Product form generation method based on deep learning and adaptive modeling

Abstract

The invention discloses a product form generation method based on deep learning and adaptive modeling, and relates to the technical field of data processing by analyzing and evaluating collected data, performing evaluation calculation according to an evaluation result and extracting key features. According to the product form generation method based on deep learning and adaptive modeling, acquired data are analyzed and evaluated, evaluation calculation is performed according to an evaluation result, key features are extracted, and parameters of a generative adversarial network are adjusted based on the key features by using an adaptive modeling technology, so that the product form generation efficiency is improved. Diversified initial schemes are generated based on the generative adversarial network, the initial schemes form a digital product to be transmitted, dynamic factor analysis is realized according to market data, target market audiences are accurately identified, the advertisement putting efficiency is improved, and the advertisement putting efficiency is improved through the generative adversarial network. An advertisement putting strategy is dynamically adjusted according to real-time market feedback and user behavior data, and precise marketing is achieved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a product morphology generation method based on deep learning and adaptive modeling. Background Art

[0002] In the process of cross-border independent station product development (multi-language websites, adaptive websites, cross-border foreign trade malls), products that are adjusted to suit the region can better open up foreign markets.

[0003] The reference patent name is: A product form generative design method based on deep learning and parametric modeling (patent publication number: CN115270288A, patent publication date: 2022-11-01), which solves the technical problem of rapid generation of three-dimensional models of complex products based on user intentions. The key points of its technical solution are to integrate professional field knowledge in design into the generative design algorithm, explore the user's style and image perception of product styling elements, and use the degree of conformity between styling and image as a design guide to help designers accurately grasp design features and user needs, reduce design risks, and at the same time, establish a generative product design system through deep learning and parametric modeling technology to assist designers in quickly generating design solutions in the conceptual design stage.

[0004] Based on the statements in the above documents, existing overseas product development work is easily hindered by culture and language, and the process of product form generation fails to cover various factors for excellent existing digital products, resulting in deviations when extracting features or affecting the effectiveness of new digital products in actual applications. To this end, the present invention provides a product form generation method based on deep learning and adaptive modeling. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a product form generation method based on deep learning and adaptive modeling, which solves the problem that the existing overseas product expansion work is easily hindered by culture and language, and the product form generation process fails to cover various factors for excellent existing digital products, resulting in deviations in feature extraction or affecting the effectiveness of new digital products in actual applications.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a product morphology generation method based on deep learning and adaptive modeling, specifically comprising the following steps:

[0007] A1. Collect target market user data from multiple channels such as social media, search engines, and e-commerce platforms, cleanse and pre-process the collected data, and transmit and store the processed data via wireless communication technology.

[0008] A2. Analyze and evaluate the collected data, perform evaluation calculations based on the evaluation results, and extract key features. Based on these key features and using adaptive modeling techniques, adjust the parameters of the generative adversarial network. Generate diverse initial solutions based on the generative adversarial network, and transform these initial solutions into digital products for transmission.

[0009] A3. Establish a real-time feedback mechanism to collect advertising effectiveness data and adaptively update and optimize the model based on the feedback data.

[0010] Preferably, the operation of cleaning the collected data in A1 is:

[0011] Data cleaning operations are used to process missing values, outliers, and duplicate values in the data; data denoising operations are used to eliminate random errors or irrelevant signals in the data; and data normalization operations are used to scale the data to a uniform range.

[0012] Natural language processing technology is used to translate the collected copy data, and computer vision technology is used to identify and replace cultural elements in image and video data to form copy content that is easy to identify.

[0013] Preferably, the operation of preprocessing the collected data in A1 is:

[0014] a11. The collected data includes image data and parameter data, and is preliminarily classified according to image category and parameter category;

[0015] a12. Perform an extraction operation based on the classified image data and parameter data. Set the extraction window, using the data type required for subsequent evaluation processing. Use the type of different data sources as the first search tag, and the title of the required data type as the second search tag. The content of the second search tag is input after the search is completed based on the first search tag. The data after the two searches are completed is extracted and used for evaluation.

[0016] a13. And form a data table according to the category of the search tag for subsequent data tracing.

[0017] Preferably, the wireless communication technology in A1 is specifically:

[0018] Perform spectrum analysis on external signal data through fast Fourier transform, and convert the received time-domain air interface signal into a frequency-domain signal sequence;

[0019] Combining the existing operating frequency of the communication equipment and the determined interference frequency, a suitable undisturbed frequency is selected as the transmission frequency of the wireless communication technology, and the collected data is transmitted through the 5G network based on the transmission frequency.

[0020] Preferably, the operation of analyzing and evaluating the collected data in A2 is:

[0021] B1. Extract product type labels based on the product direction you need to form, and extract relevant collected data based on the product type labels;

[0022] B2. Determine the evaluation value based on the evaluation, conversion, and browsing status of the collected data, and then determine the existing digital products with better performance by sorting the evaluation values.

[0023] B3. Select the G best digital products from the previous section and conduct analysis.

[0024] Preferably, the operation of deriving the evaluation value in B2 by combining the evaluation situation, conversion situation and browsing situation in the collected data is:

[0025] b21. The specific number of evaluations in each digital product data related to the product type label is marked as M. a , and set the evaluation quantity threshold to P1, which will satisfy M a Extract digital product data ≥P1 and determine the number of positive reviews in the corresponding digital product data as N a , determine the number of positive reviews without text comments and automatic reviews is L a , and the praise rate [(N a -L a ) / (M a -L a )];

[0026] b22. Based on the digital product data determined in step b21, determine the total number of views in the digital product data as R b , and set the browsing duration evaluation threshold as P2, and mark the number of browsing times in the total browsing time that is greater than the browsing duration evaluation threshold P2 as R c , then the actual number of views is (R b -R c );

[0027] b23, and extract the consultation quantity from the confirmed digital product data and mark it as S c , and the conversion rate of page views is , and the number of completed transformations is T c , then the consultation conversion rate is (T c / S c );

[0028] b24. Assign weights to the praise rate, pageview conversion rate, and consultation conversion rate to derive evaluation values, and determine the existing better digital products based on the ranking of the evaluation values.

[0029] Preferably, the expression for obtaining the evaluation value by weight assignment in b24 is:

[0030] ;

[0031] F is the evaluation value, α is the weighted value of the positive rate, β is the weighted value of the pageview conversion rate, γ is the weighted value of the consultation conversion rate, γ>α>β, and α+β+γ=1.

[0032] Preferably, the operation of extracting key features based on the evaluation results in A2 is:

[0033] C1. Extract the G better digitized products and extract the image data and text data on the digitized products;

[0034] C2. Using a matching function to identify the text data in each digital product, the matching function sets an extended window to traverse the text data to obtain vocabulary tags, extracts the vocabulary tags from each digital product for matching and comparison, and sorts them according to the number of times the vocabulary tags appear to determine common features of the text data;

[0035] C3. Extract the color and layout of the image data to determine the common features of the image data, and then combine the common features of the text data and the common features of the image data to form key features.

[0036] Preferably, the operation of generating diversified initial solutions based on the generative adversarial network in A2 extracts key features and then introduces the generative adversarial network for training and optimization to form a digital product with key features, and transmits the digital product to the target market for display.

[0037] Preferably, the real-time feedback mechanism in A3 is:

[0038] D1. After the digital product is displayed in the target market, data on the effectiveness of the digital product launch is extracted within the cycle time.

[0039] D2. Then, the same operations as steps b21 to b23 are used to identify the delivery effect data. If the final data obtained is lower than the expected data set before delivery, the optimization data is fed back to update the digital product.

[0040] This invention provides a product morphology generation method based on deep learning and adaptive modeling. Compared with the existing technology, it has the following advantages:

[0041] 1. This product form generation method based on deep learning and adaptive modeling analyzes and evaluates the collected data, performs evaluation calculations and extracts key features based on the evaluation results, adjusts the parameters of the generative adversarial network based on the key features and uses adaptive modeling technology, generates diversified initial solutions based on the generative adversarial network, and constructs the initial solutions into digital products for transmission. It realizes dynamic factor analysis based on market data, accurately identifies the target market audience, improves advertising delivery efficiency, and dynamically adjusts advertising delivery strategies based on real-time market feedback and user behavior data through the generative adversarial network to achieve precision marketing.

[0042] 2. This product form generation method based on deep learning and adaptive modeling collects target market user data through multiple channels, cleans and pre-processes the collected data, and transmits and stores the processed data via wireless communication technology. It uses natural language processing and computer vision technology to perform cross-cultural adaptation of advertising content, overcome cultural differences and language barriers, and thus facilitate the accurate conversion of collected data into digital products.

[0043] 3. This product form generation method based on deep learning and adaptive modeling extracts data on the effectiveness of digital product launches within a certain period of time after the digital product is displayed in the target market. If the final data obtained is lower than the expected data set before launch, the optimization data is fed back to update the digital product. Through real-time strategy optimization, the advertising launch strategy is dynamically adjusted according to real-time market feedback and user behavior data to achieve precision marketing. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is an operational flow chart of the product morphology generation method of the present invention;

[0045] Figure 2 This is an operational flow chart of data preprocessing of the present invention;

[0046] Figure 3 An operational flow chart for data analysis and evaluation of the present invention;

[0047] Figure 4 The flowchart of the operation of extracting key features of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] See also Figure 1-Figure 4 , the present invention provides two technical solutions:

[0050] Example 1: A method for generating product morphology based on deep learning and adaptive modeling, specifically comprising the following steps:

[0051] A1. Collect target market user data from multiple channels such as social media, search engines, and e-commerce platforms, cleanse and pre-process the collected data, and transmit and store the processed data via wireless communication technology.

[0052] A2. Analyze and evaluate the collected data, perform evaluation calculations based on the evaluation results, and extract key features. Based on these key features and using adaptive modeling techniques, adjust the parameters of the generative adversarial network. Generate diverse initial solutions based on the generative adversarial network, and transform these initial solutions into digital products for transmission.

[0053] A3. Establish a real-time feedback mechanism to collect advertising effectiveness data and adaptively update and optimize the model based on the feedback data.

[0054] Generative adversarial networks are a mature technology. They consist of a generative model and a discriminative model. The generative model is responsible for capturing the distribution of sample data, while the discriminative model is generally a binary classifier that determines whether the input is real data or a generated sample. The optimization process of this model is a "binary minimax game" problem. During training, one of the parties (the discriminative network or the generative network) is fixed, and the parameters of the other model are updated, and the iterations are repeated. Ultimately, the generative model can estimate the distribution of sample data. The emergence of generative adversarial networks has greatly promoted the research on unsupervised learning and image generation.

[0055] By analyzing and evaluating the collected data, evaluating and calculating and extracting key features based on the evaluation results, adjusting the parameters of the generative adversarial network based on the key features and using adaptive modeling technology, generating diversified initial solutions based on the generative adversarial network, and constructing the initial solutions into digital products for transmission, dynamic factor analysis is achieved based on market data, accurately identifying the target market audience, improving the efficiency of advertising delivery, and through the generative adversarial network, dynamically adjusting the advertising delivery strategy based on real-time market feedback and user behavior data to achieve precision marketing.

[0056] In the embodiment of the present invention, the operation of cleaning the collected data in A1 is:

[0057] Data cleaning operations are used to process missing values, outliers, and duplicate values in the data; data denoising operations are used to eliminate random errors or irrelevant signals in the data; and data normalization operations are used to scale the data to a uniform range.

[0058] Natural language processing technology is used to translate the collected copy data, and computer vision technology is used to identify and replace cultural elements in image and video data to form copy content that is easy to identify.

[0059] In the embodiment of the present invention, the operation of pre-processing the collected data in A1 is:

[0060] a11. The collected data includes image data and parameter data, and is preliminarily classified according to image category and parameter category;

[0061] a12. Perform an extraction operation based on the classified image data and parameter data. Set the extraction window, using the data type required for subsequent evaluation processing. Use the type of different data sources as the first search tag, and the title of the required data type as the second search tag. The content of the second search tag is input after the search is completed based on the first search tag. The data after the two searches are completed is extracted and used for evaluation.

[0062] a13. And form a data table according to the category of the search tag for subsequent data tracing.

[0063] User data from the target market is collected through multiple channels, and the collected data is cleaned and pre-processed. The processed data is transmitted and stored via wireless communication technology. Natural language processing and computer vision technology are used to adapt advertising content across cultures, overcoming cultural differences and language barriers, thereby facilitating the accurate conversion of collected data and digital products.

[0064] In the embodiment of the present invention, the wireless communication technology in A1 is specifically:

[0065] Perform spectrum analysis on external signal data through fast Fourier transform, and convert the received time-domain air interface signal into a frequency-domain signal sequence;

[0066] Combining the existing operating frequency of the communication equipment and the determined interference frequency, a suitable undisturbed frequency is selected as the transmission frequency of the wireless communication technology, and the collected data is transmitted through the 5G network based on the transmission frequency.

[0067] In the embodiment of the present invention, the operation of analyzing and evaluating the collected data in A2 is:

[0068] B1. Extract product type labels based on the product direction you need to form, and extract relevant collected data based on the product type labels;

[0069] B2. Determine the evaluation value based on the evaluation, conversion, and browsing status of the collected data, and then determine the existing digital products with better performance by sorting the evaluation values.

[0070] B3. Select the G best digital products from the previous section and conduct analysis.

[0071] In the embodiment of the present invention, the operation of deriving the evaluation value in B2 by combining the evaluation status, conversion status, and browsing status in the collected data is:

[0072] b21. The specific number of evaluations in each digital product data related to the product type label is marked as M. a , and set the evaluation quantity threshold to P1, which will satisfy M a Extract digital product data ≥P1 and determine the number of positive reviews in the corresponding digital product data as N a , determine the number of positive reviews without text comments and automatic reviews is L a , and obtain the praise rate ;

[0073] b22. Based on the digital product data determined in step b21, determine the total number of views in the digital product data as R b , and set the browsing duration evaluation threshold as P2, and mark the number of browsing times in the total browsing time that is greater than the browsing duration evaluation threshold P2 as R c , then the actual number of views is;

[0074] b23, and extract the consultation quantity from the confirmed digital product data and mark it as S c , and the conversion rate of page views is , and the number of completed transformations is T c , then the consultation conversion rate is (T c / S c );

[0075] b24. Assign weights to the praise rate, pageview conversion rate, and consultation conversion rate to derive evaluation values, and determine the existing better digital products based on the ranking of the evaluation values.

[0076] In the embodiment of the present invention, the expression for weight assignment in b24 to obtain the evaluation value is:

[0077] ;

[0078] F is the evaluation value, α is the weighted value of the positive rate, β is the weighted value of the pageview conversion rate, γ is the weighted value of the consultation conversion rate, γ>α>β, and α+β+γ=1.

[0079] In the embodiment of the present invention, the operation of extracting key features based on the evaluation results in A2 is:

[0080] C1. Extract the G better digitized products and extract the image data and text data on the digitized products;

[0081] C2. Using a matching function to identify the text data in each digital product, the matching function sets an extended window to traverse the text data to obtain vocabulary tags, extracts the vocabulary tags from each digital product for matching and comparison, and sorts them according to the number of times the vocabulary tags appear to determine common features of the text data;

[0082] C3. Extract the color and layout of the image data to determine the common features of the image data, and then combine the common features of the text data and the common features of the image data to form key features.

[0083] In an embodiment of the present invention, the operation of generating diversified initial solutions based on the generative adversarial network in A2 extracts key features and then introduces the generative adversarial network for training and optimization to form a digital product with key features, and then transmits the digital product to the target market for display.

[0084] In the embodiment of the present invention, the real-time feedback mechanism in A3 is:

[0085] D1. After the digital product is displayed in the target market, data on the effectiveness of the digital product launch is extracted within the cycle time.

[0086] D2. Then, the same operations as steps b21 to b23 are used to identify the delivery effect data. If the final data obtained is lower than the expected data set before delivery, the optimization data is fed back to update the digital product.

[0087] After digital products are displayed in the target market, data on the effectiveness of digital product launches is extracted within the launch cycle. If the final data obtained is lower than the expected data set before launch, the optimization data is fed back to update the digital products. Through real-time strategy optimization, the advertising launch strategy is dynamically adjusted according to real-time market feedback and user behavior data to achieve precision marketing.

[0088] In the process of realizing digital product generation, personalized content recommendations are combined. The characteristics and behavioral data of the target market audience are analyzed through deep learning models to build user portraits. Based on the user portraits, personalized recommendation algorithms are used, such as existing collaborative filtering and deep learning recommendation models, to generate personalized advertising content recommendations. The personalized advertising content recommendations are then displayed to the target market audience to improve the attractiveness and conversion rate of the advertising content.

[0089] The difference between Example 2 and Example 1 is that the same data is extracted and analyzed by the existing product form generation method and the product form generation method of the present invention, and the display operation of the tested digital product in the target market is formed, and various data within the cycle time are recorded and extracted. The specific results are shown in Table 1:

[0090] Table 1 Data results table

[0091]

[0092] Experimental results show that the product morphology generation method of the present invention is significantly superior to the existing product morphology generation method in terms of the expression of various parameter data, and the number of views, consultations and transactions are much higher than the existing product morphology generation method, which proves its superiority and accuracy in practical applications.

[0093] Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0094] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0095] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A product morphology generation method based on deep learning and adaptive modeling, characterized by: The specific steps include: A1. Collect target market user data from multiple channels such as social media, search engines, and e-commerce platforms, cleanse and pre-process the collected data, and transmit and store the processed data via wireless communication technology. A2. Analyze and evaluate the collected data, perform evaluation calculations based on the evaluation results, and extract key features. Based on these key features and using adaptive modeling techniques, adjust the parameters of the generative adversarial network. Generate diverse initial solutions based on the generative adversarial network, and transform these initial solutions into digital products for transmission. A3. Establish a real-time feedback mechanism to collect advertising effectiveness data and adaptively update and optimize the model based on the feedback data.

2. The method for generating product morphology based on deep learning and adaptive modeling according to claim 1, characterized in that: The operations for cleaning the collected data in A1 are: Data cleaning operations are used to process missing values, outliers, and duplicate values in the data; data denoising operations are used to eliminate random errors or irrelevant signals in the data; and data normalization operations are used to scale the data to a uniform range. Natural language processing technology is used to translate the collected copy data, and computer vision technology is used to identify and replace cultural elements in image and video data to form copy content that is easy to identify.

3. The method for generating product morphology based on deep learning and adaptive modeling according to claim 1, characterized in that: The operations for preprocessing the collected data in A1 are: a11. The collected data includes image data and parameter data, and is preliminarily classified according to image category and parameter category; a12. Perform an extraction operation based on the classified image data and parameter data. Set the extraction window, using the data type required for subsequent evaluation processing. Use the type of different data sources as the first search tag, and the title of the required data type as the second search tag. The content of the second search tag is input after the search is completed based on the first search tag. The data after the two searches are completed is extracted and used for evaluation. a13. And form a data table according to the category of the search tag for subsequent data tracing.

4. The method for generating product morphology based on deep learning and adaptive modeling according to claim 1, characterized in that: The wireless communication technology in A1 is specifically: Perform spectrum analysis on external signal data through fast Fourier transform, and convert the received time-domain air interface signal into a frequency-domain signal sequence; Combining the existing operating frequency of the communication equipment and the determined interference frequency, a suitable undisturbed frequency is selected as the transmission frequency of the wireless communication technology, and the collected data is transmitted through the 5G network based on the transmission frequency.

5. The method for generating product morphology based on deep learning and adaptive modeling according to claim 1, characterized in that: The operations of analyzing and evaluating the collected data in A2 are: B1. Extract product type labels based on the product direction you need to form, and extract relevant collected data based on the product type labels; B2. Determine the evaluation value based on the evaluation, conversion, and browsing status of the collected data, and then determine the existing digital products with better performance by sorting the evaluation values. B3. Select the G best digital products from the previous section and conduct analysis.

6. The method for generating product morphology based on deep learning and adaptive modeling according to claim 5, characterized in that: The operation of obtaining the evaluation value by combining the evaluation, conversion and browsing conditions in the collected data in B2 is: b21. The specific number of evaluations in each digital product data related to the product type label is marked as M. a , and set the evaluation quantity threshold to P1, which will satisfy M a Extract digital product data for products with values greater than or equal to P1 and determine the number of positive reviews in the corresponding digital product data as N. a , determine the number of positive reviews without text comments and automatic reviews is L a , and the praise rate [(N a -L a ) / (M a -L a )]; b22. Based on the digital product data determined in step b21, determine the total number of views in the digital product data as R b , and set the browsing duration evaluation threshold as P2, and mark the number of browsing times in the total browsing time that is greater than the browsing duration evaluation threshold P2 as R c , then the actual number of views is (R b -R c ); b23, and extract the consultation quantity from the confirmed digital product data and mark it as S c , and the conversion rate of page views is [S c / (R b -R c )], and the number of completed transformations is T c , then the consultation conversion rate is (T c / S c ); b24. Assign weights to the praise rate, pageview conversion rate, and consultation conversion rate to derive evaluation values, and determine the existing better digital products based on the ranking of the evaluation values.

7. The method for generating product morphology based on deep learning and adaptive modeling according to claim 6, characterized in that: The expression for weight assignment in b24 to obtain the evaluation value is: ； F is the evaluation value, α is the weighted value of the positive rate, β is the weighted value of the pageview conversion rate, γ is the weighted value of the consultation conversion rate, γ>α>β, and α+β+γ=1.

8. The method for generating product morphology based on deep learning and adaptive modeling according to claim 5, characterized in that: The operation of extracting key features based on the evaluation results in A2 is: C1. Extract the G better digitized products and extract the image data and text data on the digitized products; C2. Using a matching function to identify the text data in each digital product, the matching function sets an extended window to traverse the text data to obtain vocabulary tags, extracts the vocabulary tags from each digital product for matching and comparison, and sorts them according to the number of times the vocabulary tags appear to determine common features of the text data; C3. Extract the color and layout of the image data to determine the common features of the image data, and then combine the common features of the text data and the common features of the image data to form key features.

9. The method for generating product morphology based on deep learning and adaptive modeling according to claim 1, characterized in that: The operation of generating diversified initial solutions based on the generative adversarial network in A2 extracts key features and then introduces the generative adversarial network for training and optimization to form a digital product with key features, and then transmits the digital product to the target market for display.

10. The method for generating product form based on deep learning and adaptive modeling according to claim 6, characterized in that: The real-time feedback mechanism in A3 is: D1. After the digital product is displayed in the target market, data on the effectiveness of the digital product launch is extracted within the cycle time. D2. Then, the same operations as steps b21 to b23 are used to identify the delivery effect data. If the final data obtained is lower than the expected data set before delivery, the optimization data is fed back to update the digital product.

Images