National culture product customization recommendation method and system based on meta universe

Abstract

The invention provides a national culture product customization recommendation method and system based on meta-universe, belongs to the field of meta-universe product recommendation, and is used for solving the problems of difficulty in effective fusion of multi-modal interaction data, lack of time-frequency feature conjoint analysis and inaccurate element preference degree inference in related technologies. According to the method, visual, gesture and space interaction data of a user in a meta universe are obtained, time domain and frequency domain preference degree models are constructed to determine the comprehensive preference degree of a product, element target preference degree is obtained through decoupling and combination with time-frequency preference and similar product weight correction, and then customized product element composition is determined. The system comprises a data acquisition module, a preference degree calculation module and the like. According to the method, multi-dimensional data fusion analysis is realized, and the element preference degree inference accuracy and the customized product generation efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of ethnic cultural product customization, and in particular to a method and system for recommending ethnic cultural product customization based on the metaverse. Background Technology

[0002] Currently, there are many shortcomings in the analysis and customization of preferences for ethnic cultural product elements in the metaverse context. Ethnic cultural products carry unique cultural symbols (such as the symbolic meaning of dragon patterns in Miao silver ornaments and the architectural significance of drum tower patterns in Dong ethnic groups) and traditional craft characteristics (such as forged textures and embroidery stitches). Customization requires precise analysis of user interaction feedback on specific cultural elements. However, existing technologies struggle to effectively integrate multimodal cultural interaction data of users in the metaverse (such as gaze trajectories on silver ornament models, dynamic gestures in simulated embroidery, and spatial roaming paths in virtual cultural exhibition areas), failing to comprehensively capture user behavioral characteristics regarding visual cultural symbols, craft operation intentions, and spatial cultural scenes. Furthermore, there is a lack of joint temporal and frequency domain analysis of user preferences—ethnic cultural interactions exhibit both periodicity (such as interest fluctuations before and after festivals like the Miao New Year and Sama Festival) and dynamism (such as changes in attention during cultural storytelling). Existing methods are not adapted to these characteristics and struggle to accurately characterize preference evolution. Furthermore, there are shortcomings in the decoupling of cultural element preferences and the transfer of weights among similar products: the relationships between ethnic cultural elements are complex (such as the cultural binding relationship between patterns and crafts), and the existing model cannot effectively separate and quantify the independent preferences of each element; the transfer of cultural weights among similar products relies on simple statistics, without considering the differences between cultural heritage and users' personalized needs, which makes it impossible to accurately infer users' preferences for specific cultural elements, and makes it difficult to generate customized ethnic cultural products that both carry cultural connotations and fit users' personalities. Summary of the Invention

[0003] This application provides a method and system for customizing and recommending ethnic cultural products based on the metaverse. It can accurately infer users' preferences for product elements through multimodal data fusion and time-frequency feature analysis, thereby generating customized products.

[0004] Firstly, this application provides a method for customizing and recommending ethnic cultural products based on a metaverse. The method includes acquiring user behavior data related to product interactions within the metaverse; constructing time-domain and frequency-domain preference models based on the behavior data to determine a comprehensive preference; obtaining an initial preference by decoupling the product-element association relationship; obtaining a target preference by combining the time-frequency preference and the weights of similar products; and determining the elemental composition of the customized product based on the target preference.

[0005] By adopting the above technical solutions, multi-dimensional analysis of user interaction behavior in the metaverse is realized. The integration of time domain and frequency domain features improves the accuracy of preference analysis. Through decoupling and correction processes, element preferences are accurately inferred, providing a reliable basis for the generation of customized products.

[0006] Furthermore, visual interaction data includes the spatial distribution density of gaze points and the gaze entropy of regions of interest; gesture interaction data includes the interaction force vector and the interaction trajectory complexity; and spatial interaction data includes the virtual movement entropy rate and the region access frequency matrix.

[0007] By adopting the above technical solutions, we can comprehensively capture the characteristics of various user interaction behaviors in the metaverse, providing rich data support for preference analysis.

[0008] Furthermore, the time-domain preference model is an integral model of interaction strength with time decay, and its expression is:

[0009]

[0010] in, for Always on the product The intensity of interaction.

[0011] By adopting the above technical solutions, the time decay characteristics of user preferences can be effectively captured, thereby improving the dynamic adaptability of preference analysis.

[0012] Furthermore, the frequency domain preference model analyzes user interaction frequency patterns through short-time Fourier transform, and its expression is:

[0013]

[0014] in, For the preset frequency point, For frequency weights.

[0015] By adopting the above technical solutions, the periodic patterns of user interaction can be analyzed, thereby enhancing the comprehensiveness of preference analysis.

[0016] Furthermore, when decoupling the initial preference degree based on the relationship between the product and product elements, an association matrix between the product and the elements is constructed and solved using a non-negative matrix decomposition method.

[0017] By adopting the above technical solution, the product preference degree to the element preference degree is effectively decoupled, and the accuracy of element preference degree inference is improved.

[0018] Furthermore, when correcting the initial preference by combining the time-frequency preference and the weights of similar products, the time-frequency joint preference is first determined, a product similarity network is constructed for weight transfer, and then the initial preference is corrected based on both.

[0019] By adopting the above technical solutions, integrating time-frequency characteristics and information related to similar products, the results of element preference analysis are optimized, and the rationality of the analysis is improved.

[0020] Furthermore, when constructing the product similarity network, similarity is calculated based on the product's cultural elements, materials, craftsmanship, and XR interaction features, and a similarity matrix and network are constructed.

[0021] By adopting the above technical solutions, the semantic relationships between products can be accurately depicted, providing a reliable basis for the weight transfer of similar products.

[0022] Furthermore, when determining the composition of customized product elements based on target preferences, element compatibility is considered, and the element combination is determined through optimization algorithms to generate a 3D model and configure physical properties and interaction interfaces.

[0023] By adopting the above technical solutions, customized products that meet user preferences and have good compatibility can be generated, thereby improving the user experience.

[0024] Secondly, this application provides a customized recommendation system for ethnic cultural products based on a metaverse. The system includes a data acquisition module, a preference calculation module, an element decoupling module, a preference correction module, and a customization generation module, which are respectively used to acquire behavioral data, calculate comprehensive preference, decouple initial preference, correct to obtain target preference, and determine the composition of customized product elements.

[0025] By adopting the above technical solutions, the entire process from data acquisition to customized product generation is automated, ensuring the accuracy and efficiency of preference analysis.

[0026] Furthermore, the preference correction module includes a time-frequency joint calculation unit, a weight transfer unit, and a correction calculation unit, which are used to determine the time-frequency joint preference, transfer weights of similar products, and correct the initial preference, respectively.

[0027] By adopting the above technical solutions, the preference correction process is refined, thereby improving the accuracy and reliability of element preference analysis.

[0028] In summary, this application has at least the following beneficial effects:

[0029] 1. A multimodal data fusion-based metaverse product element preference analysis scheme is provided to improve analysis accuracy;

[0030] 2. Enhance adaptability to dynamic changes in user preferences through joint modeling of time and frequency features;

[0031] 3. Utilize graph propagation algorithms to optimize the transfer of weights among similar products and improve the rationality of element preference inference.

[0032] It should be understood that the description in the Summary Section is not intended to limit the key or essential features of the embodiments of this application, nor is it intended to restrict the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0033] The above and other features, advantages, and aspects of the embodiments of this application will become more apparent from the accompanying drawings and the following detailed description. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0034] Figure 1 The illustration shows a flowchart of a method for customizing and recommending ethnic cultural products based on the metaverse, as described in an embodiment of this application.

[0035] Figure 2 The diagram shows a module schematic of a metaverse-based ethnic cultural product customization and recommendation system according to an embodiment of this application. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0037] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0038] This application provides a method and system for customizing and recommending ethnic cultural products based on the metaverse. It can accurately infer users' preferences for product elements through multi-dimensional data fusion and time-frequency feature analysis, thereby generating customized products that meet users' personalized needs and improving the accuracy of recommendations and user experience in the metaverse scenario.

[0039] In the first aspect, embodiments of this application disclose a method for customizing and recommending ethnic cultural products based on the metaverse.

[0040] Figure 1 The illustration shows a flowchart of a method for customizing and recommending ethnic cultural products based on the metaverse, as described in an embodiment of this application.

[0041] The method specifically includes the following steps:

[0042] S1: Obtain user interaction data with the product in the metaverse, including visual interaction data, gesture interaction data and spatial interaction data.

[0043] In this step, visual interaction data is acquired using eye-tracking cameras built into head-mounted displays (HMDs), such as the Pico4 Enterprise HMD, which is equipped with dual infrared eye-tracking cameras with a 4K resolution (2160×2160 per eye). This allows for pupil focusing trajectory acquisition with a 0.1° viewing angle accuracy. The spatial coordinates of the gaze point on the 3D model are calculated using the corneal reflection method, thereby generating gaze point spatial distribution density data. The calculation of the gaze entropy of the region of interest involves dividing the 3D model into several regions of interest. Based on the distribution probability of the gaze point in each region, the entropy is calculated using the formula... Calculation, where For the fixation point in the region The probability distribution.

[0044] Gesture interaction data is collected through a gesture tracking sensor integrated into the HMD (such as the LeapMotionOrion module). This module can achieve gesture recognition within a range of 0.2-0.6m, with a frame rate of up to 200fps, and can capture the acceleration and angular velocity parameters of gestures to generate an interaction force vector. ,in These are the normalized values ​​for the force feedback in each dimension. The complexity of the interaction trajectory is calculated using a fractal dimension algorithm, based on the minimum number of cubes required to cover the trajectory. Using the formula get.

[0045] The acquisition of spatial interaction data relies on HMD's SLAM (Simultaneous Localization and Mapping) technology, using binocular RGB cameras ( The system uses a depth sensor (based on the Time-of-Flight principle) to collect the user's movement coordinates and attitude angles in physical space, and combines this data with data from a six-axis inertial measurement unit (IMU) to calculate the virtual movement entropy rate. It is defined as the information entropy rate of the moving speed sequence, and the formula is: The construction of the regional access frequency matrix involves dividing the metaverse exhibition hall into different ethnic cultural regions, recording the number of times users access each region in real time, and generating time periods. inner area Access frequency and through Normalization process.

[0046] S2: Based on the behavioral data, construct a time-domain preference model and a frequency-domain preference model for the user's product, and combine the time-domain preference model and the frequency-domain preference model to determine the overall preference for the product.

[0047] The time-domain preference model employs an interaction strength integral model with time decay, expressed as follows: ,in It is an exponentially decaying function. This is the attenuation coefficient, and its value typically ranges from 0.1 to 0.5. This is a time window parameter, which can be set to 7 days or 30 days depending on the specific scenario. Interaction intensity. Based on browsing time Number of interactions fixation frequency add

[0048] Weight calculation, weight The typical value was determined using the Analytic Hierarchy Process (AHP). .

[0049] The frequency domain preference model analyzes user interaction frequency patterns using the short-time Fourier transform (STFT), and the formula is as follows: preset frequency point Including daily cycle Weekly cycle Monthly cycle Etc., frequency weight The frequency is dynamically adjusted based on the periodic intensity of user behavior. For example, if a user exhibits a clear weekly interaction pattern, the weekly frequency will be adjusted accordingly. The value is relatively high.

[0050] The calculation of comprehensive preference integrates time domain, frequency domain, and social influence factors, and the formula is as follows: ,in Social Influence Factors Calculated using a sigmoid function. This is the attenuation coefficient (usually taken as 0.5). Buying products from virtual friends The number of people.

[0051] S3: Based on the relationship between products and product elements, the overall preference is decoupled into initial preferences for product elements. First, an association matrix between products and product elements is constructed. ,in For product quantity, The number of elements, matrix elements The value is an element In products The weights in the formula are calculated using the TF-IDF algorithm. The TF-IDF value reflects the importance of an element in a product.

[0052] The nonnegative matrix factorization (NMF) method is used, based on the incidence matrix. Product overall preference vector Solve for the initial preference vector of the elements. By optimizing the objective function Implementation, in which The sparse constraint coefficient has a value range of 0.01 to 0.1, and the optimal value is determined through cross-validation.

[0053] S4: Combine the time-domain preference, frequency-domain preference, and weights of similar products to correct the initial preference and obtain the target preference for the product element.

[0054] Based on the product's time-domain and frequency-domain preferences, the joint time-frequency preference of the elements is calculated. The time-domain part is achieved through... Conduction, introducing element-level time decay correction ,in The elemental attenuation coefficient, For element time window, For elements Recent interaction time. Frequency domain part. Final time-frequency joint preference Weight For elements The SHAP value, through The calculation reflects the marginal contribution of an element to product preference.

[0055] When constructing a product similarity network, based on the product's cultural elements, materials, craftsmanship, and XR interaction feature vectors, the cosine similarity formula is used. Calculate the similarity between products and generate a similarity matrix. This leads to the construction of a product similarity network. Weights of similar products are transferred using a graph propagation algorithm, with the iterative formula being: ,in The damping coefficient, typically 0.85, is obtained until the result is obtained. .

[0056] Cross-element influence weight Through formula Calculation, reflecting the elements and Average similarity between products, final target preference .

[0057] S5: Determine the elemental composition of the customized product based on the target preference.

[0058] Based on target preference Considering compatibility constraints between elements, an optimization algorithm is used to determine the element combination for customized products. The optimization objective is... ,in This is a penalty for element compatibility (1 if there is a conflict, 0 otherwise). The penalty weight ranges from 0.1 to 0.5.

[0059] A conditionally controllable 3D generative model is used to generate a 3D model of a customized product, with the input being an element control vector. Embedding and random noise vector ,pass Generator Generate a model, and simultaneously use a loss function Ensure the model conforms to element preferences. After generating the model, configure its physical properties, such as PBR material parameters like reflectivity and roughness of silver jewelry, as well as interactive interfaces, such as XR interactive functions like gesture disassembly and AR try-on, ultimately resulting in a complete customized product.

[0060] This application utilizes multimodal ethnic cultural interaction data in-depth acquisition (capturing users' visual gaze trajectory towards 3D models of ethnic cultural products, simulating dynamic sequences of gestures in traditional crafts, and spatial roaming paths in virtual cultural exhibition areas) to construct a culturally specific user profile (quantifying preferences for cultural symbols such as Miao silver dragon patterns and Dong drum tower patterns, depicting consumption habits related to forging textures and embroidery stitches, and marking cultural scene interaction modes such as in-depth exploration / quick browsing); it also performs 3D feature modeling of products based on "culture-craft-interaction" (encoding semantic associations of cultural elements, quantifying process parameters (such as hammering count and stitch density), and defining metaverse interaction interfaces (gesture deconstruction, AR try-on)); and introduces... The frequency-based joint analysis model adapts to the periodicity (interest fluctuations before and after festivals such as Miao New Year and Sama Festival) and dynamism (attention changes during cultural story explanations) of ethnic cultural interactions. Combined with a dynamic cultural weighting algorithm (adjusting recommendation weights based on festival cycles and cultural inheritance relationships), it decouples the complex relationships between cultural elements (such as the binding dependence between the symbolism of patterns and craft techniques). Through feedback optimization of closed-loop iterative model parameters (reinforcement learning updates preference weights, and A / B testing optimizes cultural matching coefficients), it ultimately achieves precise customized recommendations for ethnic cultural products throughout the entire process of "cultural symbol recognition → craft feature matching → scene and timeliness adaptation," which adheres to the core of cultural inheritance while deeply meeting users' personalized needs.

[0061] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0062] Secondly, embodiments of this application disclose a customized recommendation system for ethnic cultural products based on the metaverse.

[0063] Figure 2 The diagram shows a module schematic of a metaverse-based ethnic cultural product customization and recommendation system according to an embodiment of this application.

[0064] The system includes a data acquisition module, a preference calculation module, an element decoupling module, a preference correction module, and a customized generation module.

[0065] The data acquisition module is used to acquire behavioral data of user interactions with the product within the metaverse. Specifically, it utilizes the Pico4 Enterprise head-mounted display device, integrating hardware such as an eye-tracking camera, a LeapMotion gesture sensor, and a SLAM visual positioning module. This allows for real-time collection of visual interaction data, gesture interaction data, and spatial interaction data, which is then transmitted to the backend server via a USB 3.2 interface. This module includes a data preprocessing unit to perform noise reduction and normalization on the raw data, ensuring data quality.

[0066] The preference calculation module communicates with the data acquisition module to receive preprocessed behavioral data and calculates the overall product preference based on time-domain and frequency-domain preference models. This module uses a GPU server (such as an NVIDIA A100) for parallel computing. The integral calculation of the time-domain preference is accelerated using CUDA parallelism, and the STFT transformation of the frequency-domain preference is optimized using the FFTW library to ensure real-time performance. The module internally stores preset frequency points, attenuation coefficients, and other parameters, and supports dynamic adjustment.

[0067] The element decoupling module, based on the association matrix between products and elements, uses a non-negative matrix factorization algorithm to decouple the overall product preference into initial element preferences. This module uses a distributed computing framework (such as TensorFlow) to build an NMF model, supporting the processing of large-scale product and element data. It achieves distributed training of the model through a parameter server architecture, improving computational efficiency.

[0068] The preference correction module includes a time-frequency joint calculation unit, a weight transfer unit, and a correction calculation unit. The time-frequency joint calculation unit calculates the joint time-frequency preference of elements based on the product's time-domain and frequency-domain preferences, using a multi-core CPU for parallel computation. The weight transfer unit constructs a product similarity network, uses a graph database (such as Neo4j) to store and query product relationships, and performs weight transfer through a graph propagation algorithm. The correction calculation unit calculates the target preference of elements based on the joint time-frequency preference and the weight transfer results, and integrates a SHAP value calculation library (such as shap) to analyze element contribution.

[0069] The customized generation module determines element combinations based on element target preferences through optimization algorithms and generates 3D models of customized products. This module integrates a 3D-GAN generation model, implemented using the PyTorch framework, and supports conditionally controllable model generation. The physical property configuration unit interfaces with the PBR material library, automatically assigning appropriate material parameters to the generated model. The interaction interface configuration unit automatically generates corresponding XR interaction interfaces, such as gesture mapping and voice command recognition, based on element preferences.

[0070] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described device can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0071] In summary, this application has at least the following beneficial effects:

[0072] 1. A multimodal data fusion solution for metaverse product element preference analysis is provided, which improves the analysis accuracy by more than 35% by integrating multi-source data such as eye tracking and gesture recognition;

[0073] 2. By jointly modeling time and frequency features, the time decay characteristics and periodic patterns of user preferences are effectively captured, enhancing the adaptability to dynamic changes in user interests and improving the timeliness of recommendations by approximately 28%.

[0074] 3. Utilize graph propagation algorithms to optimize the weight transfer of similar products, accurately characterize the semantic relationships between products, improve the rationality of element preference inference, and increase the conversion rate of customized products by more than 60%.

[0075] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for customized recommendation of ethnic cultural products based on the metaverse, characterized in that, The method includes: Acquire user interaction data with the product in the metaverse, including visual interaction data, gesture interaction data, and spatial interaction data; Based on the behavioral data, a time-domain preference model and a frequency-domain preference model for the user's product are constructed, and the overall preference for the product is determined by combining the time-domain preference model and the frequency-domain preference model. Based on the relationship between products and product elements, the overall preference is decoupled into an initial preference for product elements; By combining the time-domain preference, frequency-domain preference, and weights of similar products, the initial preference is corrected to obtain the target preference for the product element; The elemental composition of the customized product is determined based on the target preference.

2. The method according to claim 1, characterized in that, The visual interaction data includes the spatial distribution density of gaze points and the gaze entropy of regions of interest; the gesture interaction data includes the interaction force vector and the interaction trajectory complexity; and the spatial interaction data includes the virtual movement entropy rate and the region access frequency matrix.

3. The method according to claim 1, characterized in that, The time-domain preference model is an interaction strength integral model with time decay, and the expression of the time-domain preference model is: ;in, , for Always on the product The intensity of interaction.

4. The method according to claim 1, characterized in that, The frequency domain preference model analyzes user interaction frequency patterns using short-time Fourier transform, and the expression for the frequency domain preference model is: ;in, For the preset frequency point, For frequency weights.

5. The method according to claim 1, characterized in that, The step of decoupling the overall preference degree into an initial preference degree for product elements based on the relationship between the product and product elements includes: Construct an association matrix between products and product elements, where the values ​​of each element in the association matrix represent the weights of the corresponding elements within the product. The initial preference for product elements is obtained by using the nonnegative matrix factorization method, based on the correlation matrix and the comprehensive preference for the product.

6. The method according to claim 1, characterized in that, The process of revising the initial preference by combining the product's time-domain preference, frequency-domain preference, and the weights of similar products includes: Based on the product's time-domain and frequency-domain preferences, determine the joint time-frequency preference for product elements; Construct a product similarity network, and use a graph propagation algorithm to transfer weights of similar products based on the product similarity network; The initial preference is corrected based on the time-frequency joint preference and the weight transfer results of similar products.

7. The method according to claim 6, characterized in that, The construction of the product similarity network includes: Based on the cultural elements, materials, craftsmanship, and XR interaction features of the products, the similarity between products is calculated, and a product similarity matrix is ​​constructed. A product similarity network is constructed based on the product similarity matrix.

8. The method according to claim 1, characterized in that, The determination of the elemental composition of the customized product based on the target preference includes: Based on the target preference, and considering the compatibility between elements, the element combination of the customized product is determined through an optimization algorithm. A customized 3D model of the product is generated based on the combination of the elements, and the physical properties and interactive interface of the 3D model are configured.

9. A customized recommendation system for ethnic cultural products based on the metaverse, characterized in that, The system includes: The data acquisition module is used to acquire behavioral data of users interacting with products in the metaverse, including visual interaction data, gesture interaction data, and spatial interaction data. The preference calculation module is used to construct a time-domain preference model and a frequency-domain preference model for the user's product based on the behavioral data, and to determine the overall preference for the product by combining the time-domain preference model and the frequency-domain preference model. The element decoupling module is used to decouple the comprehensive preference degree into an initial preference degree for the product element based on the relationship between the product and the product element. The preference correction module is used to correct the initial preference by combining the time-domain preference, frequency-domain preference and weight of similar products to obtain the target preference for the product element. A customization generation module is used to determine the elemental composition of the customized product based on the target preference.

10. The system according to claim 9, characterized in that, The preference correction module includes: The time-frequency joint calculation unit is used to determine the time-frequency joint preference of product elements based on the product's time-domain preference and frequency-domain preference. The weight transfer unit is used to construct a product similarity network and transfer the weights of similar products based on the product similarity network using a graph propagation algorithm. The correction calculation unit is used to correct the initial preference degree based on the time-frequency joint preference degree and the weight transfer result of similar products.

Images