d1U)⊙(d2V)]

[0053] Among them, W o represents the weight matrix of deep fusion, layerNorm(·) represents layer normalization, represents the GeLU function, U and V are low-rank projection matrices, f deep A semantic feature encoding vector is deeply integrated into the student portrait-recommended resource.

[0054] Furthermore, the semantic feature encoding vectors of the low-level fusion of student profiles and recommended resources, the mid-level fusion of student profiles and recommended resources, and the deep-level fusion of student profiles and recommended resources are subjected to a progressively complementary perception fusion of student profiles and recommended resources to obtain a semantic collaborative encoding vector for student profiles and recommended resources. It should be understood that low-level feature encoding preserves the precise correspondence between explicit ability indicators and resource access conditions (e.g., the numerical matching of credit points with course requirements), mid-level feature encoding reveals the topological adaptability of ability modules and demand structures (e.g., the mapping relationship between project experience combinations and technology stack dependencies), and deep-level feature encoding implies the semantic consistency between career potential and resource strategic value (e.g., the implicit resonance between innovative thinking patterns and industrial upgrading trends). Relying solely on a single level of features for decision-making will lead the recommendation system into the dilemma of detail bias, structural discontinuity, or semantic loss of focus. For example, excessive focus on course score matching may ignore the transfer value of students' project experience, or relying solely on deep semantic associations may miss candidate resources that meet the basic skill requirements. Therefore, in order to build a multi-scale collaborative decision-making system and achieve a full-dimensional dynamic balance between education supply and industry demand, in the technical solution of this application, the student portrait-recommended resource low-level fusion semantic feature coding vector, the student portrait-recommended resource middle-level fusion semantic feature coding vector and the student portrait-recommended resource deep-level fusion semantic feature coding vector are subjected to student portrait-recommended resource semantic progressive complementary perception fusion to obtain the student portrait-recommended resource semantic collaborative coding vector.

[0055] Specifically, low-level fusion features ensure the objectivity of explicit competence standards and establish a reliable baseline for the recommendation system; mid-level fusion features analyze the dynamic adaptation relationship between competence elements and resource structure, supporting personalized training path planning; deep-level fusion features capture the long-term synergistic value of development potential and strategic demands, driving the forward-looking allocation of industry-education resources. Through a progressive fusion mechanism, the system can verify and strengthen the contribution of features at different levels of abstraction layer by layer. For example, on the basis of ensuring that students meet the technical threshold of the project (low level), it further evaluates whether their existing experience modules support the needs of the project stage (middle level), and finally determines whether their innovation potential is in line with the direction of the enterprise's technology layout (deep level). This multi-level collaborative mechanism not only optimizes the accuracy of current resource matching, but also builds a dynamically evolving decision-making framework for industry-education integration by exploring the multi-scale relationship between student characteristics and industry trends, so that the education chain and the industrial chain form a deeply coupled innovation ecosystem at the micro-operation, meso-organization and macro-strategy levels.

[0056] In particular, it should be understood that low- and mid-level fusion features (such as explicit skill matching and modular capability structures) differ fundamentally from deep-level semantic features (such as professional value orientation and industrial strategic direction) in terms of information expression dimensions and timescales. Low- and mid-level fusion encoding focuses on the immediate adaptation of current capabilities and resource needs, while deep-level feature encoding maps long-term development potential and strategic synergy value. Direct global fusion can easily lead to conflicts between short-term needs and long-term goals.

[0057] Therefore, in a preferred example of the present application, in the process of performing student portrait-recommended resource semantic progressive complementary perception fusion on the student portrait-recommended resource low-level fusion semantic feature coding vector, the student portrait-recommended resource middle-level fusion semantic feature coding vector and the student portrait-recommended resource deep-level fusion semantic feature coding vector, first, the student portrait-recommended resource low-level fusion semantic feature coding vector and the student portrait-recommended resource middle-level fusion semantic feature coding vector are subjected to cross-level preliminary interactive fusion based on order parameter phase complementarity to obtain the student portrait-recommended resource cross-level preliminary fusion semantic feature coding vector; then, the student portrait-recommended resource cross-level preliminary fusion semantic feature coding vector and the student portrait-recommended resource deep-level fusion semantic feature coding vector are subjected to student portrait-recommended resource semantic interaction based on order parameter field constraint and path integral optimization to obtain the student portrait-recommended resource semantic collaborative coding vector. That is, through the order parameter phase complementarity mechanism, the system can identify dynamic phase differences between low- and mid-level features (e.g., a student's course grades meet the requirements but their project experience structure is incomplete), adjust feature weights in cross-level interactions, and form a preliminary fusion vector that preserves the satisfaction of explicit conditions while reflecting structural adaptation. Furthermore, path integral optimization is introduced, placing the preliminary fusion results and deep semantic features under the constraints of the order parameter field. Through dynamic adjustment of the integral path, the accumulation of non-integrability in feature interactions (e.g., the logical gap between short-term skill matching and long-term innovation potential) is eliminated, achieving progressive alignment of features at multiple time scales.

[0058] Specifically, the phase-complementary interaction between low- and mid-level layers dynamically adjusts the contribution weights of explicit condition verification and structural adaptation analysis through a gating mechanism. For example, while ensuring that students meet the hard skill requirements of a position, the recommendation priority is adjusted based on the modularity of project experience within the mid-level features. Deep path integral optimization, through field constraint modeling of strategic semantic continuity, transforms the interaction between students' innovative thinking patterns (deep coding) and the evolutionary trends of the enterprise's technological roadmap (resource deep coding) into an integrable path, thus avoiding inaccurate recommendations caused by strategic goal drift. This phased fusion mechanism enables the system to both respond to changes in students' current status and incorporate individual development into the overall perspective of the evolution of the industrial ecosystem. This cross-level fusion mechanism not only improves the accuracy of resource matching but also, through the co-evolutionary modeling of spatiotemporal features, builds a sustainably optimized dynamic adaptation ecosystem for industry-education integration, enabling two-way empowerment of talent development and industrial upgrading at both the micro-adaptation and macro-strategy levels.

[0059] In this example, the following semantic progressive complementary perception fusion formula is used to perform a student portrait-recommended resource semantic progressive complementary perception fusion on the student portrait-recommended resource low-level fusion semantic feature coding vector, the student portrait-recommended resource middle-level fusion semantic feature coding vector, and the student portrait-recommended resource deep-level fusion semantic feature coding vector to obtain a student portrait-recommended resource semantic collaborative coding vector; wherein, the semantic progressive complementary perception fusion formula is:

[0060]

[0061]

[0062] Among them, W g represents the gate weight matrix, Sigmoid(·) represents the Sigmoid function, G gate represents the gate order parameter, f low-mid Preliminary cross-level fusion semantic feature encoding vector for the student portrait-recommendation resource, W′ q Represents the optimized query weight matrix, W′ k Denotes the optimized key weight matrix, W′ v Represents the optimized value weight matrix, v q Indicates query, v k represents the key vector, v v represents the value vector, v f Generate a semantic collaborative encoding vector for the student portrait-recommended resource.

[0063] Specifically, in the above process, when the student portrait-recommended resource low-level fusion semantic feature coding vector, the student portrait-recommended resource middle-level fusion semantic feature coding vector and the student portrait-recommended resource deep-level fusion semantic feature coding vector are subjected to progressive complementary perceptual fusion, the student portrait-recommended resource low-level fusion semantic feature coding vector f low and student portrait-recommended resource mid-level fusion semantic feature encoding vector f mid Essentially generates the gate order parameter G gate , then based on the gating order parameter G gate The phase complementarity is used for fusion, and the semantic feature encoding vector f is further fused with the student portrait-recommendation resource deep level. deep Through the weight matrix W q 、W k and W v The order parameter field is constrained to achieve interaction. Here, the interaction is suppressed due to the instability of the constraint structure of the order parameter field, thereby affecting the expression effect of the student portrait-recommended resource semantic collaborative coding vector.

[0064] Therefore, for the pre-trained matrix W q 、W k and W v , first calculate the decomposition field path integral representation of the order parameter field combined with each matrix:

[0065]

[0066] Then, define the field uniform alignment loss function:

[0067]

[0068] where ‖·‖* represents the nuclear norm of the matrix, i.e., the sum of the eigenvalues ​​of the matrix. ω is a scaling hyperparameter.

[0069] In this way, the path integral of each decomposition field of the order parameter field is used to suppress the accumulation of non-integrability of the constraint connection of the order parameter field, so as to avoid the geometric phase accumulation that hinders the interaction of the local field structure. Then, the field uniformity equivalence of the sum of the eigenvalues ​​can be used to avoid the instability of the constraint structure solution, thereby realizing the weight matrix W q 、W k and W v The interaction is maintained to improve the expression effect of the student portrait-recommended resource semantic collaborative encoding vector.

[0070] Specifically, S52 performs feature decoding on the student portrait-recommended resource semantic collaborative coding vector to obtain a comprehensive fitness score. It should be understood that although the student portrait-recommended resource semantic collaborative coding vector, which has undergone multi-scale interactive fusion, contains full-dimensional correlation information between student ability traits and resource needs, its high-dimensional distributed representation is difficult to directly map into a sortable decision indicator. Traditional scoring mechanisms rely on manual rules or shallow regression, which cannot capture the dynamic coupling effect of career potential and industry trends in the deep semantic space, and can easily cause recommendation results to deviate from actual development needs.

[0071] Therefore, in order to build a conversion bridge from complex semantic understanding to precise decision-making output and realize the quantitative alignment of the value of education supply and industry demand, in the technical solution of this application, the student portrait-recommended resource semantic collaborative coding vector is passed through a comprehensive fitness generation model based on a decoder to obtain a comprehensive fitness score. In this process, the decoder reconstructs the core semantic elements of the student portrait-recommended resource semantic collaborative coding vector, and converts the implicit association between the student growth trajectory and the resource evolution path (such as the matching degree between the depth of academic research and the basic research investment of the enterprise) into an explicit fitness score. This conversion mechanism not only retains the global perspective of multi-level feature interaction, but also focuses on key fitness dimensions through the attention mechanism. For example, in the process of balancing skill matching and career development potential, the weight distribution of course recommendations and job recommendations is dynamically adjusted. Through the feature decoupling and reconstruction of the decoder, the system can identify non-explicit but strategically valuable fitness relationships. This quantitative mechanism not only improves the accuracy of personalized recommendations, but also provides a scientific and operational decision-making basis for the dynamic configuration of industry-education resources through the interpretable mapping of semantic collaborative coding to fitness scores, and promotes the formation of a closed-loop optimization ecology for talent training and industrial upgrading.

[0072] In particular, the S6 outputs the recommendable resource content data corresponding to the top K largest comprehensive fitness scores in the set of comprehensive fitness scores as a recommended resource content recommendation list. It should be understood that although the comprehensive fitness score set generated through multi-level semantic interaction quantifies the full-dimensional association between student characteristics and resource elements, the raw data that has not been sorted and optimized is difficult to directly convert into an operational recommendation strategy. For example, a student may be suitable for multiple technical research and development positions and innovative practice projects at the same time. If all high-scoring options are simply listed, it is easy to cause decision-making information overload and fail to focus on the core development path. By setting the K value threshold, the system can avoid the choice paradox caused by over-recommendation while retaining the diversity of personalized recommendations, ensuring that the recommendation list is both accurate and operational.

[0073] Therefore, in order to build a dynamically optimized recommendation decision-making mechanism and achieve the maximum value matching between education supply and industry demand, in the technical solution of this application, the recommendable resource content data corresponding to the top K maximum comprehensive fitness scores in the set of comprehensive fitness scores are output as a recommendation list of recommendable resource content. It is worth mentioning that the K value screening mechanism avoids missing potential highly adaptable resources (such as the transferable ability requirements implied in cross-disciplinary positions) by weighing the breadth and depth of recommendations, and prevents the recommendation list from being too long and weakening the decision-making efficiency. For example, for students with interdisciplinary research potential, the system retains the recommendation of core professional positions while incorporating a small number of cutting-edge cross-disciplinary opportunities through K value control to guide them to explore multiple development paths. This mechanism design enables the recommendation system to meet immediate adaptation needs while reserving flexible space for career development. This dynamically balanced recommendation strategy not only improves the efficiency of individual growth, but also promotes the deep coupling of the education chain and the industrial chain through the efficient allocation of resources, and builds a sustainable and optimized intelligent decision-making closed loop for the integration of industry and education.

[0074] In summary, the industry-education integration management method based on big data analysis according to the embodiment of this application is explained. By acquiring and deeply mining the multi-dimensional portrait data of students and the content data of recommended resources, the characterization of student portraits and the semantic characterization of resource content are realized, and the multi-scale semantic progressive interaction technology is used to effectively improve the adaptability between portraits and recommended resources. In this way, not only can accurate recommendations for individual differences be achieved, but also changes in students' growth trajectories can be dynamically responded to, providing scientific decision-making basis for colleges and enterprises.

[0075] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A production-education integration management method based on big data analysis, characterized in that: include: Obtain multi-dimensional portrait data of target students; Get a collection of recommended resource content data; Performing student portrait characterization on the multi-dimensional portrait data of the target student object to obtain the current state portrait vector of the target student; Performing resource content characterization on each piece of recommendable resource content data in the set of recommendable resource content data to obtain a set of recommendable resource content semantic feature vectors; Inputting each of the recommended resource content semantic feature vectors and the target student's current status portrait vector in the set of recommended resource content semantic feature vectors into a portrait-recommended resource matching analysis network to obtain a set of comprehensive fitness scores; The recommended resource content data corresponding to the top K maximum comprehensive fitness scores in the set of comprehensive fitness scores are output as a recommended resource content recommendation list.

2. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 1 is characterized in that: Student multi-dimensional portrait data includes student portrait structured data and student portrait unstructured data; among them, student portrait structured data includes academic performance, courses taken and major direction; student portrait unstructured data includes project reports, course assignment summaries, interest self-descriptions, and career goal self-descriptions.

3. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 2 is characterized in that: Perform student portrait characterization on the multi-dimensional portrait data of the target student object to obtain the current state portrait vector of the target student, including: The student portrait structured data is passed through the student state portrait encoder based on the BERT model to obtain the current state portrait vector of the first target student; Performing structured encoding on each unstructured data in the student portrait unstructured data to obtain a set of student portrait embedding encoding vectors; The set of student portrait embedding encoding vectors is passed through the student state portrait encoder based on the BERT model to obtain the current state portrait vector of the second target student; The first target student's current state portrait vector and the second target student's current state portrait vector are fused to obtain the target student's current state portrait vector.

4. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 3 is characterized in that: The collection of recommendable resource content data includes recommendable resource content structured data and recommendable resource content unstructured data; among them, the recommendable resource content structured data includes course prerequisites, a list of technology stacks required for the project, and hard requirements for the position; the recommendable resource content unstructured data includes course outlines, project descriptions, and job responsibilities.

5. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 4 is characterized in that: Input each of the recommended resource content semantic feature vectors and the target student's current status portrait vector in the set of recommended resource content semantic feature vectors into the portrait-recommended resource matching analysis network to obtain a set of comprehensive adaptation scores, including: Perform multi-scale semantic progressive interaction on the semantic feature vectors of the recommended resource content and the target student's current status portrait vector to obtain the student portrait-recommended resource semantic collaborative encoding vector; Feature decoding is performed on the student portrait-recommended resource semantic co-encoding vector to obtain a comprehensive fitness score.

6. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 5 is characterized in that: Multi-scale semantic progressive interaction is performed on the semantic feature vectors of the recommended resource content and the target student's current status portrait vector to obtain the student portrait-recommended resource semantic collaborative encoding vector, including: Perform deep nonlinear transformation on the semantic feature vector of the recommended resource content and the target student's current status portrait vector respectively to obtain the recommended resource mid-level latent feature coding vector, the student portrait mid-level latent feature coding vector, the recommended resource deep-level latent feature coding vector and the student portrait deep-level latent feature coding vector; Perform multi-level cross-modal semantic fusion on the semantic feature vectors of the recommended resource content and the target student's current status portrait vector to obtain the student portrait-recommended resource low-level fusion semantic feature encoding vector, the student portrait-recommended resource mid-level fusion semantic feature encoding vector, and the student portrait-recommended resource deep-level fusion semantic feature encoding vector; The student portrait-recommended resource low-level fusion semantic feature coding vector, the student portrait-recommended resource mid-level fusion semantic feature coding vector and the student portrait-recommended resource deep-level fusion semantic feature coding vector are subjected to student portrait-recommended resource semantic progressive complementary perception fusion to obtain the student portrait-recommended resource semantic collaborative coding vector.

7. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 6 is characterized in that: Perform deep nonlinear transformation on the semantic feature vector of the recommended resource content and the target student's current status portrait vector respectively to obtain the recommended resource mid-level latent feature coding vector, the student portrait mid-level latent feature coding vector, the recommended resource deep-level latent feature coding vector and the student portrait deep-level latent feature coding vector, including: Multi-level implicit feature extraction is performed on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain the mid-level implicit feature coding vector of the recommended resource, the mid-level implicit feature coding vector of the student portrait, the deep-level implicit feature coding vector of the recommended resource and the deep-level implicit feature coding vector of the student portrait.

8. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 7 is characterized in that: Multi-level cross-modal semantic fusion is performed on the semantic feature vectors of the recommended resource content and the target student's current status portrait vector to obtain the student portrait-recommended resource low-level fusion semantic feature encoding vector, the student portrait-recommended resource mid-level fusion semantic feature encoding vector, and the student portrait-recommended resource deep-level fusion semantic feature encoding vector, including: Perform low-level feature fusion on the semantic feature vector of the recommended resource content and the target student's current status portrait vector to obtain a student portrait-recommended resource low-level fusion semantic feature encoding vector; Performing mid-level feature fusion on the mid-level latent feature coding vector of the recommended resources and the mid-level latent feature coding vector of the student portrait to obtain the mid-level fused semantic feature coding vector of the student portrait-recommended resources; Deep feature fusion is performed on the deep latent feature coding vector of the recommended resources and the deep latent feature coding vector of the student portrait to obtain the deep-level fusion semantic feature coding vector of the student portrait and the recommended resources.

9. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 8 is characterized in that: The student portrait-recommended resource low-level fusion semantic feature coding vector, the student portrait-recommended resource mid-level fusion semantic feature coding vector, and the student portrait-recommended resource deep-level fusion semantic feature coding vector are subjected to student portrait-recommended resource semantic progressive complementary perception fusion to obtain the student portrait-recommended resource semantic collaborative coding vector, including: Performing cross-level preliminary interactive fusion of the student portrait-recommended resource low-level fusion semantic feature coding vector and the student portrait-recommended resource mid-level fusion semantic feature coding vector based on order parameter phase complementarity to obtain the student portrait-recommended resource cross-level preliminary fusion semantic feature coding vector; The student portrait-recommendation resource cross-level preliminary fusion semantic feature coding vector and the student portrait-recommendation resource deep-level fusion semantic feature coding vector are subjected to student portrait-recommendation resource semantic interaction based on order parameter field constraint and path integral optimization to obtain the student portrait-recommendation resource semantic collaborative coding vector.

10. The power equipment diagnosis method integrating knowledge graph and deep learning according to claim 9 is characterized in that: Perform feature decoding on the student profile-recommended resource semantic co-encoding vector to obtain a comprehensive fitness score, including: The student portrait-recommended resource semantic collaborative encoding vector is passed through a decoder-based comprehensive fitness generation model to obtain a comprehensive fitness score.

Images