Intangible cultural heritage information promotion method and system

Abstract

The invention provides an intangible cultural heritage information promotion method and system, and the method comprises the steps: S2, extracting implicit features such as strength rhythm from structured coded data, generating a reproducible virtual simulation module, and obtaining a digital representation form of experience knowledge; s4, aiming at the output guidance sequence, integrating a real-time feedback mechanism, and performing comparative analysis by obtaining practical video data uploaded by a user to obtain a deviation correction suggestion so as to optimize a knowledge reproduction process; s5, updating a content library of the dynamic platform on the basis of the deviation correction suggestion, and if the correction suggestion exceeds a preset correction threshold value, triggering automatic iteration to obtain an enhanced cross-regional interaction model; and S7, generating a virtual interaction scene according to the adaptive real-time guidance resource, simulating the response of a handicraft, and obtaining final skill reproduction output to support wide propagation. According to the method, experience knowledge digitization and real-time feedback are combined, so that the inheritance efficiency and cross-region learning experience of traditional handicrafts are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of intangible cultural heritage processing technology, and in particular to a method and system for promoting intangible cultural heritage information. Background Technology

[0002] Intangible cultural heritage, as an important component of human cultural diversity, carries the historical memory and cultural wisdom of nations, making its protection and transmission particularly crucial in the context of globalization. Intangible cultural heritage skills are not merely technical knowledge systems, but also embody the experience and cultural value passed down through generations of artisans. However, with the acceleration of modernization, many intangible cultural heritage skills face the dilemma of interrupted transmission and insufficient promotion. How to effectively transmit and widely disseminate intangible cultural heritage skills through modern technology has become a crucial issue urgently needing to be addressed in the field of cultural preservation. The core of this field lies in using innovative methods to transmit the essence of traditional skills to a wider audience in a replicable and scalable form, while preserving their cultural connotations and unique techniques.

[0003] Currently, the transmission of intangible cultural heritage skills mainly relies on traditional apprenticeship systems or local training programs. While this approach ensures the authenticity of the skills, it is limited by geographical and temporal constraints, making large-scale knowledge dissemination difficult. Some digital attempts, such as video recording or textual compilation, attempt to overcome these limitations, but often remain at the superficial level of documentation, lacking a systematic expression of the craftsmen's techniques and experiential knowledge. This results in the core skills and techniques being difficult to accurately capture and transmit during the transmission process, making it difficult for learners to grasp the essence of the skills through existing methods. More importantly, these methods typically cannot support real-time interaction and dynamic updates across regions, failing to meet the diverse needs of modern society for the popularization and inheritance of intangible cultural heritage skills.

[0004] In the transmission of intangible cultural heritage skills, the core technical challenge lies in how to structurally encode the techniques and experiential knowledge of artisans. This encoding not only requires transforming complex movements, strengths, rhythms, and other details into storable digital forms, but also ensuring that learners can understand and reproduce this information intuitively. For example, in traditional embroidery, the angle of the stitches, the pressure applied, and the combination of different materials all rely on the artisan's long-term practical experience. This experience is often implicit and difficult to fully express through language or simple videos. The lack of structured encoding leads to inefficient transmission of skill knowledge, making it difficult for learners to grasp the core of the skills through existing digital content.

[0005] Another technical challenge stems from how to effectively transfer encoded knowledge across regions and time. Even if skill knowledge is successfully encoded, without a dynamic and interactive platform, learners still struggle to obtain real-time guidance or feedback in different scenarios. For example, a learner attempting to learn an intangible cultural heritage skill in a remote area, lacking direct interaction with the inheritor, cannot solve specific practical problems relying solely on static digital content. This lack of a transmission platform limits the promotion of skill knowledge to a small scale, hindering the widespread dissemination of intangible cultural heritage.

[0006] Therefore, how to structurally encode the techniques and experience of artisans and transmit these skills across regions and time through a dynamic and interactive digital platform has become a key issue in the promotion of intangible cultural heritage. This issue involves not only the technical aspects of encoding and storage, but also how to design a system that supports real-time interaction and dynamic updates to ensure that skills and knowledge can be effectively reproduced and passed on in different cultural and regional contexts.

[0007] By addressing the aforementioned issues, the inheritance and promotion of intangible cultural heritage skills can be liberated from traditional limitations, enabling the widespread dissemination of cultural value. How to preserve the authenticity of the skills while achieving efficient and dynamic knowledge transfer through digital means becomes the core business problem this research aims to solve. Summary of the Invention

[0008] To address the technical problems raised in the background section, a first aspect of the present invention provides a method for promoting information on intangible cultural heritage, the method comprising: S1. By collecting videos of artisans' operations and sensor data, the action sequences are grouped and processed to obtain structured coded data. S2. Implicit features such as force and rhythm are extracted from the structured coded data to generate reproducible virtual simulation modules, obtaining a digital representation of experiential knowledge. S3. An interactive database is constructed based on the digital representation. If a user queries specific skill details, matching items are retrieved from the database, and the first similarity threshold of the search results is determined to determine the output guidance sequence. S4. A real-time feedback mechanism is integrated into the output guidance sequence. By acquiring and analyzing practice video data uploaded by users, deviation correction suggestions are obtained to optimize the knowledge reproduction process. S5. The content library of the dynamic platform is updated based on the deviation correction suggestions. If the correction suggestions exceed a preset correction threshold, automatic iteration is triggered to obtain an enhanced cross-regional interactive model. S6. Personalized learning paths are distributed from the enhanced cross-regional interactive model. User regional data is classified, and the classification results are used to determine suitable real-time guidance resources. S7. Virtual interactive scenes are generated based on the suitable real-time guidance resources, simulating the artisan's response, to obtain the final skill reproduction output to support widespread dissemination.

[0009] As a further description of the above technical solution: Step S1 involves collecting video footage of the artisan's actions and sensor data, grouping the action sequences to obtain structured coded data, including: Step S11: Record the artisan's movements with a camera and collect movement parameters with an accelerometer and gyroscope to obtain the original movement sequence; Step S12: Extract the first feature vector from the original action sequence; Step S13: Group the first feature vector into action groups; Step S14: Generate the first structured data based on the action grouping results; Step S15: Generate encoded data from the first structured data; Step S16: Extract action patterns from the encoded data to obtain an action pattern set; Step S17: Generate compressed data based on the action pattern set.

[0010] As a further description of the above technical solution: Step S11, recording the movements of the craftsman through a camera and collecting movement parameters through an accelerometer and a gyroscope to obtain the original movement sequence, includes: the original movement sequence is composed of a video frame sequence and a sensor time sequence.

[0011] As a further description of the above technical solution: Step S12, extracting the first feature vector from the original action sequence, includes: for the video frame sequence, calculating the optical flow vector and aggregating it into trajectory features; for the sensor time series, calculating the average acceleration and the standard deviation of angular velocity within a fixed time window to form the first feature vector.

[0012] As a further description of the above technical solution: Step S13, grouping the first feature vector by action, includes: grouping the first feature vector by using the K-means clustering algorithm, iteratively allocating cluster centers based on Euclidean distance, and if the distance from the first feature vector to the center is less than a preset distance threshold, they are grouped into the same group to obtain the action grouping result.

[0013] As a further description of the above technical solution: Step S14, generating first structured data based on the action grouping results, includes: for each action sequence, extracting the cluster center as a representative vector, and combining the timestamp and action label to generate the first structured data.

[0014] As a further description of the above technical solution: Step S15, generating encoded data from the first structured data, includes: for the first structured data, using a sequential encoding method, mapping action labels and cluster centers to fixed-length codes to obtain encoded data.

[0015] As a further description of the above technical solution: Step S7, generating a virtual interactive scene based on the adapted real-time guidance resources, simulating the artisan's response, and obtaining the final skill reproduction output to support widespread dissemination, includes: Step S71: Obtain behavioral data from the activities of artisans through a preset data acquisition module, extract features from the behavioral data, and obtain a set of behavioral features; Step S72: If the completeness of the behavioral feature set is greater than the preset completeness threshold, then perform time series analysis on the behavioral feature set to generate the artisan response sequence. Step S73: Based on the artisan's response sequence, construct a virtual interactive scene and obtain virtual scene data; Step S74: By mapping the virtual scene data with the artisan's response sequence, dynamic interactive content is generated to obtain the interactive simulation output; Step S75: If the average frame rate of the interactive simulation output is greater than the preset frame rate threshold, then the interactive simulation output is optimized to obtain the skill reproduction content. Step S76: Based on the content of the technique reproduction, disseminate information through multiple channels to achieve widespread dissemination and output.

[0016] As a further description of the above technical solution: In step S72, if the completeness of the behavioral feature set is greater than a preset completeness threshold, then a time series analysis is performed on the behavioral feature set to generate a craftsman's response sequence, including: The completeness of the behavioral feature set is calculated using the following formula: Where J represents the completeness of the behavioral feature set, F represents the actual number of behavioral features extracted, and T represents the expected number of behavioral features extracted.

[0017] A second aspect of the present invention provides an intangible cultural heritage information promotion system, which promotes intangible cultural heritage information using the method described above. The system includes: a data acquisition and structuring module, used to collect videos of artisans' operations and sensor data, group and process action sequences to obtain structured coded data; a latent feature extraction and simulation module, used to extract latent features such as intensity and rhythm from the structured coded data, generate a reproducible virtual simulation module, and obtain a digital representation of experiential knowledge; an interactive query and matching module, used to construct an interactive database based on the digital representation, and if a user queries specific skill details, retrieve matching items from the database, determine a first similarity threshold of the search results to determine the output guidance sequence; and real-time feedback and... The invention comprises the following modules: a correction module, which integrates a real-time feedback mechanism for the output guidance sequence; a dynamic platform iteration module, which updates the content library of the dynamic platform based on the correction suggestions; and a personalized learning distribution module, which distributes personalized learning paths from the enhanced cross-regional interaction model, classifies user regional data, and determines the appropriate real-time guidance resources based on the classification results. A virtual interaction generation module generates virtual interactive scenarios based on the appropriate real-time guidance resources, simulates the response of artisans, and obtains the final skill reproduction output to support widespread dissemination. The technical solution provided by this invention has the following beneficial effects: This invention discloses a method and system for promoting intangible cultural heritage information. Addressing the challenges of structured expression of experiential knowledge, low efficiency of cross-regional dissemination, and insufficient learning feedback in the transmission of traditional handicrafts, this invention collects videos of artisans' operations and sensor data. A clustering algorithm is used to group action sequences, generating structured coded data and extracting implicit features such as force and rhythm. A deep learning model is used to train a virtual simulation module, forming a reproducible digital representation. Based on this, an interactive database is constructed, allowing users to query skill details and receive guidance sequences. A real-time feedback mechanism is integrated, generating deviation correction suggestions by comparing user practice videos, dynamically updating the content library, triggering automatic iteration, optimizing the cross-regional interactive model, and ultimately generating personalized learning paths and virtual interactive scenarios, achieving precise guidance and skill dissemination. This invention significantly improves the efficiency of traditional handicraft transmission and enhances the cross-regional learning experience by combining digitized experiential knowledge with real-time feedback. Attached Figure Description

[0018] Figure 1 This is a flowchart of a method for promoting intangible cultural heritage information according to the present invention.

[0019] Figure 2This is a schematic diagram of a method for promoting intangible cultural heritage information according to the present invention.

[0020] Figure 3 This is another schematic diagram of a method for promoting intangible cultural heritage information according to the present invention.

[0021] Figure 4 This is a schematic diagram of the structure of an intangible cultural heritage information promotion system according to the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0023] like Figures 1-3 As shown, in a first aspect, the present invention provides a method for promoting information on intangible cultural heritage, specifically including: S1 involves collecting videos of artisans performing their tasks and sensor data, then using a clustering algorithm to group the action sequences to obtain structured coded data. The clustering algorithm categorizes similar techniques based on feature vector distance calculations.

[0024] Optionally, this step also includes: Step S11: Record the artisan's movements with a camera and collect movement parameters with an accelerometer and gyroscope to obtain the original movement sequence.

[0025] Preferably, the original action sequence consists of a video frame sequence and a sensor time sequence.

[0026] Step S12: Extract the first feature vector from the original action sequence.

[0027] Optionally, for video frame sequences, optical flow vectors are calculated using the OpenCV library and aggregated into trajectory features; for sensor time series, the average acceleration and standard deviation of angular velocity are calculated within a fixed time window to form the first feature vector.

[0028] Step S13: The first feature vector is grouped using the K-means clustering algorithm.

[0029] Preferably, cluster centers are assigned iteratively based on Euclidean distance. If the distance from the first feature vector to the center is less than a preset distance threshold, they are grouped into the same group to obtain the action grouping result.

[0030] Step S14: Generate the first structured data based on the action grouping results.

[0031] Preferably, for each action sequence, the cluster center is extracted as a representative vector, and combined with the timestamp and action label to generate the first structured data.

[0032] Step S15: Generate encoded data from the first structured data.

[0033] Preferably, for the first structured data, a sequential encoding method is used to map the action labels and cluster centers into fixed-length codes to obtain encoded data.

[0034] Step S16: Extract action patterns from the encoded data to obtain an action pattern set.

[0035] Optionally, the PrefixSpan algorithm can be used to analyze repetitive sequences in the encoded data. If the frequency of a sequence is higher than a preset frequency threshold, it is determined to be a typical action pattern.

[0036] Step S17: Generate compressed data based on the action pattern set.

[0037] Optionally, for a set of action patterns, Huffman coding can be used to compress repetitive sequences to generate compressed data.

[0038] For example, in a scene where a potter is making ceramics, motion data collected by cameras and sensors allows for precise analysis of the process. The original motion sequence consists of video frames and sensor time series. The camera records the potter's movements while throwing the potter's wheel at 30 frames per second, generating a continuous video frame sequence; simultaneously, accelerometers and gyroscopes collect hand acceleration and angular velocity data at a frequency of 100Hz, forming a time series. This multimodal data acquisition ensures the comprehensiveness of motion information, providing a foundation for subsequent feature extraction.

[0039] Specifically, when extracting the first feature vector, optical flow vectors are calculated using the OpenCV library for the video frame sequence. Optical flow reflects the movement trajectory of the potter's hand during pottery throwing, such as sliding along the direction of the potter's wheel rotation. By aggregating the optical flow vectors, trajectory features are obtained, describing the hand's movement pattern. For the sensor time series, the average acceleration is calculated within a 0.5-second time window; for example, the average acceleration of the hand moving forward and backward is approximately 2.5 m / s². Simultaneously, the standard deviation of angular velocity is calculated; for example, the standard deviation of rotation around the Z-axis is 0.8 rad / s. These statistics constitute the first feature vector, capturing the dynamic characteristics of the action and aiding in subsequent cluster analysis.

[0040] In one embodiment, K-means clustering is used to group the first feature vector. K is set to 5, representing five typical actions, such as pinching, rotating, and flattening. Based on Euclidean distance, cluster centers are iteratively assigned; if the distance from a vector to a center is less than a distance threshold of 0.3, it is grouped into one group. The grouping results divide the ceramic artist's continuous actions into discrete action units, facilitating structured processing.

[0041] For example, the first structured data is generated from action groups. Each action group is represented by a vector with the cluster center, combined with a timestamp and action label, such as "10:00:05 pinch-pull action vector [2.5, 0.8]". This structured data clearly records the time and features of the action sequence, which is convenient for pattern mining.

[0042] Specifically, a sequential encoding method is used when generating coded data. Action labels are mapped to fixed-length codes, such as "pinch and pull" as 01 and "rotate" as 02. These codes are then combined with cluster center vectors to form coded data. This encoding preserves the temporal and feature information of the action sequence, providing a foundation for pattern analysis.

[0043] In one embodiment, the PrefixSpan algorithm is used to extract motion patterns. A frequency threshold of 0.2 is set; if the "pinch-pull-rotate-flatten" sequence appears in more than 20% of the encoded data, it is identified as a typical motion pattern. This typical motion pattern reflects the core process of a potter's wheel-throwing, which helps in process optimization.

[0044] For example, when generating compressed data, Huffman coding is used to compress the set of action patterns. Frequently occurring "pinch-pull-rotate" sequences are assigned shorter codes, such as 10, to reduce data storage space. This compression method reduces storage costs while retaining key information about the action patterns, facilitating transmission and analysis.

[0045] Understandably, the above method, through multimodal data acquisition, feature extraction, clustering, encoding, and compression, forms an efficient motion analysis workflow. The final compressed data can be used for process transmission, motion optimization, or automated teaching, demonstrating significant technical value.

[0046] S2 extracts implicit features such as intensity and rhythm from structured coded data, and uses a deep learning model to train and generate reproducible virtual simulation modules, thereby obtaining a digital representation of experiential knowledge.

[0047] Optionally, this step also includes: Step S21: Obtain the encoding format from the structured data, use preprocessing methods to extract the original features, and obtain the initial representation of strength and rhythm.

[0048] In step S22, if the initial representation contains noise, a convolutional neural network is used to denoise the data to obtain clear strength and rhythm features.

[0049] Step S23: Based on the clear strength and rhythm characteristics, a long short-term memory network is used for temporal modeling to obtain a dynamic representation of the latent features.

[0050] Step S24: Train the deep learning model through dynamic representation to generate a virtual simulation module and obtain a reproducible simulation unit.

[0051] Step S25: Extract experiential knowledge from the simulation unit, and perform knowledge transformation using an encoder-decoder structure to obtain a structured knowledge representation.

[0052] Step S26: For the structured knowledge representation, a data mapping method is used to generate a digital representation, and the final output result is obtained.

[0053] Step S27: If the output result does not match the preset digitization threshold, the parameters of the deep learning model are adjusted to obtain the optimized digitization representation.

[0054] In one possible implementation, when obtaining the encoding format from structured data, the original features can be extracted through preprocessing methods.

[0055] For example, a craftsman's motion data includes video frames and sensor time series. Preprocessing can extract color histograms from the video frame sequence and calculate peak acceleration and rate of change of angular velocity from the sensor data to form an initial representation of force and rhythm. Assuming the color histogram extracted from the video frames reflects the brightness changes in the action area, and the sensor data records hand force in the range of 0 to 10 Newtons and angular velocity change in the range of 0 to 5 radians per second, a preliminary feature vector is obtained. This method can capture the dynamic characteristics of the motion, providing a foundation for subsequent processing.

[0056] For example, if the initial representation contains noise, such as lighting variations in video frames or jitter interference from sensors, a convolutional neural network can be used for noise reduction.

[0057] Specifically, convolutional neural networks extract spatial features from video frames and filter out lighting-independent noise through multiple convolutional operations. For sensor data, one-dimensional convolution is used to smooth the time series, preserving the core variations in intensity and rhythm. Assuming the input intensity data contains random noise, the intensity values ​​stabilize within the range of 2 to 8 Newtons after noise reduction, and the time interval error of rhythm features is reduced to within 0.1 seconds. This processing significantly improves data quality and provides reliable input for time series modeling.

[0058] In one embodiment, a long short-term memory network can be used for temporal modeling based on clear strength and rhythm characteristics.

[0059] For example, when studying a craftsman's weaving movements, the network input consists of force values ​​per second and rhythm intervals, while the output is a dynamic representation of latent features. Assuming the force sequence of the weaving movement is 3, 5, and 4 Newtons, and the rhythm intervals are 0.5 seconds and 0.7 seconds, the Long Short-Term Memory (LSTM) network captures the periodic changes in the movement through memory units, generating a dynamic representation reflecting the smoothness of the movement. This representation reflects the continuity and regularity of the movement, providing support for subsequent simulations.

[0060] For example, training a deep learning model using dynamic representations can generate virtual simulation modules. Assuming the model takes a dynamic representation of a weaving motion as input, training it produces a reproducible simulation unit capable of simulating the movement trajectory of a hand in three-dimensional space. The simulation unit can output coordinates per second, such as (0.1, 0.2, 0.3) meters, reflecting the precise location of the movement. This module can provide a reference for automated equipment, enhancing the accuracy of motion reproduction.

[0061] In one possible implementation, empirical knowledge is extracted from the simulation unit, and knowledge transformation can be performed using an encoder-decoder structure.

[0062] For example, the encoder compresses the trajectory data of the analog unit into a low-dimensional vector, and the decoder converts it into a structured knowledge representation, such as the intensity range and rhythm frequency of the movement. Assuming the output knowledge representation is "intensity 4-6 Newtons, rhythm 0.6 seconds / cycle," this representation is easy to store and analyze, providing convenience for subsequent applications.

[0063] For example, for structured knowledge representations, digital representations can be generated through data mapping. Assuming the knowledge representation is mapped to binary encoding, a force of 4-6 Newtons is mapped to 0 and 1, and a tempo of 0.6 seconds is mapped to 10, the final output is generated. This digital representation facilitates system integration and transmission. If the output does not match the preset digital threshold, such as a tempo frequency deviation exceeding 0.1 seconds, the learning rate or number of layers in the deep learning model can be adjusted, and the digital representation can be regenerated after optimization to ensure the result meets expectations. This method can improve the system's adaptability and accuracy.

[0064] S3: Construct an interactive database based on the digital representation. If a user queries specific technique details, retrieve matching items from the database and determine the first similarity threshold of the search results to determine the output guidance sequence.

[0065] Optionally, this step also includes: Step S31: Extract data fields from the structured representation, implement key-value pair mapping through Python dictionary, generate the storage structure of the interactive database, and obtain the initial database.

[0066] Step S32: Based on the initial database, import data using Elasticsearch and create an inverted index to generate a query-optimized index table, resulting in a database that can be retrieved quickly.

[0067] Step S33: If the user submits a query request, relevant data fields are retrieved from the database that can be quickly searched based on the index table. The TF-IDF vector cosine similarity between the query and the data fields is calculated using the Scikit-learn library to obtain a similarity score.

[0068] Step S34: For the similarity score, a preset first similarity threshold is used for comparison. If the similarity score exceeds the first similarity threshold, it is determined as a valid match and a matching result set is obtained.

[0069] Step S35: Extract technique details from the matching result set, serialize and encode them using JSON format, generate a structured guide sequence, and obtain the preliminary output sequence.

[0070] Step S36: Using the initial output sequence, the sorted sequence is arranged in descending order of similarity score using Python's sorted function to generate the final guidance sequence and obtain the output result of the user's query.

[0071] For example, when extracting data fields from a structured representation, a sequence of notes from a musical composition can be used as a data source. The note sequence contains information such as pitch, duration, and dynamics. A Python dictionary is used to map pitch as the key and duration and dynamics as the value, forming key-value pairs.

[0072] For example, the C4 note corresponds to a duration of 0.5 seconds and a dynamic range of 60, generating a key-value pair {C4:{duration:0.5, dynamic range:60}}. This mapping method facilitates subsequent storage and retrieval, maintaining the clarity of the data structure.

[0073] In one possible implementation, the initial database could be built on a relational database, such as SQLite, storing key-value pairs as a table structure. The table would contain fields such as note value, duration, and dynamics, with each row recording complete information for a single note.

[0074] For example, the C4 note is recorded as a single row, with field values ​​of C4, 0.5, and 60. This structured storage facilitates data management and provides a foundation for subsequent indexing.

[0075] Specifically, when importing data into Elasticsearch and creating an inverted index, you can import the note sequence field into the Elasticsearch cluster. The inverted index uses pitch as its core field, recording the document position where each pitch appears.

[0076] For example, the C4 pitch appears in positions 3 and 5 of document 1, and the D4 pitch appears in position 1 of document 2. After the index table is generated, searching for the C4 pitch can quickly locate the relevant document, improving retrieval efficiency.

[0077] For example, when retrieving data fields based on an index table, if a user queries "C4 notes with an intensity greater than 50", the system locates the C4 pitch documents using the inverted index, then filters for records with an intensity greater than 50, and returns the matching results. This approach reduces the time spent on a full table scan and is suitable for real-time query scenarios.

[0078] In one possible implementation, the Scikit-learn library can be used to analyze the semantic relevance between the query and the data fields when calculating the cosine similarity of TF-IDF vectors.

[0079] For example, when a user queries "fast tempo notes," the system vectorizes the query text and the duration descriptions of notes in the database, then calculates the cosine similarity. Assuming the cosine value of the description vector for a C4 note with a duration of 0.5 seconds is 0.85, exceeding the preset first similarity threshold of 0.8, it is considered a valid match. This method ensures that the query results are highly relevant to the user's intent.

[0080] Specifically, when extracting technique details from the matching result set, the combination patterns of intensity and duration can be extracted from the matched note records.

[0081] For example, a C4 note with a dynamic range of 60 and a duration of 0.5 seconds reflects a certain playing style. This can be serialized into a structured instruction sequence using JSON, such as {note:C4, dynamic range:60, duration:0.5}. This serialization method facilitates data transfer between systems and subsequent processing.

[0082] For example, the final guidance sequence is generated using Python's `sorted` function. Assuming the matching results contain the notes C4, D4, and E4 with similarity scores of 0.85, 0.75, and 0.90 respectively, sorting them generates the sequence [E4, C4, D4]. The output received by the user is arranged from highest to lowest relevance, intuitively reflecting the priority of the query and facilitating quick location of core information.

[0083] In one possible implementation, the above process can be applied to a music creation assistance system to generate note sequence recommendations that match user queries. Whether it's data extraction, storage, indexing, or query matching, each step is tightly integrated, forming a highly efficient music information processing chain that significantly improves user experience and data processing efficiency.

[0084] S4 integrates a real-time feedback mechanism for the output guidance sequence. It obtains deviation correction suggestions by comparing and analyzing the practical video data uploaded by users to optimize the knowledge reproduction process.

[0085] Optionally, this step also includes: Step S41: Extract action sequence features from the practice video data uploaded by the user, and use the OpenPose tool to obtain the first action sequence feature point set.

[0086] Step S42: Compare the feature point set of the first action sequence with the preset standard model, calculate the deviation value L between the feature point set of the first action sequence and the standard model, and obtain the first deviation analysis result.

[0087] Optionally, the deviation value L between the feature point set of the first action sequence and the standard model can be calculated using the following formula: in,( , ) represents the coordinates of the feature point set of the first action sequence. , () is the standard coordinate system.

[0088] Step S43: If the first deviation analysis result exceeds the preset first deviation threshold, then a first action correction suggestion is generated using a linear regression algorithm, and the content of the first correction suggestion is determined.

[0089] Step S44: Based on the first correction suggestion, generate the first optimized action sequence data to obtain the first adjusted action feature point set.

[0090] Step S45: Compare the first adjusted action feature point set with the preset standard model again, calculate the consistency score of reproduction, and judge the reproduction effect.

[0091] Preferably, the reproduction consistency score G is calculated using the following mean squared error formula: in, This is the first adjusted action feature point. Here, n is the standard point, and n is the number of points. Step S46: If the consistency score is lower than the preset consistency score threshold, the first action correction suggestion is iteratively updated to generate the second optimized action sequence data.

[0092] For example, when extracting motion sequence features from user-uploaded practice video data, consider a user-uploaded dance practice video. The goal is to analyze whether the dancer's movements conform to standard steps. Using the OpenPose tool, the system first identifies key points of the dancer in the video, such as shoulders, elbows, and knees, forming a first set of motion sequence feature points. These feature points are recorded in coordinate form, for example, shoulders at (100, 200) and knees at (150, 300). By capturing the coordinate changes in each frame, the system generates time-series data of the motion for subsequent comparison.

[0093] In one possible implementation, when comparing with a preset standard model, it is assumed that the standard model is a professional dancer's step template, containing the standard coordinates of the same key points.

[0094] For example, the standard shoulder coordinates are (105, 195). The average deviation value L is obtained by calculating the Euclidean distance between the feature points of the first action sequence and the standard point.

[0095] For example, the shoulder deviation is 5 pixels, the knee deviation is 10 pixels, and the average deviation L is 7.5. If the preset first deviation threshold is 8, then this deviation is within an acceptable range, indicating that the movement is close to the standard.

[0096] Specifically, if the deviation value L exceeds the first deviation threshold, for example, L = 12, the system will trigger a linear regression algorithm to generate correction suggestions. The algorithm analyzes key points with large deviations and infers the trend of user action deviation.

[0097] For example, if the system detects that a user is not raising their knee high enough, it suggests "raising the left knee to hip level." These suggestions are based on models trained with historical data and provide specific guidance for deviation patterns.

[0098] For example, when generating the first optimized motion sequence data, the system adjusts the feature point coordinates based on the correction suggestions. Suppose that after the user's adjustment, the knee coordinates change from (150, 300) to (150, 280), which is closer to the standard coordinates (150, 275). The first adjusted set of motion feature points reflects the user's improved motion state.

[0099] In one possible implementation, the consistency / inconsistency score is calculated using the mean squared error formula.

[0100] For example, the error between the adjusted knee coordinates (150, 280) and the standard coordinates (150, 275) is 5 pixels. The average error of all keypoints is calculated to obtain the score G. If G is lower than a preset consistency score threshold, such as 0.01, it indicates that the movement is highly consistent and no further correction is needed. If G is higher, such as 0.05, a second round of correction suggestions needs to be generated iteratively, such as "slow down the pace and keep the knee high."

[0101] It's worth noting that when iterating and updating suggestions, the system may consider the context of the action. For example, if the user's pace is too fast, causing deviation, the suggestion might include "control the pace to 1 step per second." This ensures the suggestions are more targeted and improves the accuracy of the user's actions.

[0102] For example, the advantage of this method lies in its ability to provide users with rapid and precise feedback on movement improvements through data-driven analysis and iterative optimization. This approach is particularly suitable for scenarios requiring high-precision movements, such as dance instruction or sports training, significantly improving learning efficiency.

[0103] S5 updates the content library of the dynamic platform based on the deviation correction suggestions. If the correction suggestions exceed the preset correction threshold, automatic iteration is triggered to obtain an enhanced cross-regional interaction model.

[0104] Optionally, this step also includes: Step S51: Obtain update instructions from the deviation correction suggestions, and use the spaCy tool to parse the suggestion text, extract keywords, and determine the scope of the content library update.

[0105] Step S52: Extract user behavior log data from the MySQL database according to the update range, and generate a content recommendation set using a collaborative filtering algorithm.

[0106] Step S53: If the cosine similarity between the content recommendation set and the regional feature extraction result is lower than the preset second similarity threshold, the content distribution strategy is adjusted by the k-means clustering algorithm.

[0107] Step S54: Based on the adjusted content distribution strategy, obtain cross-regional interaction data and use Pandas tools to process the data to generate an initial augmented model.

[0108] Step S55: Extract interaction effect evaluation indicators from the initial enhancement model to determine whether the content distribution efficiency meets expectations.

[0109] Step S56: If the content distribution efficiency does not meet expectations, then use the Apache Spark tool to regenerate deviation correction suggestions based on the interaction effect evaluation metrics.

[0110] For example, when obtaining update instructions from deviation correction suggestions, the spaCy tool can be used to parse the text and extract keywords to determine the scope of content library updates. As a natural language processing tool, spaCy can efficiently identify key information such as actions and objects in text. Suppose that after analyzing a user-uploaded yoga practice video, the deviation correction suggestion is "arms raised too low, need to be adjusted to shoulder level." After parsing, spaCy extracts the keywords "arms," ​​"raise," "angle," and "shoulder level," thereby determining the teaching modules related to arm movements in the content library to be updated. Keyword extraction not only accurately locates the update scope but also reduces redundant processing of irrelevant content, improving the targeting of content updates.

[0111] In one possible implementation, when retrieving user behavior log data from a MySQL database, behavioral data related to arm movements can be filtered based on the user's historical practice records, such as practice duration and frequency of movement completion.

[0112] For example, database records show that user A's arm movement completion rate was only 60% in the past 10 practice sessions, indicating a persistent deviation in this movement. Based on this, collaborative filtering algorithms can analyze behavioral data from similar user groups and recommend instructional videos or movement guidance content suitable for user A, such as "arm extension stretch tutorial." This recommendation method enhances the personalized adaptability of content through group behavioral patterns.

[0113] Specifically, if the cosine similarity between the content recommendation set and the regional features is lower than a preset second similarity threshold, such as 0.8, then the content distribution strategy needs to be adjusted using the k-means clustering algorithm. Assuming users come from different regions, with users in the east preferring dynamic yoga and users in the west favoring static stretching, the k-means algorithm can cluster users into two groups based on region and preference, and then reallocate the content distribution weights.

[0114] For example, dynamic yoga videos are prioritized for users in the eastern region, while static stretching courses are shown to users in the western region. This adjustment ensures that content distribution better meets localized needs and increases user engagement.

[0115] For example, when acquiring cross-regional interaction data, Pandas tools can be used to process user practice feedback data across different regions, such as completion rates and feedback comments. Suppose that after processing, it is found that the average completion rate for dynamic yoga among users in the eastern region is 85%, while the completion rate for static stretching among users in the western region is 70%. Based on this, the initial augmentation model can prioritize enhancing dynamic content recommendations for users in the eastern region, while simultaneously providing supplementary teaching resources for users in the western region. Pandas' efficient data processing capabilities support the rapid generation of models, optimizing the accuracy of content distribution.

[0116] In one possible implementation, when extracting interaction effect evaluation metrics from the initial enhancement model, attention can be paid to the duration of user practice and the accuracy of action completion.

[0117] For example, the model shows that after adjustment, user A's practice time increased from 20 minutes to 30 minutes, and the accuracy of the movements improved from 60% to 80%. If the distribution efficiency still does not meet expectations, such as the accuracy not reaching 85%, then Apache Spark is used to process large-scale interactive data and regenerate bias correction suggestions.

[0118] For example, it might be suggested to add "arm strength training" content for user A and adjust the video push frequency. This approach ensures continuous improvement in content distribution through data-driven iterative optimization.

[0119] Understandably, all the above steps are closely interconnected, from keyword extraction to content recommendation and then to distribution strategy adjustments, forming a complete content optimization loop. The implementation methods of each technical topic support each other, ensuring that deviations in user action practice are effectively corrected, while significantly improving the efficiency and personalization of content distribution.

[0120] S6 distributes personalized learning paths from the enhanced cross-regional interaction model, uses clustering algorithms to classify user regional data, and determines the appropriate real-time guidance resources based on the classification results.

[0121] Optionally, this step also includes: Step S61: Use the K-means clustering algorithm to classify the user geographic data, obtain geographic labels, and obtain the classification set of user groups.

[0122] Step S62: Based on the classification set, use Tableau tool to analyze user behavior data, generate personalized learning paths, and determine the mapping relationship between path content and regional tags.

[0123] Step S63: Using the Apriori algorithm, process the personalized learning path and regional labels to obtain the association rule results and obtain the resource allocation parameters.

[0124] Step S64: If the resource allocation parameters meet the preset allocation parameter threshold, then generate a real-time guidance resource allocation scheme based on the association rule results and determine the priority of resource distribution.

[0125] Step S65: Based on the priority of resource distribution, obtain matching resource content from the real-time guidance resource library to obtain an adapted resource set.

[0126] Step S66: Distribute the adapted resource set to the user terminal according to the pre-established allocation rules, obtain the distribution log, and determine the completion status of the distribution.

[0127] Step S67: If the distribution log shows a completion status, then update the personalized learning path based on the user behavior data to obtain an optimized learning path set, and use the optimized learning path set as input to re-analyze the user behavior data using Tableau tool.

[0128] For example, in scenarios where resource allocation is optimized based on geographic labels, the K-means clustering algorithm can be used to classify user geographic data. Suppose an online education platform needs to provide personalized learning resources to users in different regions. First, it collects data such as the user's city and learning preferences. Then, using the K-means algorithm, users are divided into several groups based on geographic characteristics, such as East China, South China, and Northwest China.

[0129] For example, if a user is from Shanghai, the algorithm will categorize them into the "Active Group in East China" and generate a corresponding regional tag. This classification method, by analyzing geographical location and user behavior data, ensures that subsequent resource allocation is highly correlated with regional characteristics.

[0130] Specifically, Tableau can be used to analyze user behavior data and generate personalized learning paths. For example, if the East China population prefers short video courses while the Northwest population favors longer documents, Tableau can generate learning paths tailored to different groups by visualizing data such as user click-through rates and learning duration.

[0131] For example, users in East China might receive a learning path containing 10-minute video lessons, while users in Northwest China might receive a path for in-depth reading documents. These paths are mapped to geographic tags to ensure content suitability.

[0132] In one embodiment, the Apriori algorithm is used to discover association rules between learning paths and geographic tags. Assuming analysis reveals that users in East China are more likely to participate in interactive quizzes after watching video courses, the Apriori algorithm generates the rule: Video Course → Interactive Quiz, with a support of 0.8 and a confidence level of 0.9. These rules help determine resource allocation parameters, such as prioritizing quiz resources for users in East China. If the parameters meet allocation thresholds (e.g., support greater than 0.75), a real-time guidance resource allocation scheme is generated, prioritizing quizzes over documents.

[0133] For example, a real-time resource allocation scheme can extract content from a resource library based on priority. Assuming the resource library contains three types of resources: videos, documents, and quizzes, users in East China will receive a combination of quizzes and videos first, forming a suitable resource set. This set is then pushed to users according to preset rules (such as priority distribution). Distribution logs show that 95% of users in East China received the resources within 24 hours and confirmed the distribution was complete.

[0134] Specifically, learning paths can be further optimized based on distribution logs and user behavior data. For example, if logs show high quiz completion rates but insufficient video viewing time among users in East China, the system can adjust the paths, increasing the proportion of short videos, thus creating an optimized set of learning paths. These sets are then re-entered into Tableau to analyze changes in user behavior, such as an increase in quiz participation to 85%, thereby continuously improving the accuracy of resource allocation. This data-driven optimization ensures that resource distribution is highly aligned with user needs, enhancing learning effectiveness.

[0135] In one embodiment, the optimized set of learning paths may further incorporate user feedback.

[0136] For example, if users in the Northwest region report that the document content is too complex, the system can adjust it to a more concise version, re-analyze behavioral data, and verify the optimization effect. This closed-loop mechanism ensures the adaptability and efficiency of resource allocation through continuous iteration.

[0137] S7 generates virtual interactive scenes based on adapted real-time guidance resources, simulates artisan responses through deep learning models, and obtains the final skill reproduction output to support widespread dissemination.

[0138] Optionally, this step also includes: Step S71: Obtain behavioral data from the activities of artisans through a preset data acquisition module, and use a ResNet model to extract features from the behavioral data to obtain a set of behavioral features.

[0139] Step S72: If the completeness of the behavioral feature set is greater than the preset completeness threshold, then the behavioral feature set is analyzed by time series analysis using the LSTM model to generate the artisan response sequence.

[0140] Preferably, the completeness J of the behavioral feature set is calculated using the following formula: Where F is the actual number of behavioral features extracted, and T is the expected number of behavioral features extracted.

[0141] Step S73: Based on the artisan's response sequence, use Unity tools to construct a virtual interactive scene and obtain virtual scene data.

[0142] Step S74: By mapping the virtual scene data with the artisan's response sequence, dynamic interactive content is generated, resulting in an interactive simulation output.

[0143] Step S75: If the average frame rate of the interactive simulation output is greater than the preset frame rate threshold, the interactive simulation output is optimized using the Blender tool to obtain the skill reproduction content.

[0144] Step S76: Based on the content of the skill reproduction, use the YouTube API to disseminate information through multiple channels and achieve widespread dissemination.

[0145] In one possible implementation, behavioral data is acquired from the activities of artisans through a pre-set data acquisition module, involving real-time monitoring of the artisans' work environment.

[0146] For example, sensor devices and cameras can be used to capture data such as the hand movements, tool usage frequency, and work rhythm of artisans when making ceramics.

[0147] Specifically, sensors record the force and time interval of each strike of the clay, while cameras capture the movement trajectory, forming a multidimensional dataset with timestamps. This method comprehensively records the behavioral characteristics of artisans, providing a reliable data foundation for subsequent analysis.

[0148] For example, when using the ResNet model to extract features from data, video frames captured by a camera can be input into the model to extract deep features of a craftsman's movements, such as the speed and angle changes of his hand. Assuming the expected extraction of 100 behavioral features, and 90 actually extracted, the completeness J of the behavioral feature set is 90 / 100 = 0.9, which is higher than the completeness threshold of 0.8, indicating that the behavioral feature extraction results meet the requirements. This high completeness ensures the accuracy of subsequent analysis and lays the foundation for generating an accurate behavioral feature set.

[0149] In one embodiment, a response sequence is generated by performing time-series analysis on a set of behavioral features of artisans using an LSTM model.

[0150] For example, by inputting the features of continuous kneading, throwing, and finishing actions in ceramic making into an LSTM, the model can capture the temporal dependencies between actions and generate sequence data that describes the rhythm of a craftsman's production.

[0151] Preferably, this sequence can reflect the habits of craftsmen at different stages of production, such as slowing down the trimming speed after rapid throwing, providing a dynamic basis for the construction of subsequent virtual scenes.

[0152] Specifically, when using Unity tools to build virtual interactive scenes, a virtual ceramic workshop can be generated based on the response sequence.

[0153] For example, a virtual scene may contain a dynamic image of a craftsman whose movements are synchronized with a sequence generated by an LSTM, such as simulating the rotation speed and changes in hand pressure during pottery throwing.

[0154] It should be noted that Unity uses real-time rendering technology to smoothly present the movements of artisans in a virtual environment, providing users with an immersive experience. The generation of virtual scene data enhances the ability to digitally reproduce craftsmanship.

[0155] In one possible implementation, dynamic interactive content is generated through the mapping relationship between virtual scene data and response sequences.

[0156] For example, based on the sequence data of a craftsman's wheel-throwing motions, a virtual scene can generate an interactive module that allows users to simulate the wheel-throwing process using a mouse or touchscreen. If the average frame rate of the interactive simulation output reaches 40, which is higher than the frame rate threshold of 30, it indicates that the smoothness is acceptable. This high smoothness ensures that users have a near-realistic interactive experience in the virtual environment.

[0157] For example, when optimizing interactive simulation output using Blender tools, details of ceramic models in virtual scenes can be enhanced, such as adding realistic lighting effects or clay textures.

[0158] Specifically, Blender can adjust the edge smoothness and material reflectivity of models, making the reproduction of craftsmanship more visually realistic. This optimization improves the presentation quality of digital handicraft content.

[0159] In one embodiment, when using the YouTube API for multi-channel information dissemination, optimized technique reproduction content can be uploaded to the YouTube platform.

[0160] For example, a short video showcasing virtual ceramic making can be generated, automatically adding subtitles and background music, and distributed to multiple relevant channels via API, such as handicraft teaching and cultural heritage channels.

[0161] It should be noted that this method of dissemination can attract a wide audience interested in handicrafts, thereby improving the efficiency and influence of the dissemination of skills.

[0162] like Figure 4 As shown, in a second aspect, the present invention provides an intangible cultural heritage information promotion system, which promotes intangible cultural heritage information using the method described above. The system mainly includes: a data acquisition and structuring module, used to collect videos of artisans' operations and sensor data, and to group action sequences using a clustering algorithm to obtain structured coded data, wherein the clustering algorithm categorizes similar technique details based on feature vector distance calculation; a latent feature extraction and simulation module, used to extract latent features such as intensity and rhythm from the structured coded data, and to train a reproducible virtual simulation module using a deep learning model to obtain a digital representation of experiential knowledge; and an interactive query and matching module, used to construct an interactive database based on the digital representation, and to retrieve matching items from the database if a user queries specific technique details, determining a first similarity threshold for the search results. The system comprises four modules: a value to determine the output guidance sequence; a real-time feedback and correction module to integrate a real-time feedback mechanism for the output guidance sequence, which compares and analyzes user-uploaded practice video data to obtain deviation correction suggestions to optimize the knowledge reproduction process; a dynamic platform iteration module to update the content library of the dynamic platform based on the deviation correction suggestions, which triggers automatic iteration if the correction suggestions exceed a preset correction threshold to obtain an enhanced cross-regional interaction model; a personalized learning distribution module to distribute personalized learning paths from the enhanced cross-regional interaction model, which uses a clustering algorithm to classify user regional data and determines the appropriate real-time guidance resources based on the classification results; and a virtual interaction generation module to generate virtual interaction scenes based on the appropriate real-time guidance resources, which simulates the artisan's response through a deep learning model to obtain the final skill reproduction output to support widespread dissemination. The above description is merely a preferred embodiment of this application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of this application. For example, technical solutions formed by replacing the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for promoting information on intangible cultural heritage, characterized in that, The method includes: S1. By collecting videos of artisans' operations and sensor data, the action sequences are grouped and processed to obtain structured coded data. S2. Implicit features such as force and rhythm are extracted from the structured coded data to generate reproducible virtual simulation modules, obtaining a digital representation of experiential knowledge. S3. An interactive database is constructed based on the digital representation. If a user queries specific skill details, matching items are retrieved from the database, and the first similarity threshold of the search results is determined to determine the output guidance sequence. S4. A real-time feedback mechanism is integrated into the output guidance sequence. By acquiring and analyzing practice video data uploaded by users, deviation correction suggestions are obtained to optimize the knowledge reproduction process. S5. The content library of the dynamic platform is updated based on the deviation correction suggestions. If the correction suggestions exceed a preset correction threshold, automatic iteration is triggered to obtain an enhanced cross-regional interactive model. S6. Personalized learning paths are distributed from the enhanced cross-regional interactive model. User regional data is classified, and the classification results are used to determine suitable real-time guidance resources. S7. Virtual interactive scenes are generated based on the suitable real-time guidance resources, simulating the artisan's response, to obtain the final skill reproduction output to support widespread dissemination.

2. The method for promoting intangible cultural heritage information according to claim 1, characterized in that, Step S1 involves collecting video footage of the artisan's actions and sensor data, grouping the action sequences, and obtaining structured coded data, including: Step S11: Record the artisan's movements with a camera and collect movement parameters with an accelerometer and gyroscope to obtain the original movement sequence; Step S12: Extract the first feature vector from the original action sequence; Step S13: Group the first feature vector into action groups; Step S14: Generate the first structured data based on the action grouping results; Step S15: Generate encoded data from the first structured data; Step S16: Extract action patterns from the encoded data to obtain an action pattern set; Step S17: Generate compressed data based on the action pattern set.

3. The method for promoting intangible cultural heritage information according to claim 2, characterized in that, Step S11 involves recording the artisan's movements using a camera and collecting movement parameters using an accelerometer and gyroscope to obtain an original movement sequence, which includes: the original movement sequence consisting of a video frame sequence and a sensor time sequence.

4. The method for promoting intangible cultural heritage information according to claim 3, characterized in that, Step S12, which extracts the first feature vector from the original action sequence, includes: for the video frame sequence, calculating the optical flow vector and aggregating it into trajectory features; for the sensor time series, calculating the average acceleration and the standard deviation of angular velocity within a fixed time window to form the first feature vector.

5. The method for promoting intangible cultural heritage information according to claim 4, characterized in that, Step S13, which involves grouping the first feature vector into action groups, includes: using the K-means clustering algorithm to group the first feature vector, iteratively assigning cluster centers based on Euclidean distance, and if the distance from the first feature vector to the center is less than a preset distance threshold, then they are grouped into the same group to obtain the action grouping result.

6. The method for promoting intangible cultural heritage information according to claim 5, characterized in that, Step S14, generating first structured data based on action grouping results, includes: for each action sequence, extracting cluster centers as representative vectors, and combining timestamps and action tags to generate first structured data.

7. A method for promoting intangible cultural heritage information according to claim 6, characterized in that, Step S15, generating encoded data from the first structured data, includes: for the first structured data, using a sequential encoding method, mapping action labels and cluster centers to fixed-length codes to obtain encoded data.

8. The method for promoting intangible cultural heritage information according to claim 1, characterized in that, Step S7, which generates a virtual interactive scene based on the adapted real-time guidance resources, simulates the artisan's response, and obtains the final skill reproduction output to support widespread dissemination, includes: Step S71: Obtain behavioral data from the activities of artisans through a preset data acquisition module, extract features from the behavioral data, and obtain a set of behavioral features; Step S72: If the completeness of the behavioral feature set is greater than the preset completeness threshold, then perform time series analysis on the behavioral feature set to generate the artisan response sequence. Step S73: Based on the artisan's response sequence, construct a virtual interactive scene and obtain virtual scene data; Step S74: By mapping the virtual scene data with the artisan's response sequence, dynamic interactive content is generated to obtain the interactive simulation output; Step S75: If the average frame rate of the interactive simulation output is greater than the preset frame rate threshold, then the interactive simulation output is optimized to obtain the skill reproduction content. Step S76: Based on the content of the technique reproduction, disseminate information through multiple channels to achieve widespread dissemination and output.

9. A method for promoting intangible cultural heritage information according to claim 8, characterized in that, In step S72, if the completeness of the behavioral feature set is greater than a preset completeness threshold, then a time series analysis is performed on the behavioral feature set to generate a craftsman's response sequence, including: The completeness of the behavioral feature set is calculated using the following formula: Where J represents the completeness of the behavioral feature set, F represents the actual number of behavioral features extracted, and T represents the expected number of behavioral features extracted.

10. A system for promoting information on intangible cultural heritage, characterized in that, The system for promoting intangible cultural heritage information using the method described in any one of claims 1-9 includes: a data acquisition and structuring module for grouping and processing action sequences by collecting videos of artisans' operations and sensor data to obtain structured coded data; a latent feature extraction and simulation module for extracting latent features such as intensity and rhythm from the structured coded data to generate a reproducible virtual simulation module, thereby obtaining a digital representation of experiential knowledge; an interactive query and matching module for constructing an interactive database based on the digital representation, retrieving matching items from the database if a user queries specific skill details, and determining a first similarity threshold of the search results to determine the output guidance sequence; and a real-time feedback and correction module for adjusting the output... The system integrates a real-time feedback mechanism with a guidance sequence, comparing and analyzing user-uploaded practice video data to obtain deviation correction suggestions to optimize the knowledge reproduction process. A dynamic platform iteration module updates the platform's content library based on these correction suggestions; if the suggestions exceed a preset threshold, automatic iteration is triggered to obtain an enhanced cross-regional interaction model. A personalized learning distribution module distributes personalized learning paths from the enhanced cross-regional interaction model, classifies user regional data, and determines the appropriate real-time guidance resources based on the classification results. A virtual interaction generation module generates virtual interactive scenarios based on the appropriate real-time guidance resources, simulating artisan responses to obtain the final skill reproduction output for widespread dissemination.

Images