Big data mining method and system applied to VR shopping malls

By determining the interactive mode tags in the VR mall and generating sample data sequences, using the interest prediction network, the problem of new product prediction in the VR mall is solved, the accuracy of product promotion and market response speed is improved, and the trend of user interest is realized Accurate identification and efficient resource allocation.

CN118864069BActive Publication Date: 2024-12-03GUANGZHOU ELECTRONICS TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411347698.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-26
Publication Date
2024-12-03
Estimated Expiration
2044-09-26

AI Technical Summary

Technical Problem

The lack of in-depth data mining and correlation analysis in the VR market makes it difficult to accurately predict users' future purchase intentions and product market potential, especially new products on the shelves. Traditional methods are difficult to directly apply, and the potential connections between different products are ignored.

Method used

By determining the interaction mode tags of VR mall interaction events, a sample data sequence is generated, and the interest prediction network is used to predict the user interest and interaction parameters of the new product after being converted into the activation mode, and the product is optimized based on the characteristic connections of activated products. Promotion strategy.

Benefits of technology

It significantly improves the accuracy of product promotion and market response speed, accurately identify user interest trends, provides scientific basis for formulating product listing strategies, promotes efficient allocation of resources, and quickly responds to changes in consumer demand.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118864069B_ABST
    Figure CN118864069B_ABST
Patent Text Reader

Abstract

The present application provides a big data mining method and system for VR shopping malls, which improves the accuracy of product promotion and market response speed. By accurately determining the interaction mode labels of VR shopping mall interaction events and screening out potential target customer groups based on preset audience label features, the user's interest trend in specific VR products can be effectively identified. Furthermore, by utilizing the big data of product interaction migration in the preparation stage that has feature connections with activated products, generating sample data sequences and training an interest prediction network, it is possible to accurately predict the user interest that may be aroused by new products (in presentation preparation mode) after being converted to activation mode, and then deduce the predicted VR interaction parameters, which not only provides a scientific basis for the formulation of product launch strategies, but also promotes the efficient allocation of resources, allowing shopping malls to more accurately grasp market opportunities and quickly respond to changes in consumer demand.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a big data mining method and system applied to VR shopping malls. Background Art

[0002] With the rapid development of virtual reality (VR) technology, VR shopping malls are gradually gaining favor among consumers as a new shopping experience. In VR shopping malls, users can browse products, try on clothes, experience home layout, etc., which greatly enhances the fun and interactivity of shopping. However, compared with traditional e-commerce, a major challenge facing VR shopping malls is how to efficiently and accurately analyze and utilize user interaction data in a virtual environment to optimize product display, improve user experience, and promote sales conversion.

[0003] At present, data mining in VR shopping malls mainly relies on basic user behavior records, such as clicks, browsing time, etc. Although these data can reflect user interests and preferences to a certain extent, they lack in-depth feature mining and association analysis, making it difficult to comprehensively and accurately predict users' future purchase intentions and the market potential of products. In particular, for newly listed products (i.e., products in presentation preparation mode), traditional data analysis methods are often difficult to apply directly due to the lack of historical interaction data.

[0004] In addition, although different products have different attributes, users may show some common behavior patterns or interests during browsing and interaction. For example, for products with similar functions or design styles, users may show similar interaction characteristics. However, existing technologies often ignore this potential connection between products, resulting in limited breadth and depth of data mining.

[0005] Therefore, there is an urgent need for a data mining method that can deeply mine VR shopping mall interaction data, effectively associate different product attribute knowledge points, and accurately predict the interaction trends of new products. Summary of the invention

[0006] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides a big data mining method applied to a VR shopping mall, the method comprising:

[0007] Determine an interaction mode label of a VR mall interaction event for presenting a first VR product and a second VR product; the product attribute knowledge point of the first VR product is different from the product attribute knowledge point of the second VR product; the presentation mode of the first VR product is a presentation preparation mode; the presentation mode of the second VR product is a presentation activation mode;

[0008] If the interaction mode label of the VR shopping mall interaction event meets the preset audience label feature, a first sample data sequence is generated based on the VR interaction migration big data of the first VR product at this stage; the VR interaction migration big data of the first VR product at this stage has a characteristic connection with the VR interaction parameters of the second VR product at this stage; the first sample data sequence includes the first sample data obtained by the VR interaction migration big data of the third VR product at this stage, and the first training supervision data obtained according to the VR product migration interest of the third VR product at this stage; the product attribute knowledge point of the third VR product is consistent with the product attribute knowledge point of the first VR product and the interaction mode label corresponding to the third VR product matches the interaction mode label corresponding to the first VR product; the presentation mode of the third VR product is the presentation activation mode;

[0009] An interest prediction network generated by AI learning using the first sample data and the first training supervision data is obtained, and the VR interaction migration big data of the first VR product in this stage is loaded into the interest prediction network to generate a predicted VR product migration interest of the first VR product when it is switched to the presentation activation mode; the predicted VR product migration interest is used to determine the predicted VR interaction parameters of the first VR product when it is switched to the presentation activation mode.

[0010] On the other hand, an embodiment of the present application also provides a VR shopping mall platform system, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiments of the present application significantly improve the accuracy of product promotion and the speed of market response. By accurately determining the interaction mode labels of VR shopping mall interaction events and screening out potential target customer groups based on preset audience label features, the user's interest trend in specific VR products can be effectively identified. Furthermore, by utilizing the big data of product interaction migration in the preparation stage that has characteristic connections with activated products, generating sample data sequences and training an interest prediction network, it is possible to accurately predict the user interest that may be aroused by new products (in presentation preparation mode) after being converted to activation mode, and then deduce the predicted VR interaction parameters, which not only provides a scientific basis for the formulation of product launch strategies, but also promotes the efficient allocation of resources, allowing shopping malls to more accurately grasp market opportunities and quickly respond to changes in consumer demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1It is a schematic diagram of the execution flow of the big data mining method applied to the VR shopping mall provided in the embodiment of the present application.

[0013] Figure 2 It is a schematic diagram of the hardware architecture of the VR shopping mall platform system provided in the embodiment of the present application. DETAILED DESCRIPTION

[0014] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a big data mining method applied to a VR shopping mall provided by an embodiment of the present application. The big data mining method applied to a VR shopping mall is introduced in detail below.

[0015] Step S110, determining an interaction mode label for a VR mall interaction event for presenting a first VR product and a second VR product. The product attribute knowledge point of the first VR product is different from the product attribute knowledge point of the second VR product. The presentation mode of the first VR product is a presentation preparation mode. The presentation mode of the second VR product is a presentation activation mode.

[0016] In detail, the VR mall interaction event is a behavior event generated when a user interacts with elements such as products, pages, buttons, etc. in a virtual reality (VR) mall. For example, a user browses products in a VR mall, clicks on products to view details, adds products to a shopping cart, purchases products, etc.

[0017] The interaction mode label is used to identify the behavior pattern or interest tendency of the user in the VR shopping mall interaction event. Examples include: "initial attention", "deep understanding", "purchase decision", etc. These labels reflect the different levels of interest and interaction depth of users in the products.

[0018] The product attribute knowledge points are information points used to describe the product's features, functions, specifications, etc. For example, for a VR gaming helmet, its product attribute knowledge points may include high-resolution display, low-latency response, comfortable wearing design, etc.

[0019] The presentation mode is the way or state of displaying products in VR shopping malls, reflecting the promotion stage or user exposure of the products. For example, "presentation preparation mode" may mean that the product has just been launched and is being marketed; "presentation activation mode" may mean that the product has attracted market attention and is in an active promotion state.

[0020] That is, in this embodiment, it is assumed that the VR mall is a comprehensive virtual shopping platform that provides various types of goods, including electronic products, clothing, household items, etc.

[0021] The first VR product is a new type of virtual reality gaming helmet, and its product attribute knowledge points mainly include high-resolution display, low-latency response, comfortable wearing design, etc. Since the product has just been launched and is still in the early stages of market promotion, its presentation mode is presentation preparation mode.

[0022] The second VR product is a popular virtual sports fitness equipment, and its product attribute knowledge points include multiple sports modes, accurate sports monitoring, personalized training programs, etc. Since this product has received high attention and sales in the market, its presentation mode is the presentation activation mode.

[0023] VR mall interaction events may include users' browsing behavior, click operations, purchase intention expressions, etc. By analyzing these interaction behaviors, the server determines the corresponding interaction mode tags.

[0024] For example, after entering the VR mall, user A first browses the introduction page of the first VR product (game helmet), but does not perform further operations, such as viewing detailed parameters, adding to the shopping cart, etc. The server marks this behavior as an interaction mode label of "initial attention".

[0025] After entering the mall, user B directly clicks the purchase button of the second VR product (sports and fitness equipment) and completes the payment process. The server marks this behavior as an interactive mode label of "purchase decision".

[0026] Step S120, if the interaction mode label of the VR mall interaction event meets the preset audience label feature, a first sample data sequence is generated based on the VR interaction migration big data of the first VR product at this stage. The VR interaction migration big data of the first VR product at this stage has a characteristic connection with the VR interaction parameters of the second VR product at this stage. The first sample data sequence includes the first sample data obtained by the VR interaction migration big data of the third VR product at this stage, and the first training supervision data obtained based on the VR product migration interest of the third VR product at this stage. The product attribute knowledge points of the third VR product are consistent with the product attribute knowledge points of the first VR product, and the interaction mode label corresponding to the third VR product matches the interaction mode label corresponding to the first VR product. The presentation mode of the third VR product is the presentation activation mode.

[0027] In detail, the VR interaction migration big data at this stage refers to a large amount of data generated by users interacting with products in a VR mall at the current stage (such as when the product is in a certain presentation mode), which may include users' browsing time, page visit depth, number of consultations, etc., and is used to analyze users' interest in and behavior patterns of products.

[0028] The sample data is sample data used for machine learning or data analysis, which is usually selected from a large amount of data and is representative. For example, the data obtained from the VR interaction migration big data of the third VR product in this stage can be used as the first sample data to train the interest prediction network.

[0029] In supervised learning, the training supervision data is data used to guide model training, including input data and corresponding output (or label). For example, the data obtained based on the interest of the VR product transfer in this stage of the third VR product can be used as the first training supervision data to guide the training of the interest prediction network.

[0030] In this embodiment, it is assumed that the preset audience label feature is "users who have a high interest in and desire to explore virtual reality technology." In the above example, the "initial attention" interaction mode label of user A does not meet the preset audience label feature, while the "purchase decision" interaction mode label of user B meets the preset audience label feature.

[0031] Taking the first VR product (gaming helmet) as an example, the VR interaction migration big data at this stage may include the user's dwell time when browsing the product page, the depth of the page browsed (such as whether the user viewed technical specifications, user reviews and other detailed content), whether the user consulted with customer service, and other information.

[0032] The server obtains multiple VR product coding features. The VR products represented by these coding features are all in the activation mode, and the product attribute knowledge points are the same as the first VR product (that is, they are all virtual reality related devices). Assume that these products include two other VR game helmets C and D of different brands.

[0033] For products C and D, the server checks whether their corresponding interaction mode tags match those of the first VR product. If they match (for example, both are "initial attention" or "deep understanding", etc.), they are included in the subsequent analysis.

[0034] Assuming that the interaction mode tag of product C matches the first VR product, the server processes based on the VR interaction migration big data of product C at this stage and the VR interaction migration data range of the first VR product.

[0035] For example, the VR interaction migration data range of the first VR product is set to a browsing dwell time between 30 seconds and 120 seconds. If the user browsing dwell time of product C is 80 seconds, which is within this range, the server loads the coding feature of product C into the related VR product coding feature set.

[0036] The server determines a first global number of VR product coding features in the relevant VR product coding feature set and compares it with a first threshold value. The first threshold value is determined based on the loss between a priori predicted VR interaction parameters (such as predicted purchase conversion rate) and a priori actual VR interaction parameters (actual purchase conversion rate) of similar VR products in the past period of time (such as one month).

[0037] Assume that the first threshold value is set to 5, and the number of the current relevant VR product coding feature set is 8 (greater than the first threshold value), the server generates a first sample data sequence based on the VR product coding features in this set.

[0038] Specifically, the product in the set (such as product C) is output as the third VR product, and its VR interaction migration big data (such as browsing dwell time, page depth, etc.) and VR product migration interest degree (such as the degree of interest in the product, which can be comprehensively evaluated by the user's subsequent purchase behavior or re-browsing behavior) are obtained. The VR interaction migration big data of this stage is used as the first example data, and the VR product migration interest degree of this stage is used as the first training supervision data, thereby generating a first example data sequence including the first example data and the first training supervision data.

[0039] If the number in the relevant VR product coding feature set is less than the first threshold value, the server optimizes the VR interaction migration data range, for example, expanding the browsing residence time range to 20 seconds to 150 seconds, and then reprocesses the interaction migration big data of other VR products based on the optimized range to generate the first sample data sequence.

[0040] Step S130, obtaining an interest prediction network generated by AI learning using the first sample data and the first training supervision data, loading the VR interaction migration big data of the first VR product in this stage into the interest prediction network, and generating a predicted VR product migration interest of the first VR product when it is switched to the presentation activation mode. The predicted VR product migration interest is used to determine the predicted VR interaction parameters of the first VR product when it is switched to the presentation activation mode.

[0041] In detail, the interest prediction network is a model trained by a machine learning algorithm, and can predict the user's interest in a product based on input data. For example, it can be a network trained by loading the first sample data into a basic machine learning network, and can predict the user's interest in the first VR product after switching to the presentation activation mode.

[0042] The predicted VR product migration interest is obtained through the interest prediction network. The user's possible interest in the product after it is switched to another presentation mode is predicted by the interest prediction network. It is the user's interest in the first VR product after it is switched to the presentation activation mode, and is used to guide product promotion and marketing strategies.

[0043] The predicted VR interaction parameters are parameters of the user's possible interaction with the product in the future, which are calculated through a model based on the predicted interest or other data. They may include predicted purchase probability, expected number of shares, expected re-browsing time interval, etc., and are used to evaluate the market potential of the product and user behavior trends.

[0044] That is, in this embodiment, the server first obtains a basic machine learning network, such as a multi-layer neural network, and loads the first sample data generated previously into the basic machine learning network for training.

[0045] During the training process, the basic machine learning network will try to predict the interest point based on the first sample data input, and then compare the prediction result with the first training supervision data to determine the first loss function value. By continuously adjusting the parameters of the network, the loss function value is gradually reduced, thereby training an interest prediction network that can accurately predict the interest.

[0046] When the training is completed, the server loads the VR interaction migration big data (such as user browsing time, page depth, etc.) of the first VR product (game helmet) in this stage into the interest prediction network.

[0047] The basic machine learning network generates a prediction value based on the input data, which is the predicted VR product transfer interest level when the first VR product is switched to the presentation activation mode. This interest level can reflect the user's interest in the product after switching from the preparation mode to the activation mode.

[0048] Based on the predicted VR product migration interest, the server further determines the predicted VR interaction parameters of the first VR product when it is switched to the presentation activation mode. For example, the predicted VR interaction parameters may include the user's purchase intention probability, the possibility of sharing the product, the frequency of re-browsing, etc.

[0049] Assuming that the predicted VR product migration interest obtained through the interest prediction network is high, the server predicts that the user has a high probability of purchasing the product, and may accordingly increase the recommendation weight of the product in the VR mall, improve its display priority, or push more relevant discount information and supporting product recommendations to the user.

[0050] Through the above steps, VR products in different presentation modes can be analyzed and predicted based on the user's interactive behavior in the VR mall, providing data support for optimizing product display and marketing strategies.

[0051] That is, the embodiments of the present application significantly improve the accuracy of product promotion and the speed of market response. By accurately determining the interaction mode labels of VR shopping mall interaction events and screening out potential target customer groups based on preset audience label features, the user's interest trend in specific VR products can be effectively identified. Furthermore, by utilizing the big data of product interaction migration in the preparation stage that has feature connections with activated products, generating sample data sequences and training an interest prediction network, it is possible to accurately predict the user interest that may be aroused by new products (in presentation preparation mode) after being converted to activation mode, and then deduce the predicted VR interaction parameters, which not only provides a scientific basis for the formulation of product launch strategies, but also promotes the efficient allocation of resources, allowing shopping malls to more accurately grasp market opportunities and quickly respond to changes in consumer demand.

[0052] In a possible implementation, step S120 includes:

[0053] Step S121, determining a VR interaction migration data range based on the VR interaction migration big data of the first VR product in this stage.

[0054] Step S122, obtaining multiple VR product coding features. One VR product coding feature is used to represent one VR product. The presentation modes corresponding to the VR products represented by the multiple VR product coding features are all the presentation activation modes. The product attribute knowledge points corresponding to the multiple VR products are all the same as the product attribute knowledge points of the first VR product.

[0055] Step S123, obtaining VR commodity coding features respectively representing X VR commodities from the multiple VR commodity coding features. The interaction mode labels respectively corresponding to the X VR commodities all match the interaction mode label corresponding to the first VR commodity. The X VR commodities include the third VR commodity. X is a positive integer, and X is not greater than the global number of the multiple VR commodity coding features.

[0056] Step S124, generating a first sample data sequence based on the VR interaction migration big data corresponding to the X VR product coding features and the VR interaction migration data range of this stage.

[0057] In another example, let's continue to assume that we are currently processing product data in a VR mall, which offers a wide variety of virtual reality products, including gaming equipment, fitness equipment, educational tools, and so on.

[0058] First of all, the first VR product is a newly designed virtual reality painting tool set, whose product attribute knowledge points include high-precision brush sensing, rich color library, convenient creation and saving functions, etc. Since it has just been launched, it is in presentation preparation mode. The big data of VR interaction migration in this stage shows that the average browsing time of users on its page is 40 seconds, the proportion of in-depth browsing (viewing detailed function introductions, user cases, etc.) is 30%, and the proportion of users consulting customer service is 10%. Based on these data, the server determines the range of VR interaction migration data, for example, the browsing dwell time range is set to 30-60 seconds, the in-depth browsing ratio range is set to 20%-40%, and the consultation customer service ratio range is set to 5%-15%.

[0059] Next, the server obtains multiple VR product coding features. Assume that 50 VR product coding features are obtained, each coding feature represents a VR product. These products are tools related to virtual reality creation, such as VR drawing boards of different brands, 3D modeling tools, etc., and they are all in the presentation activation mode.

[0060] Then, among these 50 VR product coding features, the server selects X VR product coding features that match the first VR product interaction mode label. Assume that the interaction mode label of the first VR product is "exploration", which means that the user is mainly learning and exploring the functions and features of the product without a clear purchase intention. After screening, the server found 10 matching VR product coding features, representing 10 different VR products, including the third VR product.

[0061] Next, based on the VR interaction migration big data of this stage corresponding to these 10 VR product coding features and the VR interaction migration data range determined previously, the server begins to generate the first sample data sequence.

[0062] Taking the product corresponding to one of the VR product coding features Z1 as an example, the VR interaction migration big data of this stage shows that the user browsing time is 50 seconds, the in-depth browsing ratio is 35%, and the consultation with customer service ratio is 12%. These data are matched with the VR interaction migration data range and found to be within the range. The server loads Z1 into the relevant VR product coding feature set.

[0063] The server counts the number of VR commodity coding features in the relevant VR commodity coding feature set, assuming that the number in the current set is 7. This number is compared with the first threshold value (assuming it is 5). Since 7 is greater than 5, the server generates the first sample data sequence based on the VR commodity coding features in this set.

[0064] The specific generation process is as follows: the server outputs the VR products in the collection (such as the products corresponding to Z1) as the third VR products, and obtains the VR interaction migration big data of this stage, such as the user browsing residence time of 50 seconds, the deep browsing ratio of 35%, the consulting customer service ratio of 12%, and the VR product migration interest of this stage (assuming that the comprehensive evaluation of subsequent user re-browsing, adding to favorites, etc. is "high interest"). The data such as browsing residence time, deep browsing ratio and consulting customer service ratio are used as the first sample data, and the evaluation of "high interest" is used as the first training supervision data, thereby generating the first sample data sequence including the first sample data and the first training supervision data.

[0065] For another product corresponding to VR product coding feature Z2, the VR interaction migration big data in this stage is browsing dwell time 25 seconds, in-depth browsing ratio 15%, and customer service consultation ratio 3%. Matching with the VR interaction migration data range, it is found that none of them are within the range, and the server will not load Z2 into the relevant VR product coding feature set.

[0066] Through the above steps, the server completes the process of generating the first sample data sequence based on the VR interaction migration big data of the first VR product in this stage. These sample data sequences will be used for subsequent machine learning and prediction to better understand and predict the user's interest and behavior in the first VR product.

[0067] Assume another situation. When the server counts the number of VR product coding features in the relevant VR product coding feature set, it finds that the number is 4, which is less than the first threshold value of 5. At this time, the server will optimize the range of VR interaction migration data. For example, the browsing stay time range is expanded to 20-70 seconds, the depth browsing ratio range is adjusted to 15%-45%, and the customer service consultation ratio range is adjusted to 3%-18%.

[0068] Then, the server re-screens and processes the 50 VR product coding features, including the VR interactive migration big data of this stage and the optimized VR interactive migration data range. Some product coding features that may not originally meet the range meet the requirements under the new range and are loaded into the relevant VR product coding feature set.

[0069] Assume that after optimization, the number of relevant VR product coding feature sets reaches 8, which satisfies the condition that it is greater than the first threshold value of 5. The server generates the first sample data sequence in the same way as before, providing data support for subsequent interest prediction and interaction parameter prediction.

[0070] During the entire process, the server continuously analyzes and processes a large amount of VR product data to accurately determine users' interests and behavior patterns towards different VR products, thereby providing a strong decision-making basis for the optimization and personalized recommendations of VR shopping malls.

[0071] As time goes by, more new data is generated, and the server will continue to update and optimize the VR interaction migration data range, threshold value, and sample data sequence to adapt to changes in the market and user needs. For example, if it is found that the user attention of a certain type of VR product suddenly increases, the server will adjust the relevant parameters and data processing methods accordingly to capture this trend more accurately.

[0072] Assume that over a period of time, new technological breakthroughs have occurred in the field of VR creation tools, causing changes in user needs and expectations for such products. Some important product attribute knowledge points may no longer be so critical, while new features become more attractive to users. The server will promptly perceive this change and re-evaluate and adjust the weights and ranges of product attribute knowledge points when processing VR product coding features and generating sample data sequences.

[0073] For example, new technologies have made wireless connection functions essential in VR painting tools. When filtering and comparing VR product coding features, the server will give higher priority and weight to products with more stable and low-latency wireless connection functions. At the same time, when determining the scope of VR interaction migration data, it will focus on users' browsing and attention to descriptions related to wireless connections.

[0074] In addition, the server will also consider market competition factors. If a large number of similar VR creation tools appear on the market and competition intensifies, users may pay more attention to factors such as price and brand reputation when choosing products. When generating the first sample data sequence, the server will take these factors into consideration to more accurately predict user behavior and interests.

[0075] Suppose a well-known brand launches a VR painting tool with an innovative design, which has attracted widespread attention. When processing relevant data, the server will pay special attention to the differences and similarities between this product and the first VR product, as well as the user's interaction mode with it. If the user shows a high interest in the brand's products and unique interaction behavior, the server will incorporate this information into the generation of the first sample data sequence, providing a more comprehensive and accurate reference for the prediction and optimization of the first VR product.

[0076] In addition, the user's personal preferences and usage habits will also have an impact on the server's processing. Some users may pay more attention to the ease of use of the tool, while others are more concerned about its scalability and compatibility with other software. By analyzing a large amount of user interaction data, the server can identify these different preference patterns and perform targeted processing when generating the first sample data sequence to meet the needs of different user groups.

[0077] Suppose the server finds that the needs and preferences for VR creation tools in a certain region are significantly different from those in other regions. For example, users in a certain region are more inclined to buy painting tool sets with specific color configurations and specific materials. When the server generates the first sample data sequence, it will classify and process the data according to regional differences to provide predictions and recommendations that better meet the needs of local users.

[0078] Moreover, factors such as seasons and holidays may also affect users' interest in VR products. During certain holidays, such as Christmas or New Year, users may be more inclined to buy VR products as gifts, or show a higher level of interest in products with holiday-themed designs. The server will adjust the data processing strategy and the generation method of sample data sequences within the corresponding time period to adapt to these temporary but significant changes in demand.

[0079] Suppose a VR mall launches promotional activities, such as discounts, discounts for purchases above a certain amount, and gifts. These activities may change users' interactive behaviors and purchasing decisions. When generating the first sample data sequence, the server will consider the impact of the promotional activities and analyze the users' browsing, consulting, and purchasing behaviors during the activities in order to more accurately predict the users' interests and needs after the activities.

[0080] In short, in the process of generating the first sample data sequence based on the VR interactive migration big data of the first VR product in this stage, the server needs to comprehensively consider many factors, continuously optimize and adjust the processing methods, so as to provide more valuable predictions and decision-making support, and improve the user experience and operation effect of the VR mall.

[0081] In a possible implementation, the X VR commodity coding features include a VR commodity coding feature Zi. The X current-stage VR interactive migration big data include the current-stage VR interactive migration big data Yi corresponding to the VR commodity coding feature Zi. i is a positive integer, and i is not greater than the global number of the X VR commodity coding features.

[0082] Step S124 includes:

[0083] Step S1241, matching the VR interaction migration big data Yi of this stage with the VR interaction migration data range.

[0084] Step S1242: if the VR interaction migration big data Yi of this stage is within the range of the VR interaction migration data, the VR product coding feature Zi is loaded into the relevant VR product coding feature set.

[0085] Step S1243, determining a first global number of VR product coding features in the relevant VR product coding feature set, comparing the first global number with a first threshold value, and generating a comparison result. The first threshold value is determined based on a loss between a priori predicted VR interaction parameters of the first a priori VR product within a first a priori time period and a priori actual VR interaction parameters of the first a priori VR product. The first a priori VR product belongs to the X VR products.

[0086] Step S1244: generating a first sample data sequence based on the comparison result, the VR product coding features in the related VR product coding feature set, and the VR interactive migration data range.

[0087] Still taking the above virtual reality painting tool set as an example, it is assumed that the selected X VR product coding features include Z1, Z2, Z3, etc.

[0088] For the VR interaction migration big data Y1 corresponding to Z1 in this stage, assume that its specific data is that the user browsing time is 55 seconds, the in-depth browsing ratio is 38%, and the consultation customer service ratio is 10%. The server matches these data with the previously determined VR interaction migration data range (browsing time range 30-60 seconds, in-depth browsing ratio range 20%-40%, consultation customer service ratio range 5%-15%). After comparison, it is found that Y1 is within the VR interaction migration data range, so the server loads the VR product coding feature Z1 into the relevant VR product coding feature set.

[0089] Similarly, for the VR interaction migration big data Y2 corresponding to Z2 in this stage, assuming that the data is that the user browsing time is 20 seconds, the in-depth browsing ratio is 18%, and the consultation with customer service ratio is 2%. Matching with the VR interaction migration data range, it is found that Y2 is not within the range, so Z2 will not be loaded into the relevant VR product coding feature set.

[0090] The server counts the number of VR commodity coding features in the relevant VR commodity coding feature set to determine the first global number thereof. Assume that the number in the current set is 6.

[0091] The first threshold value is determined based on the loss between the a priori predicted VR interaction parameters (such as the predicted purchase conversion rate) and the a priori actual VR interaction parameters (actual purchase conversion rate) of the first a priori VR product in the first a priori time period (such as last month). Assuming that there were 5 similar virtual reality painting tools as the first a priori VR products last month, the predicted purchase conversion rate was 20%, but the actual purchase conversion rate was only 15%. After calculating the loss, the first threshold value is determined to be 4.

[0092] The first global quantity 6 in the current relevant VR commodity coding feature set is compared with the first threshold value 4 to generate a comparison result, that is, 6 is greater than 4.

[0093] Based on this comparison result, since the first global quantity is greater than the first threshold value, the server generates a first sample data sequence based on the VR product coding features in the relevant VR product coding feature set. Specifically, for each VR product coding feature in the set (such as Z1), obtain its corresponding VR interaction migration big data of this stage (such as user browsing dwell time, in-depth browsing ratio, consultation customer service ratio, etc.), as well as the VR product migration interest of this stage obtained through subsequent user behavior evaluation (such as high, medium, low, etc.). These VR interaction migration big data are used as the first sample data, and the product migration interest is used as the first training supervision data, thereby generating a first sample data sequence including the first sample data and the first training supervision data.

[0094] Assume another situation, the first global quantity in the relevant VR product coding feature set is 3, which is less than the first threshold value 4. At this time, the server optimizes the VR interaction migration data range, such as expanding the browsing dwell time range to 25-65 seconds, adjusting the depth browsing ratio range to 18%-42%, and adjusting the consulting customer service ratio range to 4%-16%.

[0095] Then, the server re-screens and processes the VR interaction migration big data corresponding to the X VR product coding features in this stage and the optimized VR interaction migration data range, and loads the VR product coding features that meet the new range into the relevant VR product coding feature set.

[0096] Assume that after reprocessing, the number of relevant VR product code feature sets reaches 5, which satisfies the condition that it is greater than the first threshold value of 4. The server generates the first sample data sequence in the same way as before, providing accurate and effective data support for subsequent interest prediction and interaction parameter prediction.

[0097] In this process, the data processing method is continuously and accurately adjusted and optimized to adapt to the dynamic changes in the market and user behavior, thereby providing more valuable decision-making basis for the operation of VR shopping malls and product recommendations.

[0098] In a possible implementation, step S1244 includes:

[0099] Step S1244-1: if the comparison result is that the first global quantity is not less than the first threshold value, a first sample data sequence is generated based on the VR product coding features in the relevant VR product coding feature set.

[0100] Step S1244-2: If the comparison result is that the first global number is less than the first threshold value, the VR interaction migration data range is optimized to generate an optimized VR interaction migration data range. The optimized VR interaction migration data range is larger than the VR interaction migration data range.

[0101] Step S1244-3: Generate a first sample data sequence based on the X VR product coding features, which respectively include the VR interaction migration big data of this stage and the optimized VR interaction migration data range.

[0102] Still taking the above-mentioned virtual reality painting tool set as an example, it is assumed that the first global quantity in the relevant VR product coding feature set is 8, and the first threshold value is 5.

[0103] Since the comparison result is that the first global quantity 8 is not less than the first threshold value 5, the server generates a first sample data sequence based on the VR product coding features in the relevant VR product coding feature set.

[0104] Specifically, for each VR product coding feature in the set (such as Z1, Z2, Z3, etc.), the server obtains the corresponding VR interaction migration big data of this stage. Assume that the VR interaction migration big data of Z1 in this stage is that the user browsing stay time is 58 seconds, the in-depth browsing ratio is 36%, and the consultation customer service ratio is 8%, and the VR interaction migration big data of Z2 in this stage is that the user browsing stay time is 45 seconds, the in-depth browsing ratio is 30%, and the consultation customer service ratio is 12%, etc.

[0105] At the same time, the server obtains the VR product migration interest of this stage corresponding to these VR product coding features. Assume that the VR product migration interest of Z1 is "high" at this stage, and the VR product migration interest of Z2 is "medium" at this stage, and so on.

[0106] Then, the server uses these VR interaction migration big data (such as browsing dwell time, in-depth browsing ratio, customer service consultation ratio, etc. of Z1, Z2, etc.) as the first sample data, and uses the corresponding VR product migration interest level (such as "high", "medium", etc.) as the first training supervision data to generate the first sample data sequence.

[0107] In another case, it is assumed that the first global quantity in the relevant VR commodity coding feature set is 3, which is less than the first threshold value 5. At this time, the server optimizes the VR interactive migration data range.

[0108] The original VR interaction migration data range is browsing time 30-60 seconds, in-depth browsing ratio 20%-40%, and customer service consultation ratio 5%-15%. The optimized VR interaction migration data range may become browsing time 20-70 seconds, in-depth browsing ratio 15%-45%, and customer service consultation ratio 3%-18%.

[0109] Next, the server re-screens and processes the VR interaction migration big data and the optimized VR interaction migration data range based on X VR product coding features (including Z1, Z2, Z3, etc.).

[0110] For example, the original VR interaction migration big data of Z4 in this stage is that the user browsing time is 25 seconds, the in-depth browsing ratio is 18%, and the consultation ratio is 4%. It does not meet the requirements in the original range, but meets the requirements in the optimized range. The server loads Z4 into the new related VR product coding feature set.

[0111] Finally, the server generates the first sample data sequence based on the new set of relevant VR product coding features in the same way as described above, that is, the VR interaction migration big data that meets the VR product coding features in the optimized range is used as the first sample data, and the corresponding VR product migration interest is used as the first training supervision data to generate the first sample data sequence.

[0112] In a possible implementation, step S1244-1 includes:

[0113] Step S1244-11: output the VR commodity represented by the VR commodity coding feature in the related VR commodity coding feature set as the third VR commodity.

[0114] Step S1244-12, obtaining the VR interaction migration big data of the third VR product in this stage, and the VR product migration interest of the third VR product in this stage.

[0115] Step S1244-13, using the VR interaction migration big data of the third VR product in this stage as the first sample data, and using the VR product migration interest degree of the third VR product in this stage as the first training supervision data, to generate a first sample data sequence including the first sample data and the first training supervision data.

[0116] In this embodiment, it is assumed that the relevant VR product coding feature set includes three VR product coding features: Z1, Z2 and Z3.

[0117] First, the server outputs the VR products represented by Z1, Z2 and Z3 as the third VR product. For example, Z1 represents a virtual reality painting tool set of brand A, Z2 represents a similar product of brand B, and Z3 represents a related tool of brand C.

[0118] Then, the server obtains the VR interaction migration big data of these third VR products in this stage. For brand A's products, its VR interaction migration big data in this stage may include that the average user browsing time is 55 seconds, the in-depth browsing ratio is 35%, and the proportion of consulting customer service is 10%; for brand B's products, the data may be browsing time of 48 seconds, in-depth browsing ratio of 30%, and customer service consultation ratio of 8%; for brand C's products, the browsing time may be 52 seconds, in-depth browsing ratio of 32%, and customer service consultation ratio of 9%.

[0119] At the same time, the server obtains the VR product migration interest of these third VR products in this stage. Assume that the product migration interest of brand A is evaluated as "high", the product of brand B is "medium", and the product of brand C is "high".

[0120] Next, the server uses the VR interaction migration big data of brand A in this stage (browsing dwell time of 55 seconds, in-depth browsing ratio of 35%, and customer service consultation ratio of 10%) as the first sample data, and its VR product migration interest level of "high" in this stage as the first training supervision data.

[0121] Similarly, the VR interaction migration big data of brand B in this stage (browsing dwell time of 48 seconds, in-depth browsing ratio of 30%, and customer service consultation ratio of 8%) is used as the first sample data, and the VR product migration interest level of "medium" in this stage is used as the first training supervision data.

[0122] For Brand C, its VR interaction migration big data in this stage (browsing dwell time of 52 seconds, in-depth browsing ratio of 32%, and customer service consultation ratio of 9%) is used as the first sample data, and the VR product migration interest level of "high" in this stage is used as the first training supervision data.

[0123] Finally, the server generates a first sample data sequence including the first sample data and the corresponding first training supervision data. This first sample data sequence will be used for subsequent analysis and model training to better understand and predict the user's interest and behavior in similar VR products.

[0124] In a possible implementation, step S130 includes:

[0125] Step S131, obtaining a first basic machine learning network, and loading the first sample data into the first basic machine learning network.

[0126] Step S132: using the first basic machine learning network, determine the interest point prediction result corresponding to the first sample data.

[0127] Step S133: determine a first loss function value between the first training supervision data and the interest point prediction result, train the first basic machine learning network based on the first loss function value, and generate an interest prediction network.

[0128] In this embodiment, the server first obtains a first basic machine learning network, such as a multi-layer neural network structure. This network already has basic parameters and architecture, and is ready to receive data for training and optimization.

[0129] Then, the server loads the first sample data generated previously into the first basic machine learning network. Assume that the first sample data includes the VR interaction migration big data of multiple third VR products in this stage, such as the browsing dwell time of brand A's virtual reality painting tool set is 55 seconds, the in-depth browsing ratio is 35%, and the customer service consultation ratio is 10%, and the browsing dwell time of brand B's products is 48 seconds, the in-depth browsing ratio is 30%, and the customer service consultation ratio is 8%.

[0130] Using this first basic machine learning network, the first sample data is processed and analyzed to try to predict the interest points corresponding to each sample data. For example, for the product data of brand A, it is predicted that the user interest level it may arouse is "high"; for the product data of brand B, its interest level is predicted to be "medium".

[0131] Next, the server obtains the first training supervision data, that is, the actual interest in the VR product migration of the third VR products at this stage, such as the actual interest in the product of brand A is "high", and the actual interest in the product of brand B is "medium".

[0132] Then, the server determines the difference between the first training supervision data and the prediction result of the point of interest, calculates the difference through a specific loss function, and obtains a first loss function value. For example, for the product of brand A, there is a certain deviation between the predicted "higher" and the actual "high", and the specific value of the deviation is calculated through the loss function.

[0133] Based on the calculated first loss function value, the server adjusts and optimizes the parameters of the first basic machine learning network. Through multiple iterative training, the loss function value is continuously reduced, so that the prediction result of the network gradually approaches the actual training supervision data.

[0134] After multiple training and optimizations, the server finally generated an interest prediction network that can accurately predict the user's interest in similar VR products. This network can give a relatively accurate interest prediction result based on the input interaction migration data of new VR products, providing support for subsequent product recommendations and marketing strategies.

[0135] In a possible implementation, the method further includes:

[0136] Step A110, based on the predicted VR product migration interest and the VR interaction migration big data of the first VR product in this stage, determine the predicted VR interaction parameters of the first VR product when switching to the presentation activation mode.

[0137] Step A120: Determine a predicted interaction improvement rate corresponding to the first VR product based on the predicted VR interaction parameter of the first VR product and the presented VR interaction parameter of the first VR product. The predicted VR interaction parameter and the predicted interaction improvement rate are both displayed in a product information display page for the first VR product in a VR device.

[0138] In this embodiment, the server obtains the predicted VR product migration interest level of the first VR product (virtual reality painting tool set) when it is switched to the presentation activation mode, assuming that this interest level is "high". At the same time, the server also obtains the VR interaction migration big data of the first VR product at this stage, such as the average browsing stay time of users is 40 seconds, the in-depth browsing ratio is 30%, and the customer service consultation ratio is 8%.

[0139] Based on the predicted VR product migration interest level of "high" and the VR interaction migration big data of this stage, the server begins to determine the predicted VR interaction parameters of the first VR product when it switches to the presentation activation mode. The predicted VR interaction parameters may include the user's expected purchase probability, expected number of shares, expected re-browsing time interval, etc. Assume that based on the current data and interest level, the predicted purchase probability is determined to be 40%, the expected number of shares is 5 times, and the expected re-browsing time interval is 2 days.

[0140] Next, the server obtains the current presentation VR interaction parameters of the first VR product, such as the current actual purchase probability is 20%, the actual number of shares is 2 times, and the actual re-browsing time interval is 5 days.

[0141] Then, the server uses a specific algorithm to calculate the improvement ratio of the predicted VR interaction parameters relative to the presented VR interaction parameters, thereby determining the predicted interaction improvement rate corresponding to the first VR product. Assume that the improvement rate of purchase probability is 100% ((40%-20%) / 20%), the improvement rate of sharing times is 150% ((5-2) / 2), and the reduction rate of the time interval between re-browsing is 60% ((5-2) / 5).

[0142] Finally, the server displays these predicted VR interaction parameters (purchase probability 40%, sharing times 5, re-browsing interval 2 days) and predicted interaction improvement rate (purchase probability increased by 100%, sharing times increased by 150%, re-browsing interval shortened by 60%) on the product information display page for the first VR product in the VR device. When users browse the product, they can clearly see these data and understand the possible popularity of the product in the future and the improvement of the interaction effect.

[0143] In a possible implementation, the method further includes:

[0144] Step B110, if the interaction mode label of the VR mall interaction event does not meet the preset audience label feature, a second sample data sequence is generated based on the interaction mode label of the VR mall interaction event. The second sample data sequence includes the second sample data obtained by the VR interaction migration big data of the fourth VR product at this stage, and the second training supervision data obtained according to the VR interaction parameters of the fourth VR product at this stage. The product attribute knowledge points of the fourth VR product are consistent with the product attribute knowledge points of the first VR product, and the interaction mode label corresponding to the fourth VR product matches the interaction mode label corresponding to the first VR product. The presentation mode of the fourth VR product is the presentation activation mode.

[0145] Step B120, obtaining a second basic machine learning network, loading the second sample data into the second basic machine learning network, and using the second basic machine learning network to determine a VR interaction parameter prediction result corresponding to the second sample data.

[0146] Step B130, determining a second loss function value between the second training supervision data and the VR interaction parameter prediction result, training the second basic machine learning network based on the second loss function value, generating a VR interaction parameter prediction network, loading the VR interaction migration big data of the first VR product at this stage into the VR interaction parameter prediction network, and generating predicted VR interaction parameters of the first VR product when it is converted to the presentation activation mode.

[0147] Assume that the interaction mode label of the VR mall interaction event is "casual browsing", which does not meet the preset audience label characteristics of "users with high interest and desire to explore virtual reality technology".

[0148] The server generates the second sample data sequence based on the interaction mode label of "casual browsing". First, the server searches for a fourth VR product that has the same product attribute knowledge points as the first VR product (virtual reality painting tool set) and a matching interaction mode label (i.e., the same "casual browsing"). Assume that the fourth VR product found is another virtual reality painting auxiliary device.

[0149] The server obtains the VR interaction migration big data of this fourth VR product in this stage, such as the average browsing time of users is 25 seconds, the in-depth browsing ratio is 15%, and the customer service consultation ratio is 3%, and uses it as the second sample data. At the same time, based on the VR interaction parameters of the fourth VR product in this stage, such as the actual purchase times is 2 times, the sharing times is 1 time, etc., the second training supervision data is obtained.

[0150] The server obtains a second basic machine learning network, such as a neural network with a different structure, and then loads the second sample data generated above into the second basic machine learning network.

[0151] The second basic machine learning network is used to process and analyze the second sample data to determine the corresponding VR interaction parameter prediction results. For example, the predicted number of purchases is 1 and the number of shares is 0.

[0152] The server obtains the second training supervision data, namely, the actual purchase times of the fourth VR product, 2 times, and the sharing times, 1 time, and compares it with the above prediction results.

[0153] The difference between the two is calculated through a specific loss function to determine the value of the second loss function. For example, based on the difference between the actual and predicted number of purchases and shares, the loss value is calculated to be 1.5.

[0154] Based on the value of the second loss function, the server adjusts and optimizes the parameters of the second basic machine learning network, and generates a VR interaction parameter prediction network after multiple trainings.

[0155] Finally, the server loads the VR interaction migration big data of the first VR product in this stage into the VR interaction parameter prediction network to generate the predicted VR interaction parameters of the first VR product when it is converted to the presentation activation mode. For example, it is predicted that the purchase number of the first VR product may be 3 times, and the sharing number may be 2 times.

[0156] In a possible implementation, step B110 includes:

[0157] Step B111, obtaining T VR product coding features. One VR product coding feature is used to characterize one VR product. The presentation modes corresponding to the T VR products are all the presentation activation modes. The product attribute knowledge points corresponding to the T VR products are the same as the product attribute knowledge points of the first VR product. The interaction mode labels corresponding to the T VR products do not include the interaction mode label corresponding to the preset audience label feature.

[0158] Step B112, obtaining K VR product coding features from the T VR product coding features. The interaction mode labels respectively included in the K VR product coding features are consistent with the interaction mode label corresponding to the first VR product, and the product experience priorities respectively included in the K VR product coding features are the same as the product experience priorities corresponding to the first VR product. K is a positive integer, and K is not greater than the global number of multiple VR product coding features.

[0159] Step B113, taking K as the second global number, if the second global number is not less than a second threshold value, outputting the VR products represented by the K VR product coding features as the fourth VR product. The second threshold value is determined based on the loss between the a priori predicted VR interaction parameter of the second a priori VR product within the second a priori time period and the a priori actual VR interaction parameter of the second a priori VR product. The second a priori VR product belongs to the T VR products.

[0160] Step B114: if the second global number is less than the second threshold value, U VR product coding features are obtained from the T VR product coding features. The product experience priorities respectively included in the U VR product coding features are the same as the product experience priorities corresponding to the first VR product. U is a positive integer not less than U, and U is not greater than the global number of multiple VR product coding features.

[0161] Step B115: Taking U as the third global quantity, if the third global quantity is not less than the second threshold value, outputting the VR commodities represented by the U VR commodity coding features as the fourth VR commodity.

[0162] Step B116: If the third global number is less than the second threshold value, a product experience priority range is generated based on the product experience priority corresponding to the first VR product and the peripheral product experience priority of the product experience priority corresponding to the first VR product, and L VR product coding features are obtained from the T VR product coding features. The product experience priorities respectively included in the L VR product coding features all belong to the product experience priority range. L is a positive integer not less than U, and L is not greater than the global number of the multiple VR product coding features.

[0163] Step B117, taking L as the fourth global number, if the fourth global number is not less than the second threshold value, outputting the VR products represented by the L VR product coding features as the fourth VR product, taking the VR interaction migration big data of the fourth VR product in this stage as the second sample data, taking the VR interaction parameters of the fourth VR product in this stage as the second training supervision data, generating a first sample data sequence including the second sample data and the second training supervision data, and if the fourth global number is less than the second threshold value, optimizing the product experience priority range to generate an optimized product experience priority range. The optimized product experience priority range is greater than the product experience priority range.

[0164] Step B118, obtaining J VR product coding features from the T VR product coding features. The product experience priorities respectively included in the J VR product coding features all belong to the optimized product experience priority range. J is a positive integer not less than L, and J is not greater than the global number of the multiple VR product coding features.

[0165] Step B119: Generate a second sample data sequence based on the J VR product coding features.

[0166] First, the server obtains T VR product coding features. Assume that T is 100, and the VR products represented by these 100 VR product coding features are all virtual reality painting tools, and the presentation mode is all presentation activation mode. The product attribute knowledge points of these products are the same as the first VR product (a specific virtual reality painting tool set), but their corresponding interaction mode labels do not include the interaction mode labels corresponding to the preset audience label features, such as the specific interaction mode labels corresponding to "users who have a high interest and desire to explore virtual reality technology."

[0167] Then, the server selects K VR product coding features from these 100 VR product coding features, whose interaction mode labels and product experience priorities are the same as those of the first VR product. Assuming that the interaction mode label of the first VR product is "preliminary understanding" and the product experience priority is "intermediate", after screening, the server finds 20 VR product coding features that meet the conditions, that is, K=20.

[0168] The server uses K=20 as the second global number. Assuming the second threshold value is 15, since 20 is not less than 15, the server outputs the VR commodities represented by the 20 VR commodity coding features as the fourth VR commodity.

[0169] Next, the server obtains the VR interaction migration big data of the fourth VR products in this stage, such as the user's browsing time, click times, consultation status, etc., and uses it as the second sample data. At the same time, the VR interaction parameters of the fourth VR products in this stage, such as the purchase quantity, the number of collections, etc., are obtained as the second training supervision data. Then, the server generates a first sample data sequence including the second sample data and the second training supervision data.

[0170] Assume another situation, the number of K VR product coding features obtained after screening is 10, that is, the second global number is 10, which is less than the second threshold value 15. At this time, the server re-acquires U VR product coding features with the same product experience priority as the first VR product from these 100 VR product coding features. Assume that 18 are found this time, that is, U=18.

[0171] The server uses U=18 as the third global quantity. Assuming that the third threshold value is 16 at this time, since 18 is not less than 16, the server outputs the VR products represented by the 18 VR product coding features as the fourth VR product, and obtains the VR interaction migration big data and VR interaction parameters of this stage in the above manner to generate the first sample data sequence.

[0172] Assume again that the number of U VR product coding features is 12, which is less than the third threshold value of 16. At this time, the server generates a product experience priority range based on the product experience priority "medium" corresponding to the first VR product and its surrounding product experience priorities, such as "low" and "high".

[0173] The server obtains L VR product coding features belonging to the product experience priority range from these 100 VR product coding features. Assume that 25 are found, that is, L=25.

[0174] The server uses L=25 as the fourth global number. Assuming that the fourth threshold value is 20, since 25 is not less than 20, the server outputs the VR commodities represented by the 25 VR commodity coding features as the fourth VR commodity, and generates the first sample data sequence.

[0175] If the number of L VR product coding features is 15, which is less than the fourth threshold value of 20, the server optimizes the product experience priority range, for example, expanding the range to include a wider range such as "Intermediate-1" and "Intermediate+1".

[0176] Then, the server obtains J VR product coding features belonging to the priority range of the optimized product experience from the 100 VR product coding features. Assume that 30 are found, that is, J=30.

[0177] Finally, the server generates a second sample data sequence based on these J VR product coding features for subsequent analysis and model training.

[0178] Throughout the entire process, the server continuously screens and adjusts to obtain sample data of appropriate quantity and quality, providing strong support for accurately predicting and analyzing user interaction behaviors with VR products. For example, during the screening process, the server finds that certain brands or functions of VR painting tools are more popular with users within a specific product experience priority range. In subsequent recommendations and marketing strategies, these types of products can be promoted as a priority.

[0179] As the market and user needs change, the server will continue to update and optimize the screening conditions and thresholds. For example, in a certain period of time, the demand for VR painting tools with intermediate product experience priority decreases, while the demand for high-level product experience priority increases. The server will adjust the product experience priority range and threshold accordingly to ensure that more representative and timely sample data is obtained.

[0180] In addition, factors such as seasons and promotions will also affect users' attention to product experience priority. During the school season, painting tools with simple functions and low prices that are prioritized as primary products may be more popular; during large promotions, users may be more inclined to buy painting tools with advanced product experience priority but with discounts. The server will adjust the screening strategy in a timely manner according to these changes to generate a second sample data sequence that is more in line with market dynamics.

[0181] At the same time, the server will also consider the differences between different regions and user groups. Users in certain regions may be more interested in VR painting tools with specific functions or brands, and the distribution of their product experience priorities may also be different. The server will perform personalized screening and data generation based on the characteristics of these regions and user groups to provide more accurate services and recommendations.

[0182] Therefore, a large amount of VR product data can be fully utilized to generate high-quality second sample data sequences, laying a solid foundation for improving the service quality and user experience of VR shopping malls.

[0183] Figure 2 The hardware structure of the VR shopping mall platform system 100 provided in the embodiment of the present application for implementing the above-mentioned big data mining method applied to the VR shopping mall is shown as follows: Figure 2 As shown, the VR shopping mall platform system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0184] In one possible design, the VR mall platform system 100 can be a single server or a server group. The server group can be centralized or distributed (for example, the VR mall platform system 100 can be a distributed system). In some embodiments, the VR mall platform system 100 can be local or remote. For example, the VR mall platform system 100 can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the VR mall platform system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the VR mall platform system 100 can be implemented on a VR mall platform system. By way of example only, the VR mall platform system may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any aggregation thereof.

[0185] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions used by the VR shopping mall platform system 100 to execute or use to complete the exemplary methods described in this application.

[0186] In the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the big data mining method applied to the VR shopping mall in the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0187] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the above-mentioned VR shopping mall platform system 100. The implementation principles and technical effects are similar, and this embodiment will not be repeated here.

[0188] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are set. When a processor executes the computer executable instructions, the big data mining method applied to the VR mall as above is implemented.

[0189] It should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. A big data mining method applied to VR shopping malls, characterized in that: The method comprises: Determine an interaction mode label for a VR mall interaction event for presenting a first VR product and a second VR product; the product attribute knowledge point of the first VR product is different from the product attribute knowledge point of the second VR product; the presentation mode of the first VR product is a presentation preparation mode; the presentation mode of the second VR product is a presentation activation mode, the presentation preparation mode indicates that the product has just been launched and is being marketed; the presentation activation mode indicates that the product has attracted market attention and is in an active promotion state; If the interaction mode label of the VR shopping mall interaction event meets the preset audience label feature, a first sample data sequence is generated based on the VR interaction migration big data of the first VR product at this stage; the VR interaction migration big data of the first VR product at this stage has a characteristic connection with the VR interaction parameters of the second VR product at this stage; the first sample data sequence includes the first sample data obtained by the VR interaction migration big data of the third VR product at this stage, and the first training supervision data obtained according to the VR product migration interest of the third VR product at this stage; the product attribute knowledge point of the third VR product is consistent with the product attribute knowledge point of the first VR product and the interaction mode label corresponding to the third VR product matches the interaction mode label corresponding to the first VR product; the presentation mode of the third VR product is the presentation activation mode; Obtaining an interest prediction network generated by AI learning using the first sample data and the first training supervision data, loading the VR interaction migration big data of the first VR product in this stage into the interest prediction network, and generating a predicted VR product migration interest of the first VR product when it is switched to the presentation activation mode; the predicted VR product migration interest is used to determine a predicted VR interaction parameter of the first VR product when it is switched to the presentation activation mode; The first sample data sequence is generated based on the VR interaction migration big data of the first VR product in this stage, including: Determine the VR interaction migration data range based on the VR interaction migration big data of the first VR product in this stage; Acquire multiple VR commodity coding features; one VR commodity coding feature is used to represent one VR commodity; the presentation modes corresponding to the VR commodities represented by the multiple VR commodity coding features are all the presentation activation modes; the commodity attribute knowledge points corresponding to the multiple VR commodities are all the same as the commodity attribute knowledge points of the first VR commodity; Among the multiple VR commodity coding features, VR commodity coding features representing X VR commodities are obtained; the interaction mode labels corresponding to the X VR commodities respectively match the interaction mode label corresponding to the first VR commodity; the X VR commodities include the third VR commodity; X is a positive integer, and X is not greater than the global number of the multiple VR commodity coding features; Generate a first sample data sequence based on the VR interaction migration big data of this stage corresponding to the X VR commodity coding features and the VR interaction migration data range; The method further comprises: If the interaction mode label of the VR mall interaction event does not meet the preset audience label feature, a second sample data sequence is generated based on the interaction mode label of the VR mall interaction event; the second sample data sequence includes second sample data obtained by using the VR interaction migration big data of the fourth VR product at this stage, and second training supervision data obtained according to the VR interaction parameters of the fourth VR product at this stage; the commodity attribute knowledge points of the fourth VR product are consistent with the commodity attribute knowledge points of the first VR product and the interaction mode label corresponding to the fourth VR product matches the interaction mode label corresponding to the first VR product; the presentation mode of the fourth VR product is the presentation activation mode; Obtaining a second basic machine learning network, loading the second sample data into the second basic machine learning network, and using the second basic machine learning network to determine a VR interaction parameter prediction result corresponding to the second sample data; Determine a second loss function value between the second training supervision data and the VR interaction parameter prediction result, train the second basic machine learning network based on the second loss function value, generate a VR interaction parameter prediction network, load the VR interaction migration big data of the first VR product in this stage into the VR interaction parameter prediction network, and generate predicted VR interaction parameters of the first VR product when it is converted to the presentation activation mode.

2. The big data mining method applied to VR shopping malls according to claim 1 is characterized in that: The X VR commodity coding features include VR commodity coding feature Zi; The X VR interaction migration big data of this stage include the VR interaction migration big data Yi of this stage corresponding to the VR commodity coding feature Zi; i is a positive integer, and i is not greater than the global number of the X VR commodity coding features; The first sample data sequence is generated based on the VR interactive migration big data of this stage corresponding to the X VR commodity coding features and the VR interactive migration data range, including: Matching the VR interaction migration big data Yi of this stage with the VR interaction migration data range; If the VR interactive migration big data Yi of this stage is within the VR interactive migration data range, then the VR commodity coding feature Zi is loaded into the relevant VR commodity coding feature set; Determine a first global number of VR commodity coding features in the relevant VR commodity coding feature set, compare the first global number with a first threshold value, and generate a comparison result; the first threshold value is determined based on a loss between a priori predicted VR interaction parameters of a first priori VR commodity within a first priori time period and a priori actual VR interaction parameters of the first priori VR commodity; the first priori VR commodity belongs to the X VR commodities; A first sample data sequence is generated based on the comparison result, the VR commodity coding features in the relevant VR commodity coding feature set, and the VR interactive migration data range.

3. The big data mining method applied to VR shopping malls according to claim 2 is characterized in that: The generating a first sample data sequence based on the comparison result, the VR commodity coding feature in the relevant VR commodity coding feature set, and the VR interactive migration data range includes: If the comparison result is that the first global quantity is not less than the first threshold value, generating a first sample data sequence based on the VR commodity coding features in the relevant VR commodity coding feature set; If the comparison result is that the first global number is less than the first threshold value, the VR interaction migration data range is optimized to generate an optimized VR interaction migration data range; the optimized VR interaction migration data range is larger than the VR interaction migration data range; Based on the VR interaction migration big data of this stage and the optimized VR interaction migration data range respectively included in the X VR commodity coding features, a first sample data sequence is generated.

4. The big data mining method applied to VR shopping malls according to claim 3 is characterized in that: Generating a first sample data sequence based on the VR commodity coding features in the relevant VR commodity coding feature set includes: Outputting the VR commodity represented by the VR commodity coding feature in the relevant VR commodity coding feature set as the third VR commodity; Obtaining VR interaction migration big data of the third VR product at this stage, and VR product migration interest of the third VR product at this stage; The VR interaction migration big data of the third VR product in this stage is used as the first sample data, and the VR product migration interest degree of the third VR product in this stage is used as the first training supervision data to generate a first sample data sequence including the first sample data and the first training supervision data.

5. The big data mining method applied to VR shopping malls according to claim 1 is characterized in that: The obtaining of the interest prediction network generated by AI learning using the first sample data and the first training supervision data includes: Obtaining a first basic machine learning network, and loading the first sample data into the first basic machine learning network; Determine, using the first basic machine learning network, a prediction result of a point of interest corresponding to the first sample data; Determine a first loss function value between the first training supervision data and the interest point prediction result, train the first basic machine learning network based on the first loss function value, and generate an interest prediction network.

6. The big data mining method applied to VR shopping malls according to claim 1 is characterized in that: The method further comprises: Determining predicted VR interaction parameters of the first VR product when switching to a presentation activation mode based on the predicted VR product migration interest and the VR interaction migration big data of the first VR product at this stage; Based on the predicted VR interaction parameters of the first VR product and the presented VR interaction parameters of the first VR product, a predicted interaction improvement rate corresponding to the first VR product is determined; the predicted VR interaction parameters and the predicted interaction improvement rate are both displayed in a product information display page for the first VR product in a VR device.

7. The big data mining method applied to VR shopping malls according to claim 1 is characterized in that: The generating of the second sample data sequence based on the interaction mode label of the VR shopping mall interaction event includes: T VR commodity coding features are obtained; one VR commodity coding feature is used to characterize one VR commodity; the presentation modes corresponding to the T VR commodities are all the presentation activation modes; the commodity attribute knowledge points corresponding to the T VR commodities are all the same as the commodity attribute knowledge points of the first VR commodity; the interaction mode labels corresponding to the T VR commodities do not include the interaction mode label corresponding to the preset audience label feature; Among the T VR commodity coding features, K VR commodity coding features are obtained; the interaction mode labels respectively included in the K VR commodity coding features are consistent with the interaction mode label corresponding to the first VR commodity and the commodity experience priorities respectively included in the K VR commodity coding features are the same as the commodity experience priorities corresponding to the first VR commodity; K is a positive integer, and K is not greater than the global number of multiple VR commodity coding features; Taking K as the second global quantity, if the second global quantity is not less than a second threshold value, outputting the VR commodities represented by the K VR commodity coding features as the fourth VR commodity; the second threshold value is determined based on the loss between the a priori predicted VR interaction parameter of the second a priori VR commodity within a second a priori time period and the a priori actual VR interaction parameter of the second a priori VR commodity; the second a priori VR commodity belongs to the T VR commodities; Alternatively, if the second global number is less than the second threshold value, then among the T VR product coding features, U VR product coding features are obtained, and U is used as the third global number. If the third global number is not less than the second threshold value, the VR products represented by the U VR product coding features are output as the fourth VR product. The product experience priorities respectively included in the U VR product coding features are the same as the product experience priorities corresponding to the first VR product. U is a positive integer not less than U, and U is not greater than the global number of multiple VR product coding features. Alternatively, if the third global number is less than the second threshold value, a product experience priority range is generated based on the product experience priority corresponding to the first VR product and the peripheral product experience priority of the product experience priority corresponding to the first VR product; among the T VR product coding features, L VR product coding features are obtained; the product experience priorities respectively included in the L VR product coding features all belong to the product experience priority range; L is a positive integer not less than U, and L is not greater than the global number of the multiple VR product coding features; L is taken as the fourth global number, and if the fourth global number is not less than the second threshold value, the VR products respectively represented by the L VR product coding features are output as the fourth VR product; The VR interaction migration big data of the fourth VR product at this stage is used as the second sample data, and the VR interaction parameters of the fourth VR product at this stage are used as the second training supervision data to generate a first sample data sequence including the second sample data and the second training supervision data; If the fourth global number is less than the second threshold value, the product experience priority range is optimized to generate an optimized product experience priority range; the optimized product experience priority range is greater than the product experience priority range; Among the T VR product coding features, J VR product coding features are obtained; the product experience priorities respectively included in the J VR product coding features all belong to the optimized product experience priority range; J is a positive integer not less than L, and J is not greater than the global number of the multiple VR product coding features; Based on the J VR commodity coding features, a second sample data sequence is generated.

8. A VR shopping mall platform system, characterized in that: The VR shopping mall platform system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the big data mining method applied to the VR shopping mall as described in any one of claims 1 to 7 above.