Big data-based college student cultural heritage quality evaluation system

Abstract

The invention relates to the technical field of big data and education informatization, in particular to a college student cultural heritage quality evaluation system based on big data, and aims to solve the problems of inaccurate cultural heritage quality measurement, non-traceable process and non-operable intervention in the prior art. The system comprises a data acquisition layer, a data governance layer, a literacy modeling layer, a portrait generation layer and a feedback intervention layer, and multi-source education scene behavior data acquisition, Bloom cognitive dimension-based behavior labeling, four-dimensional dynamic literacy index modeling, knowledge graph portrait embedding and personalized intervention suggestion generation are carried out. And whole-process dynamic evaluation and intelligent feedback of cultural heritage quality of college students are realized. Through adoption of the technical scheme, the evaluation accuracy, interpretability and educational intervention effectiveness can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of big data and educational informatization technology, and particularly relates to a college student cultural heritage literacy evaluation system based on big data. BACKGROUND

[0002] With the increasing emphasis of the state on the integration of cultural heritage and education, improving the cultural heritage literacy of college students has become an important part of quality education in colleges and universities. Cultural heritage literacy not only involves the cognition, understanding and identification of material and intangible cultural heritage, but also covers multi-dimensional indicators such as protection awareness, communication ability and practice participation. Under this background, a scientific, dynamic and quantifiable evaluation system is urgently needed to support the accurate assessment and individualized training of college students' cultural heritage literacy level in colleges and universities. Big data technology provides a technical possibility for building such an evaluation system due to its advantages in multi-source data collection, behavior trajectory analysis and intelligent modeling. Through retrieval, a regional intangible cultural heritage management method based on big data with publication number CN119323502B is disclosed. This patent focuses on the monitoring and optimization of regional intangible cultural heritage publicity, and constructs a heat analysis model by collecting publicity information, browsing behavior and activity data, and outputs publicity suggestions. However, the service object of this technical solution is the cultural management department or the public, and its core goal is to improve the communication efficiency of intangible cultural heritage projects, rather than to quantitatively evaluate the cultural cognition level or literacy ability of specific groups (such as college students). In addition, this method does not involve key literacy dimensions such as learning behavior, knowledge mastery, value identification in the education scenario, lacks modeling and feedback mechanisms for individual literacy development trajectories, and is difficult to support the individualized and process-oriented analysis needs required by education evaluation.

[0003] On the other hand, a stratified analysis method and device based on historical and cultural data with publication number CN111651506B are disclosed. This patent constructs a scoring analysis model to spatially display the historical and cultural resources in terms of historical duration, influence, protection range, etc., to support the comprehensive management of cultural heritage. Although this solution introduces scoring parameters and visual analysis, its evaluation object is the cultural heritage itself and its regional distribution characteristics, rather than the literacy status of people. Its scoring system is based on objective heritage attributes and does not include key indicators such as learners' subjective cognition, attitude, behavior, etc. that reflect literacy levels, nor does it combine education data (such as course scores, practical activities, questionnaire feedback, digital platform interaction records, etc.) for multi-modal fusion analysis, so it cannot achieve a dynamic, comprehensive and interpretable evaluation of college students' individual or group cultural heritage literacy.

[0004] The above problems show that although the prior art has made certain progress in cultural heritage data management, knowledge organization and regional analysis, there is still a significant gap in the "humanized" literacy evaluation oriented to the higher education scene: first, there is a lack of targeted design with college students as the evaluation subject; second, a multi-dimensional literacy index system covering cognition, emotion, behavior, etc. is not constructed; third, the education process data and cultural resource data are not effectively integrated for intelligent modeling. Therefore, the present application proposes a college student cultural heritage literacy evaluation system based on big data, aiming to realize scientific quantification, process tracking and personalized feedback of college student cultural heritage literacy through multi-source education behavior data collection, multi-dimensional literacy index modeling and dynamic evaluation algorithm, so as to fill the gap of the prior art in the field of cultural education evaluation. SUMMARY

[0005] The present application provides a college student cultural heritage literacy evaluation system based on big data, aiming to solve the technical problems of inaccurate cultural heritage literacy measurement, untraceable process and inoperable intervention caused by the lack of multi-dimensional dynamic evaluation mechanism centered on learners in the prior art. In order to achieve the above application purpose, the present application constructs a systematic architecture integrating educational theory, cultural psychology model and big data intelligent analysis technology. The system collects the whole process behavior data generated by college students in multi-source heterogeneous education scenes such as course learning, campus cultural activities, social practice and digital platform interaction, carries out structured modeling based on the preset multi-dimensional cultural heritage literacy index system, and realizes dynamic portrait, difference identification and intelligent feedback of individual literacy state by using knowledge graph association reasoning and deep learning algorithm.

[0006] The system includes five functional levels of data collection layer, data governance layer, literacy modeling layer, portrait generation layer and feedback intervention layer, and realizes seamless connection of data flow and control flow between each level through standardized data interface and unified event bus.

[0007] The data acquisition layer is deployed on top of the university's existing IT infrastructure and specifically includes a teaching management system interface module, an online learning platform log acquisition module, a campus cultural activity attendance and evaluation module, a social practice task submission and review module, and a mobile cultural heritage interactive application data capture module. The teaching management system interface module connects to the university's academic affairs system API via the OAuth 2.0 protocol to obtain student course selection records, course grades, assignment submission status, and teacher comments in real time. The online learning platform log acquisition module uses the Fluentd log proxy program deployed on the MOOC platform or the university's SPOC platform server node to capture user page dwell time, video playback progress, chapter quiz answer sequence, discussion forum post content, and like / reply behavior. The campus cultural activity attendance and evaluation module integrates an NFC near-field communication chip and a QR code scanning terminal to record students' attendance time, dwell time, and subjective feedback text submitted via on-site QR code scanning for activities such as museum tours, intangible cultural heritage workshops, and cultural lectures. The social practice task submission and review module uses RESTful... The API receives unstructured documents uploaded by students, such as field investigation reports, oral history interview recordings, and cultural heritage protection plans, and simultaneously obtains structured grading items from instructors. The mobile cultural heritage interactive application data capture module is embedded into the official cultural apps of universities through the SDK to collect fine-grained behavioral events such as user navigation paths in virtual exhibition halls, interactive operations on 3D models of cultural relics, response times and error patterns in cultural knowledge Q&A sessions.

[0008] Furthermore, the data governance layer performs cleaning, alignment, labeling, and vectorization processing on the raw data from the data acquisition layer. This layer includes a data cleaning unit, an entity alignment unit, a behavior labeling unit, and a feature vector generation unit: The data cleaning unit, based on a predefined set of data quality rules, removes session records with a missing rate exceeding 85%, corrects event sequences with timestamp logical conflicts, and filters high-frequency click logs generated by bot traffic; The entity alignment unit performs cross-domain association of data records scattered across various subsystems based on the student's unified identity identifier (UID), where the UID is generated by a one-way mapping of the 18-digit student ID provided by the university's campus card system using the SHA-256 hash function to ensure privacy compliance; The behavior labeling unit assigns semantic labels to the raw behavior events, for example, labeling "correctly answering a multiple-choice question about the historical background of a certain cultural heritage" as the "understanding" dimension, and labeling "providing a critical analysis of the reasons for the endangerment of a certain intangible cultural heritage inheritor's skills in the discussion area" as the "evaluation" dimension; The feature vector generation unit converts the labeled structured behavior sequences into fixed-length numerical vectors, specifically implemented as follows: For each cognitive dimension Calculate the frequency of student i's behavior in this dimension within the time window t. Average response time Accuracy and contextual diversity index ,in The Shannon entropy is defined as the category of cultural heritage themes involved in this dimension. In the formula Let K be the percentage of student i's actions within time window t that involve the k-th category of cultural heritage themes (such as traditional architecture, folk literature, festival customs, etc., totaling K=12 categories) under the cognitive dimension d. Finally, the feature vector for each student in each time window t is represented as follows: .

[0009] As a preferred embodiment of the present invention, the literacy modeling layer constructs a four-dimensional cultural heritage literacy index system. This system consists of four primary indicators: knowledge mastery, cognitive attitude, practical participation, and value recognition. Each primary indicator has several secondary indicators, and all indicators are coupled with Piaget's theory of cognitive development stages and the internalization-externalization model in cultural psychology.

[0010] Specifically, the knowledge mastery includes three secondary indicators: breadth of basic cultural heritage knowledge, depth of understanding of core concepts, and cross-cultural comparative ability; the cognitive attitude includes three secondary indicators: cultural openness, historical reverence, and critical reflection tendency; the practical participation includes three secondary indicators: frequency of extracurricular cultural activities, intensity of social practice investment, and activity level on digital platforms; and the value identification includes three secondary indicators: sense of national identity, sense of responsibility for cultural inheritance, and willingness to innovate and transform. This indicator system is not a static weighting, but rather uses an interpretable machine learning model to dynamically calibrate the contribution of each indicator to the overall literacy level.

[0011] Specifically, the competency modeling layer deploys a two-stage regression model: The first stage uses Gradient Boosting Decision Tree (GBDT) to learn the non-linear mapping relationship between each secondary indicator and the manually assessed literacy level from historical labeled samples; The second stage introduces the SHAP (SHapley Additive exPlanations) value decomposition mechanism to calculate the marginal contribution weight of each secondary indicator on a specific student sample. , This allows for individualized weight allocation; Overall competence score The calculation formula is: ; in The standardized score of student i for the j-th secondary indicator within the time window t is normalized to the [0,1] interval using Min-Max.

[0012] Furthermore, the portrait generation layer, based on the dynamic score sequence output by the literacy modeling layer and combined with the cultural heritage ontology knowledge graph, generates multi-granularity, interpretable individual literacy portraits. The portrait generation layer includes ontology construction units, graph embedding units, and portrait reasoning units. The ontology construction unit constructs a cultural heritage ontology containing 12 top-level categories, 87 mid-level subclasses, and 532 specific project instances. This ontology is formally described using the OWL2DL language and defines 17 object attributes such as "belongs to", "influence", "inherited from", and "geographical distribution", as well as 9 data attributes such as "historical era", "endangered status", and "core techniques". The knowledge graph embedding unit maps entities and relations in the ontology to a low-dimensional dense vector space. The RotatE model is used for knowledge graph embedding training, and its loss function is negative sampled cross-entropy. ; in For the set of positive sample triples, The spurious tail entity generated by negative sampling. This represents the Euclidean distance between the head entity h and the tail entity t after rotation by relation r. For boundary hyperparameters, The function is sigmoid. The profiling and reasoning unit then uses the student's behavioral feature vector. Cross-modal alignment with the graph embedding vector is achieved through a dual-tower neural network architecture: the left tower receives... The output is a behavioral semantic vector after passing through three fully connected layers (512, 256, and 128 dimensions per layer, with ReLU activation). The tower on the right receives a collection of cultural heritage entities that students have recently interacted with. The average of its embedding vectors is obtained Then, a fully connected layer with the same structure is used to output a cultural context vector. ; Final image vector Based on this image vector, the system can generate three types of output: (1) A competency radar chart shows the relative strengths and weaknesses of the four primary indicators; (2) Knowledge gap prompts, through calculation The cosine similarity with the center vector of each cultural heritage subdomain is used to identify domains with a similarity below the threshold of 0.3. (3) Development trajectory curve, plotted The study analyzed the changing trends over consecutive semesters and fitted a quadratic polynomial to predict future trends in literacy.

[0013] As one of the core innovations of this invention, the feedback intervention layer automatically generates personalized and actionable educational intervention suggestions based on the output of the profile generation layer, and pushes them to students, counselors, and teachers through multiple channels. This layer includes a rule engine unit, a natural language generation unit, and a multi-terminal distribution unit. The rule engine unit pre-configures a decision rule base based on educational intervention theory, for example: If the value identification score is below 0.4 for two consecutive semesters and the participation in practice decreases by more than 20%, the 'cultural immersion reinforcement' intervention package will be triggered. If the critical reflection tendency score is higher than 0.7 but the knowledge mastery is lower than 0.5, the 'deep reading guidance' intervention package will be triggered. Each intervention package includes a list of recommended resources, a list of action tasks, and expected outcome indicators; The natural language generation unit uses a fine-tuned T5-base model to convert structured intervention instructions into natural language text that fits the educational context. Its input format is "[Student Name][Current Weakness][Recommended Action][Expected Improvement]", and the output example is: "Student XX, you have insufficient knowledge of traditional festival customs. It is recommended that you participate in the Dragon Boat Festival folk custom experience workshop this Saturday and complete the online knowledge self-test. It is expected that you can improve your relevant knowledge score by more than 0.2 within two weeks." The multi-terminal distribution unit pushes intervention messages to students' personal mobile apps in real time via a WebSocket long connection. At the same time, it synchronizes summary information to the counselor's workbench and embeds the task into the assignment module of the online learning platform of the relevant course through the LTI (Learning Tools Interoperability) standard protocol.

[0014] Furthermore, to ensure the reliability and scalability of the system, this invention adopts a microservice design pattern at the system architecture level. Each module of the data acquisition layer is encapsulated as an independent Docker container and elastically scheduled via a Kubernetes cluster; the data governance layer and literacy modeling layer are deployed on the Apache Spark on YARN computing framework, supporting batch and stream processing of terabyte-level logs; the graph embedding model training of the profile generation layer is executed on an NVIDIA A100 GPU cluster, and the inference service is exposed via a gRPC interface after optimization using TensorRT; the rule engine of the feedback intervention layer is implemented based on Drools 8.0, and the natural language generation model is hosted on the Triton Inference Server. All inter-service communication uses Protocol Buffers serialization format and is decoupled through an Apache Kafka message queue to ensure high throughput and low latency.

[0015] Regarding data security and privacy protection, the raw behavioral data is de-identified at the collection end, and students' real identity information is only stored in the university's unified identity authentication center. The system only uses irreversible hashed UIDs for association. All feature vectors and profile data are encrypted using TLS 1.3 during transmission and AES-256-GCM algorithm for storage. The system has a three-level access control policy: students can only view their own profiles and suggestions, counselors can view the statistical view of the classes under their jurisdiction, and researchers can only access the de-identified aggregated dataset after approval by the ethics committee.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention, by constructing a system architecture centered on university students, based on educational process data, and framed by a multidimensional literacy model, achieves a fundamental shift in cultural heritage literacy evaluation from a static, macroscopic, and object-oriented approach to a dynamic, microscopic, and subject-oriented one. The system not only addresses the shortcomings of existing technologies, such as data fragmentation, singular indicators, and delayed feedback, but also, through the deep integration of knowledge graphs and deep learning, enables interpretable characterization of literacy states and precise generation of educational interventions, providing solid technical support for the scientific and intelligent development of cultural education in universities. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a schematic diagram showing the module composition and data sources of the data acquisition layer of the present invention; Figure 3 This is a schematic diagram of the data processing flow and feature vector generation logic of the data governance layer of the present invention; Figure 4 This is a schematic diagram of the four-dimensional cultural heritage literacy index system and dynamic weight calculation mechanism of the literacy modeling layer of the present invention; Figure 5 This is a schematic diagram of the knowledge graph embedding and multimodal portrait reasoning structure of the portrait generation layer of this invention; Figure 6 This is a schematic diagram of the rule triggering, natural language generation, and multi-terminal distribution mechanism of the feedback intervention layer of this invention. Detailed Implementation

[0018] This invention provides a big data-based evaluation system for college students' cultural heritage literacy, the overall architecture of which is shown in the attached figure. Figure 1 As shown, it consists of five functional layers: data acquisition layer, data governance layer, literacy modeling layer, profile generation layer, and feedback intervention layer. Each layer is seamlessly connected to the others via standardized data interfaces and a unified event bus, ensuring a closed-loop operation from raw behavioral data collection to the generation of personalized educational intervention suggestions.

[0019] The data acquisition layer is deployed on top of the university's existing IT infrastructure and specifically includes a teaching management system interface module, an online learning platform log collection module, a campus cultural activity attendance and evaluation module, a social practice task submission and review module, and a mobile cultural heritage interactive application data capture module. The module composition and data source relationships are shown in the attached figure. Figure 2 As shown. The teaching management system interface module connects to the university's academic affairs system API via the OAuth 2.0 protocol, obtaining student course selection records, course grades, assignment submission status, and teacher comments in real time. The online learning platform log collection module uses the Fluentd log broker deployed on the MOOC platform or the university's SPOC platform server node to capture user page dwell time, video playback progress, chapter quiz answer sequence, discussion forum post content, and like / reply behavior. The campus cultural activity check-in and evaluation module integrates an NFC near-field communication chip and a QR code scanning terminal to record students' attendance time, dwell time, and subjective feedback text submitted on-site via QR code scanning for activities such as museum tours, intangible cultural heritage workshops, and cultural lectures. The social practice task submission and review module receives unstructured documents uploaded by students, such as field investigation reports, oral history interview recordings, and cultural heritage protection plans, via a RESTful API, and simultaneously obtains the structured evaluation items of the instructors. The mobile cultural heritage interactive application data capture module is embedded in the university's official cultural APP via SDK, collecting fine-grained behavioral events such as user roaming paths in virtual exhibition halls, interactive operations on 3D models of cultural relics, response time and error patterns in cultural knowledge Q&A.

[0020] Furthermore, the data governance layer performs cleaning, alignment, labeling, and vectorization processing on the raw data from the data acquisition layer. Its data processing flow and feature vector generation logic are shown in the attached figure. Figure 3 As shown. This layer includes a data cleaning unit, an entity alignment unit, a behavior labeling unit, and a feature vector generation unit. The data cleaning unit, based on a predefined set of data quality rules, removes session records with a missing rate exceeding 85%, corrects event sequences with timestamp logical conflicts, and filters high-frequency click logs generated by bot traffic. The entity alignment unit uses a unified student identity identifier to cross-domain associate data records scattered across various subsystems. The UID is generated by a one-way mapping of the 18-digit student ID provided by the university's campus card system using a SHA-256 hash function, ensuring privacy compliance. The behavior labeling unit assigns semantic labels to the original behavior events; for example, "correctly answering a multiple-choice question about the historical background of a cultural heritage" is labeled as the "understanding" dimension, and "providing a critical analysis of the reasons for the endangerment of a certain intangible cultural heritage inheritor's skills in the discussion area" is labeled as the "evaluation" dimension. The feature vector generation unit converts the labeled structured behavior sequences into fixed-length numerical vectors. Its specific implementation is as follows: For each cognitive dimension Calculate the frequency of student i's behavior in this dimension within the time window t. Average response time Accuracy and contextual diversity index ,in The Shannon entropy is defined as the category of cultural heritage themes involved in this dimension. In the formula Let K be the percentage of student i's actions within time window t that involve the k-th category of cultural heritage themes (such as traditional architecture, folk literature, festival customs, etc., totaling K=12 categories) under the cognitive dimension d. Finally, the feature vector for each student in each time window t is represented as follows: .

[0021] In a preferred embodiment of the present invention, the literacy modeling layer constructs a four-dimensional cultural heritage literacy index system. This system consists of four primary indicators: knowledge mastery, cognitive attitude, practical participation, and value recognition. Each primary indicator has several secondary indicators, and all indicators are coupled with Piaget's theory of cognitive development stages and the internalization-externalization model in cultural psychology. Specifically, knowledge mastery includes three secondary indicators: breadth of basic cultural heritage knowledge, depth of understanding of core concepts, and cross-cultural comparison ability; cognitive attitude includes three secondary indicators: cultural openness, historical reverence, and critical reflection tendency; practical participation includes three secondary indicators: frequency of extracurricular cultural activities, intensity of social practice investment, and activity level on digital platforms; and value recognition includes three secondary indicators: sense of national identity, sense of responsibility for cultural inheritance, and willingness to innovate and transform. This index system is not a static weight allocation, but rather uses an interpretable machine learning model to dynamically calibrate the contribution of each indicator to the overall literacy level. Specifically, the competency modeling layer deploys a two-stage regression model: the first stage uses Gradient Boosting Decision Tree (GBDT) to learn the non-linear mapping relationship between each secondary indicator and the competency level manually assessed by experts from historical labeled samples; the second stage introduces the SHAP (SHapley Additive exPlanations) value decomposition mechanism to calculate the marginal contribution weight of each secondary indicator on a specific student sample. , This allows for individualized weight allocation.

[0022] Overall competence score The calculation formula is: ; in The standardized score of student i for the j-th secondary indicator within the time window t is normalized to the [0,1] interval using Min-Max. The four-dimensional indicator system and dynamic weight calculation mechanism are shown in the appendix.Figure 4 As shown.

[0023] Furthermore, the portrait generation layer, based on the dynamic score sequence output by the literacy modeling layer and combined with the cultural heritage ontology knowledge graph, generates multi-granular, interpretable individual literacy portraits. Its knowledge graph embedding and multimodal portrait reasoning structure are shown in the appendix. Figure 5 As shown. This layer contains ontology construction units, graph embedding units, and image reasoning units.

[0024] The ontology construction unit constructs a cultural heritage ontology containing 12 top-level categories, 87 mid-level subclasses, and 532 specific project instances. This ontology is formally described using the OWL2DL language and defines 17 object attributes such as "belongs to", "influence", "inherited from", and "geographical distribution", as well as 9 data attributes such as "historical era", "endangered status", and "core techniques". The knowledge graph embedding unit maps entities and relations in the ontology to a low-dimensional dense vector space. The RotatE model is used for knowledge graph embedding training, and its loss function is negative sampled cross-entropy. ; in For the set of positive sample triples, The spurious tail entity generated by negative sampling. This represents the Euclidean distance between the head entity h and the tail entity t after rotation by relation r. For boundary hyperparameters, The function is sigmoid. The profiling and reasoning unit then uses the student's behavioral feature vector. Cross-modal alignment with the graph embedding vector is achieved through a dual-tower neural network architecture: the left tower receives... The output is a behavioral semantic vector after passing through three fully connected layers (512, 256, and 128 dimensions per layer, with ReLU activation). The tower on the right receives a collection of cultural heritage entities that students have recently interacted with. The average of its embedding vectors is obtained Then, a fully connected layer with the same structure is used to output a cultural context vector. ; Final image vector Based on this image vector, the system can generate three types of output: (1) A competency radar chart shows the relative strengths and weaknesses of the four primary indicators; (2) Knowledge gap prompts, through calculation The cosine similarity with the center vector of each cultural heritage subdomain is used to identify domains with a similarity below the threshold of 0.3. (3) Development trajectory curve, plotted The study analyzed the changing trends over consecutive semesters and fitted a quadratic polynomial to predict future trends in literacy.

[0025] As one of the core innovations of this invention, the feedback intervention layer automatically generates personalized and actionable educational intervention suggestions based on the output of the profile generation layer, and pushes them to students, counselors, and teachers through multiple channels. Its rule triggering, natural language generation, and multi-terminal distribution mechanisms are detailed in the appendix. Figure 6 As shown. This layer includes a rule engine unit, a natural language generation unit, and a multi-terminal distribution unit. The rule engine unit has a pre-built decision rule base based on educational intervention theory. For example, if the value recognition score is below 0.4 for two consecutive semesters and the practical participation decreases by more than 20%, the 'cultural immersion reinforcement' intervention package is triggered; if the critical reflection tendency score is above 0.7 but the knowledge mastery is below 0.5, the 'deep reading guidance' intervention package is triggered. Each intervention package includes a list of recommended resources, a list of action tasks, and expected performance indicators. The natural language generation unit uses a fine-tuned T5-base model to convert structured intervention instructions into natural language text that conforms to the educational context. Its input format is "[Student Name][Current Weakness][Recommended Action][Expected Improvement]", and the output example is: "Student XX, you have insufficient knowledge of traditional festival customs. It is recommended that you participate in the Dragon Boat Festival folk custom experience workshop this Saturday and complete the online knowledge self-test. It is expected that you can improve your relevant knowledge score by more than 0.2 within two weeks." The multi-terminal distribution unit pushes intervention messages to students' personal mobile apps in real time via WebSocket long connections, while simultaneously synchronizing summary information to the counselor's workbench via the WeChat API, and embedding the task into the assignment module of the online learning platform for relevant courses via the LTI (Learning Tools Interoperability) standard protocol.

[0026] Furthermore, to ensure the reliability and scalability of the system, this invention adopts a microservice design pattern at the system architecture level. Each module of the data acquisition layer is encapsulated as an independent Docker container and elastically scheduled via a Kubernetes cluster; the data governance layer and literacy modeling layer are deployed on the Apache Spark on YARN computing framework, supporting batch and stream processing of terabyte-level logs; the graph embedding model training of the profile generation layer is executed on an NVIDIA A100 GPU cluster, and the inference service is exposed via a gRPC interface after optimization using TensorRT; the rule engine of the feedback intervention layer is implemented based on Drools 8.0, and the natural language generation model is hosted on the Triton Inference Server. All inter-service communication uses Protocol Buffers serialization format and is decoupled through an Apache Kafka message queue to ensure high throughput and low latency.

[0027] Regarding data security and privacy protection, the raw behavioral data is de-identified at the collection end, and students' real identity information is only stored in the university's unified identity authentication center. The system only uses irreversible hashed UIDs for association. All feature vectors and profile data are encrypted using TLS 1.3 during transmission and AES-256-GCM algorithm for storage. The system has a three-level access control policy: students can only view their own profiles and suggestions, counselors can view the statistical view of the classes under their jurisdiction, and researchers can only access the de-identified aggregated dataset after approval by the ethics committee.

[0028] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A big data-based evaluation system for college students' cultural heritage literacy, characterized in that, The system comprises a data acquisition layer, a data governance layer, a literacy modeling layer, a profile generation layer, and a feedback intervention layer connected in sequence. The data acquisition layer is deployed on the university's information infrastructure and is used to collect student behavior data in real time from multiple educational scenarios. The data governance layer cleans, aligns entities, labels cognitive dimensions, and vectorizes the behavioral data to generate a fixed-length behavioral feature vector. The literacy modeling layer is based on a four-dimensional cultural heritage literacy index system and combines an interpretable machine learning model to dynamically calculate the overall literacy score of individuals. The portrait generation layer integrates the behavioral feature vector with the cultural heritage ontology knowledge graph to generate a multi-granular, interpretable individual literacy portrait. The feedback intervention layer automatically generates personalized educational intervention suggestions based on the literacy profile and pushes them to students, counselors, and teachers through multiple channels.

2. The big data-based evaluation system for college students' cultural heritage literacy as described in claim 1, characterized in that, The data acquisition layer includes a teaching management system interface module, an online learning platform log acquisition module, a campus cultural activity check-in and evaluation module, a social practice task submission and review module, and a mobile cultural heritage interactive application data capture module. The teaching management system interface module connects to the university's academic affairs system API via the OAuth 2.0 protocol to obtain student course selection records, course grades, assignment submission status, and teacher comments. The online learning platform log collection module uses Fluentd log agent program deployed on MOOC or SPOC platform server nodes to capture page dwell time, video playback progress, quiz answer sequence and discussion forum interaction behavior; The campus cultural activity check-in and evaluation module integrates an NFC chip and a QR code scanning terminal to record students' arrival time, stay duration, and subjective feedback text when participating in museum tours, intangible cultural heritage workshops, and cultural lectures. The social practice task submission and review module receives field investigation reports, oral history recordings, and preservation plan documents via RESTful API, and simultaneously obtains structured scores from teachers. The mobile cultural heritage interactive application data capture module is embedded into a university cultural APP through an SDK to collect data on virtual exhibition hall roaming paths, interactive operations of 3D models of cultural relics, and response time and error patterns of knowledge questions and answers.

3. The big data-based evaluation system for college students' cultural heritage literacy as described in claim 1, characterized in that, The data governance layer includes a data cleaning unit, an entity alignment unit, a behavior annotation unit, and a feature vector generation unit. The data cleaning unit removes session records with a missing rate exceeding 85%, corrects timestamp conflict events, and filters robot traffic logs based on a predefined set of data quality rules. The entity alignment unit associates cross-subsystem data records based on the student's unified identity identifier UID generated by hashing the student ID number of the university's campus card using SHA-256. The behavior labeling unit assigns semantic labels to behavioral events based on the sixth cognitive dimension of Bloom's Taxonomy of Educational Objectives. The feature vector generation unit targets each cognitive dimension. Calculate the frequency of student behavior within the time window t. Average response time Accuracy and contextual diversity index ,in The Shannon entropy is calculated for K=12 cultural heritage themes and combined into a 24-dimensional feature vector. .

4. The big data-based evaluation system for college students' cultural heritage literacy as described in claim 1, characterized in that, The four-dimensional cultural heritage literacy index system constructed by the literacy modeling layer includes four primary indicators: knowledge mastery, cognitive attitude, practical participation, and value recognition. The knowledge mastery level includes three secondary indicators: the breadth of basic cultural heritage knowledge, the depth of understanding of core concepts, and the ability to make cross-cultural comparisons. The cognitive attitude includes three secondary indicators: cultural openness, historical reverence, and critical reflection tendency. The level of practical participation includes three secondary indicators: frequency of extracurricular cultural activities, intensity of social practice investment, and activity level of interaction on digital platforms. The value recognition includes three secondary indicators: sense of national identity, sense of responsibility for cultural inheritance, and willingness to innovate and transform.

5. The big data-based evaluation system for college students' cultural heritage literacy according to claim 4, characterized in that, The competency modeling layer deploys a two-stage regression model to dynamically calibrate the weights of each secondary indicator: The first stage uses Gradient Boosting Decision Tree (GBDT) to learn the nonlinear mapping relationship between secondary indicators and expert-rated competence levels from historical samples; The second stage introduces the SHAP value decomposition mechanism to calculate the marginal contribution weight of each secondary indicator on a specific student sample. , ; Overall competence score Through formula Calculation, where is the standardized score of the j-th secondary indicator after Min-Max normalization to the [0,1] interval.

6. The big data-based evaluation system for college students' cultural heritage literacy according to claim 1, characterized in that, The image generation layer includes an ontology construction unit, a graph embedding unit, and an image reasoning unit; The ontology construction unit comprises a cultural heritage ontology with 12 top-level categories, 87 mid-level subclasses, and 532 instances. It is described and defined using the OWL 2 DL language, with 17 object attributes and 9 data attributes. The graph embedding unit uses the RotatE model to map ontology entities and relations to a low-dimensional vector space and is trained using the negative sampling cross-entropy loss function. The profile reasoning unit uses a dual-tower neural network to analyze student behavior feature vectors. Collection of cultural heritage entities with recent interactions Cross-modal alignment is performed using the average embedding vector, outputting a 256-dimensional image vector. Based on this vector, a literacy radar chart, knowledge blind spot prompts, and development trajectory curves are generated.

7. The big data-based evaluation system for college students' cultural heritage literacy as described in claim 6, characterized in that, The knowledge blind spot suggestion is obtained by calculating the profile vector. The cosine similarity with the center vectors of each cultural heritage subdomain is used to determine a knowledge blind spot when the similarity is below a threshold of 0.3; the development trajectory curve is based on the overall literacy score of consecutive semesters. Fitting a quadratic polynomial to predict future trends in literacy.

8. The big data-based evaluation system for college students' cultural heritage literacy as described in claim 1, characterized in that, The feedback intervention layer includes a rule engine unit, a natural language generation unit, and a multi-terminal distribution unit; The rule engine unit is pre-loaded with a decision rule base based on educational intervention theory. It dynamically triggers intervention packages based on the competency profile. Each intervention package includes a list of recommended resources, a list of action tasks, and expected performance indicators. The natural language generation unit uses a fine-tuned T5-base model to convert structured intervention instructions into natural language text that conforms to the educational context. The multi-terminal distribution unit pushes messages to students' mobile apps via a WebSocket long connection, simultaneously synchronizes summaries to the counselor's workbench via the WeChat API, and embeds the task into the online learning platform's assignment module via the LTI protocol.

9. The big data-based evaluation system for college students' cultural heritage literacy as described in claim 8, characterized in that, The decision rules of the rule engine unit include: If the value identification score is below 0.4 for two consecutive semesters and the participation in practice decreases by more than 20%, the 'cultural immersion reinforcement' intervention package will be triggered. If the critical reflection tendency score is higher than 0.7 but the knowledge mastery is lower than 0.5, the 'deep reading guidance' intervention package will be triggered.

10. The big data-based evaluation system for college students' cultural heritage literacy according to claim 1, characterized in that, The system employs a microservice architecture to achieve high reliability and scalability. Each module of the data acquisition layer is encapsulated as a Docker container and scheduled via a Kubernetes cluster. The data governance layer and literacy modeling layer are deployed on the Apache Spark on YARN framework; The graph embedding inference service of the image generation layer is exposed via gRPC interface after being optimized by TensorRT. The rule engine of the feedback intervention layer is implemented based on Drools 8.0, and the natural language generation model is hosted on Triton Inference Server; All inter-service communication uses Protocol Buffers format and is decoupled through Apache Kafka message queues; The system strictly adheres to privacy protection standards. Raw data is de-identified at the collection end, and only irreversible hash UIDs are used for association internally. Transmission is encrypted using TLS 1.3, storage is encrypted using AES-256-GCM, and a three-level access control policy is implemented.

Images