Product Management Method and System Based on Smart E-commerce Platform

Abstract

The present invention relates to the technical field of e-commerce platforms, and particularly to a product management method and system based on an intelligent e-commerce platform. The method includes the following steps: obtaining user historical behavior data; compensating for the consumption desire factors of the user historical behavior data to generate user portrait data; obtaining product image data; extracting product image features from the product image data to generate product image feature data; performing hyperparameter image recognition on the product image data to generate product attribute data; and associating the product image feature data and the product attribute data to generate product feature vector data. The present invention realizes personalized product push for users by constructing user portraits and accurately identifying and associating product image features, so as to implement a product management method and system based on an intelligent e-commerce platform.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce platforms, and particularly to a product management method and system based on an intelligent e-commerce platform. Background Art

[0002] With the rapid development of the e-commerce industry, product images have become crucial in attracting customers, increasing sales, and enhancing the user experience. Traditional e-commerce platforms use static images to display products, but this approach has some limitations. In traditional product image management methods, the tagging and classification of product images usually need to be done manually, which may lead to misclassification and inconsistency, reducing search and filtering efficiency; it is impossible to accurately associate the historical behaviors and comments of users with product image features to achieve targeted product push. Summary of the Invention

[0003] Based on this, it is necessary to provide a product management method and system based on an intelligent e-commerce platform to solve at least one of the above technical problems.

[0004] To achieve the above object, a product management method based on an intelligent e-commerce platform includes the following steps:

[0005] Step S1: Obtain user historical behavior data; perform consumption factor compensation on the user historical behavior data to generate user portrait data;

[0006] Step S2: Obtain product image data; extract product image features from the product image data to generate product image feature data; perform hyperparameter image recognition on the product image data to generate product attribute data;

[0007] Step S3: Perform data association on the product image feature data and the product attribute data to generate product feature vector data;

[0008] Step S4: Calculate the similarity between the user portrait data and the product feature vector data to generate similarity data;

[0009] Step S5: Perform personalized recommendation based on the user historical behavior data and the similarity calculation data to generate product recommendation data.

[0010] The beneficial effects of this application are as follows. By analyzing the user's historical behavior data, it is possible to understand the user's preferences, likes, and behavior patterns, including purchase records, click behaviors, and browsing habits, which helps to reveal the user's interests and needs and provide personalized recommendations and services. By analyzing and mining the user's historical behavior data, a user profile can be constructed, that is, a description and feature summary of the user, which helps to better understand the user's needs and provide more accurate product recommendations and personalized experiences. By obtaining product image data, the description and data of the product can be enriched. The product image data can provide a more intuitive and vivid product display to help users better understand and evaluate the product. By extracting features from the product image data, a visual feature representation of the product can be obtained, which helps to extract the visual features of the product's texture, color, and shape for subsequent product comparison, similarity calculation, and recommendation. By associating the product image feature data with the product attribute data, the visual features and other attribute features of the product can be comprehensively considered, which helps to construct a more comprehensive and multi-dimensional product feature vector and provide the input data required for more accurate similarity calculation and recommendation algorithms. By calculating the similarity between the user profile data and the product feature vector data, the matching degree between the user and the product can be evaluated, which helps to find the user's potential interest areas and recommend products with high similarity to the user. Based on the user's historical behavior data and similarity calculation data, personalized product recommendations can be realized. By analyzing the user's behaviors and preferences and combining the similarity evaluation of the products, products that meet the user's interests and preferences can be recommended to them, improving user satisfaction and shopping experience. Therefore, the present invention provides a product management method and system based on an intelligent e-commerce platform, which realizes personalized push of product images to users by constructing user profiles and accurately identifying and associating product image features.

[0011] Preferably, step S1 includes the following steps:

[0012] Step S11: Perform time-frequency conversion on the user historical behavior data to generate frequency-domain representation data;

[0013] Step S12: Draw a spectrogram for the frequency-domain representation data to generate spectrogram data;

[0014] Step S13: Perform peak statistics on the spectrogram data to generate periodic component data;

[0015] Step S14: Judge the periodic trend of the periodic component data to generate user habit data;

[0016] Step S15: Perform multi-factor consumption mining on the user historical behavior data to generate data on the theme to be consumed;

[0017] Step S16: Construct a user portrait model based on the user habit data and the to-be-consumed theme data, thereby generating user portrait model data;

[0018] Step S17: Use the user portrait model data to perform model prediction on the user historical behavior data, thereby generating user portrait data.

[0019] Through time-frequency conversion, the present invention can convert the user historical behavior data into frequency-domain representation data, thereby extracting the characteristics of its frequency components, which helps to discover the periodic changes or important frequency components in the data and provides a basis for subsequent spectrum analysis and periodic detection; by plotting the spectrogram, the frequency-domain representation data can be intuitively visualized as spectrogram data, and the spectrogram shows the energy distribution of different frequency components, helping us observe and analyze the frequency-domain characteristics in the data; by performing peak statistics on the spectrogram data, the periodic components corresponding to the frequency peaks can be identified, which helps to extract the periodic change components in the data and provides a basis for subsequent periodic trend judgment; by performing periodic trend judgment, it can be analyzed whether the periodic component data shows an obvious periodic change trend, which helps to identify and discover the periodic behavior habits of users; by integrating the user habit data and the to-be-consumed theme data, a user portrait model can be constructed to help the system better understand the user and provide personalized product recommendations and services; based on the constructed user portrait model, predictions and inferences can be made on the user historical behavior data, which helps to predict the future behavior and interest evolution of users.

[0020] Preferably, step S15 includes the following steps:

[0021] Step S151: Decompose the fluctuation period of the user historical behavior data, thereby generating fluctuation period decomposition data;

[0022] Step S152: Identify the periodic behavior changes in the fluctuation period decomposition data, thereby generating user behavior change data;

[0023] Step S153: Extract the product change loss parameters from the user behavior change data, thereby generating product change loss parameters;

[0024] Step S154: Analyze the regional latitude changes in the user historical behavior data, thereby generating the local climate environment data;

[0025] Step S155: Screen the product themes according to the local climate environment data, thereby generating product theme data;

[0026] Step S156: Use the product change loss parameters to correct the product themes in the product theme data, thereby generating the user's target product;

[0027] Step S157: Speculate the user's mood based on the local climate environment data and accumulate the consumption desire scores to generate a consumption desire scoring factor.

[0028] Step S158: Use the consumption desire scoring factor to compensate the user's target product to generate the data of the consumption theme to be desired.

[0029] Through the decomposition of the fluctuation cycle of the user's historical behavior data, the present invention can reveal potential repetitive patterns and periodic behaviors, which helps to understand the periodic changes of the user's behavior and provides a basis for subsequent analysis; by analyzing and identifying the decomposed data of the fluctuation cycle, the changing trends of the user's behavior can be captured, and understanding these behavior changes can help to understand the user's preferences; by extracting the product change and loss parameters, the impact degree of the user's behavior change on the product can be quantified, and these parameters can help to optimize the product characteristics, reduce the user's dissatisfaction with the product, and better meet the user's needs; the analysis of the regional latitude change can understand the behavior patterns and preference differences of users in different regions. According to the local climate environment data, the user's consumption behavior can be associated with the geographical environment, providing useful information for customized recommendations and market positioning; based on the local climate environment data for product theme screening, customized product recommendations can be made according to the differences in the user's environment, which helps to improve the relevance of the product and the user's satisfaction, and increase the user's interest and willingness to purchase the product; by using the product change and loss parameters, the product theme data can be corrected to adapt to the changes in the user's behavior and preferences, which helps to generate products that meet the user's goals and needs, and improve the user's acceptance and satisfaction of the product; by associating the mood speculation with the consumption desire, a consumption desire scoring factor can be generated, so as to understand the changes in the user's tendency and purchase willingness, providing guidance for personalized recommendations and marketing; using the consumption desire scoring factor to compensate the user's target product can adjust the product recommendation according to the user's desire and tendency.

[0030] Preferably, step S2 includes the following steps:

[0031] Step S21: Extract the picture quality conditions from the product image data to generate the quality condition data.

[0032] Step S22: Use the quality condition data to perform adaptive image enhancement on the product image data to generate the image enhancement data.

[0033] Step S23: Perform weighted pooling calculation on the image enhancement data to generate the product image feature data.

[0034] Step S24: Perform image marking on the image enhancement data to generate the marked product image data.

[0035] Step S25: Analyze the attribute labels of the marked product image data to generate attribute label analysis data;

[0036] Step S26: Perform attribute encoding on the attribute label analysis data to generate image attribute label data;

[0037] Step S27: Perform correlation fixing on the product image feature data to generate feature fixed-point scaling data;

[0038] Step S28: Construct an image recognition model from the feature fixed-point scaling data and the image attribute label data to generate image recognition model data;

[0039] Step S29: Use the image recognition model data to perform hyperparameter probability judgment on the product image feature data to generate product attribute data.

[0040] The present invention can evaluate the quality of a product image through picture quality condition extraction, including factors such as image clarity, exposure, and noise; adaptive image enhancement based on quality condition data can improve the visual effect of the image, enhance the clarity of the image, and contribute to enhancing the visual attractiveness, recognizability, and visual quality of the product image; by performing weighted pooling calculation on the image enhancement data, key product image features can be extracted, and these features can capture visual attributes such as the edges, textures, and shapes of the image. The extracted product image feature data can be used for subsequent image classification, similarity matching, retrieval, and recognition tasks; marking the product image data helps establish the connection between the picture and the product data, improving the searchability, understandability, and usability of the image; analyzing the attribute labels of the marked product image data can identify and extract key product attributes, which helps understand data on the features, specifications, functions, and uses of the product; through attribute encoding, the attribute data can be converted into a digital coding form, facilitating subsequent machine learning and data analysis tasks; the feature fixed-point scaling data can be used to optimize the distribution of the feature space and improve the performance of subsequent machine learning algorithms; the image recognition model data can be used for tasks such as automatic product classification, recognition, and detection.

[0041] Preferably, the adaptive image enhancement process is performed through an adaptive image enhancement calculation formula, where the adaptive image enhancement calculation formula is specifically:

[0042]

[0043] Wherein, I(x, y) represents the pixel value of the enhanced image, (x, y) represents the pixel coordinates on the image, N represents the number of samples of the quality condition data, i represents the index value of the quality condition data, ω(i) represents the weight factor of the i-th quality condition data, n represents a number approaching infinity, f(i, X, Y) represents the change rate of the i-th quality condition data at the position (x, y), g(i, x, y) represents the gradient mean value of the i-th quality condition data, θ(i, x, y) represents the function angle value of the i-th quality condition data, ρ(i, x, y) represents the adaptive factor of the i-th quality condition data at (x, y), and α represents the adaptive image enhancement error correction amount.

[0044] The present invention constructs an adaptive image enhancement calculation formula for performing adaptive image enhancement on product image data; in the formula The number of samples N of the quality condition data is used as a normalization factor for calculating the average value. By taking the average value processing, the influence of the sample data on the result can be reduced, ensuring the integrity and consistency of the enhanced image; ω(i) is used for weighted processing of different quality condition data. By adjusting the weight factor, the importance of different quality condition data can be adjusted, so that in the enhancement process, important quality condition data has a greater influence, thereby improving the adaptability and flexibility of the enhancement effect; The part is to perform normalization processing on the change rate and gradient mean value of the quality condition data. By calculating the logarithm of the ratio of the change rate and the gradient mean value and taking the limit, the significantly changed area can be enlarged, enhancing the contrast and detail data of the image, which helps to highlight the edge and texture features in the image; sin(θ(i, x, y)) can adjust the weight according to the feature changes in different directions by introducing angle data, thereby changing the brightness and contrast of the image; As a weight value, used to adjust the weight of the quality condition data at different pixel positions, the adaptive factor can be adaptively adjusted according to the feature changes in the local area to improve the adaptability and locality of image enhancement; α can finely adjust the enhanced image, correct possible errors and deviations, and further improve the accuracy and quality of image enhancement.

[0045] Preferably, step S23 includes the following steps:

[0046] Step S231: Scale the size of the model image of the image enhancement data to generate image input data;

[0047] Step S232: Use a preset pre-trained model to perform maximum convolution calculation, average convolution calculation, and minimum convolution calculation on the image input data to generate first initial convolution data, second initial convolution data, and third initial convolution data;

[0048] Step S233: Batch-normalize and activate the first initial convolution data, the second initial convolution data, and the third initial convolution data, thereby generating first convolution data, second convolution data, and third convolution data;

[0049] Step S234: Perform max pooling calculation on the first convolution data to generate first pooling data; perform average pooling calculation on the second convolution data to generate second pooling data; perform min pooling calculation on the third convolution data to generate third pooling data;

[0050] Step S235: Perform corresponding preset weight calculations on the first pooling data, the second pooling data, and the third pooling data respectively, thereby obtaining first confidence data, second confidence data, and third confidence data;

[0051] Step S236: Conduct result voting based on the first confidence data, the second confidence data, and the third confidence data, thereby generating feature representation confidence data;

[0052] Step S237: Generate product image feature data based on the feature representation confidence data.

[0053] In the present invention, by adjusting images of different sizes to the same size, it is ensured that the model can adapt to various input images, enhancing the generalization ability of the model and reducing the computational amount and memory consumption; by finding the maximum value in the image, important features of edges and details can be highlighted, by calculating the average value of pixels in the image, the image can be smoothed and noise can be reduced, by finding the minimum value in the image, details of dark parts and shadows can be highlighted; by normalizing the data of each batch, the training process can be accelerated, problems of gradient disappearance and gradient explosion can be reduced, the stability and convergence speed of the network can be improved, and applying an activation function to the convolution data can introduce non-linear characteristics, enhancing the expression ability of the model and enabling it to learn more complex functions; by selecting the maximum value in the pooling window, the main features in the image can be retained, by calculating the average value of pixels in the pooling window, the size of the feature map can be reduced and the features can be smoothed, by selecting the minimum value in the pooling window, details of dark parts and shadows in the image can be retained; according to different weight calculation methods, different features can be adaptively selected and emphasized, improving the expression ability of the model for input data; by integrating the voting results of different confidence data, a more comprehensive and representative feature representation can be obtained, enhancing the description ability of the image content and reducing the misjudgment risk brought by individual data; according to the feature representation confidence data, the product image features most relevant to solving a specific task can be selected and extracted, reducing redundant and noisy data and improving the effectiveness of the features.

[0054] Preferably, step S27 includes the following steps:

[0055] Step S271: Standardize the feature data of the product image to generate standardized feature data;

[0056] Step S272: Calculate the balanced rank for the standardized feature data to generate balanced rank data;

[0057] Step S273: Calculate the rank correlation coefficient for the balanced rank data to generate rank correlation coefficient data;

[0058] Step S274: Conduct correlation verification based on the rank correlation coefficient data to generate correlation verification data;

[0059] Step S275: Judge the correlation strength of the correlation verification data to generate correlation feature data;

[0060] Step S276: Perform feature fixed-point scaling on the correlation feature data to generate feature fixed-point scaling data.

[0061] Through the present invention, by standardizing the feature data, the value ranges of different features can be unified to the same scale, avoiding the influence of value differences between different features on subsequent processing; by calculating the balanced rank data, the feature values can be sorted and transformed into a rank form, making the data have strong comparability and interpretability; the rank correlation coefficient can measure the correlation degree between different features, helping us understand the correlation and mutual influence degree between features; through correlation verification, we can determine which features actually have significant correlations, thus excluding irrelevant features or features that may introduce noise, improving the accuracy of modeling and prediction; by judging the correlation verification data, the strength of the correlation can be represented in numerical form, helping us distinguish strong correlation, weak correlation, and non-correlation feature relationships; by feature fixed-point scaling, the weights of features can be adjusted, making the features that have a greater impact on the result have higher importance, improving the performance of the model on training and test data, and enhancing the generalization ability and prediction accuracy of the model.

[0062] Preferably, step S28 includes the following steps:

[0063] Step S281: Divide the dataset of the feature fixed-point scaling data and the image attribute label data to generate training set data and test set data;

[0064] Step S282: Use the support vector machine technology to construct a model to generate first recognition model data;

[0065] Step S283: Conduct the first search for hyperparameters on the first recognition model data to generate hyperparameter configuration data;

[0066] Step S284: Dynamically adjust the validation set for the training set data according to the hyperparameter configuration data, so as to generate dynamically adjusted data;

[0067] Step S285: Perform a secondary search for hyperparameters on the first recognition model data using the dynamically adjusted data, so as to generate hyperparameter data;

[0068] Step S286: Optimize the hyperparameters of the first recognition model data using the hyperparameter data, so as to generate second recognition model data;

[0069] Step S287: Evaluate the second recognition model data using the test set data, so as to generate second model evaluation data;

[0070] Step S288: Optimize the second recognition model data using the second model evaluation data, so as to generate image recognition model data.

[0071] Through partitioning the dataset, the present invention can reduce the overfitting phenomenon of the model on the training set, verify the effect of the model on unseen test data, and improve the robustness and reliability of the model; the support vector machine model can provide an explanation for the classification decision, helping us understand the data distribution and decision boundary behind the model; the first search for hyperparameters helps us determine a suitable set of hyperparameters for the support vector machine model, and the secondary search for hyperparameters can search for the best combination in a limited hyperparameter space, reducing the complexity of parameter search and accelerating the model optimization process; by adjusting the partitioning method of the validation set according to the hyperparameter configuration data, the performance of the model under different hyperparameter combinations can be better evaluated, which helps to select the best hyperparameters; by using the dynamically adjusted data for the secondary search of hyperparameters, the performance of the model can be further optimized, improving the accuracy and generalization ability of the model; using the hyperparameter data to optimize the first recognition model can improve the performance of the model, making the model fit the data more accurately and improving the robustness of the model; evaluating the second recognition model using the test set data can obtain the accuracy of the model on unseen data, verifying the generalization ability and prediction ability of the model; through the second model evaluation data, the deficiencies of the model can be identified and corresponding optimizations and improvements can be made to improve the performance and prediction accuracy of the model. After model optimization, the generated image recognition model data can be used in actual application scenarios for accurate image recognition and classification tasks.

[0072] Preferably, step S29 includes the following steps:

[0073] Step S291: Classify the product image feature data using the image recognition model data, so as to generate feature classification data;

[0074] Step S292: Calculate the hyperparameter weights for the feature classification data to generate feature calculation data;

[0075] Step S293: Predict the feature probabilities for the feature calculation data to generate feature probability prediction data;

[0076] Step S294: Sort the probability gradients for the feature probability prediction data to generate probability gradient sorting data;

[0077] Step S295: Perform category attribute association based on the preset category attribute relationship data and the probability gradient sorting data to generate product attribute data.

[0078] Through the image recognition model of the present invention, the product image feature data can be classified, different features can be effectively classified, various features existing in the product image can be recognized, and the key features in the product image can be extracted, thus helping to understand and describe the attributes and characteristics of the product; by calculating the hyperparameter weights for the feature classification data, the importance and contribution degree of different features can be determined, which is helpful for further optimizing the feature selection and modeling process; using the feature calculation data to predict the feature probabilities can obtain the probabilities of each feature appearing in the product image, helping us understand the feature distribution of the product image; by sorting the probability gradients for the feature probability prediction data, the most representative features in each product image can be determined, providing a more accurate and effective feature description; through the preset category attribute relationship data and the probability gradient sorting data, the features of the product can be associated with their corresponding attributes, further enriching and improving the description and attribute data of the product.

[0079] Preferably, the present invention also provides a product management system based on an intelligent e-commerce platform for executing the above-mentioned product management method based on an intelligent e-commerce platform, including:

[0080] A user portrait construction module, configured to obtain user historical behavior data; construct a user portrait for the user historical behavior data to generate user portrait data;

[0081] A product image feature analysis module, configured to obtain product image data; extract product image features from the product image data to generate product image feature data; perform hyperparameter image recognition on the product image data to generate product attribute data;

[0082] A product feature vector generation module, configured to perform data association on the product image feature data and the product attribute data to generate product feature vector data;

[0083] A similarity calculation module, configured to calculate the similarity between the user portrait data and the product feature vector data to generate similarity data;

[0084] A personalized recommendation module, which is used to make personalized recommendations based on user historical behavior data and similarity calculation data, so as to generate product recommendation data.

[0085] The beneficial effects of this application are as follows: By analyzing user historical behavior data, it is possible to understand users' preferences, hobbies, and behavior patterns, including purchase records, click behaviors, browsing habits, etc., which helps to reveal users' interests and needs and provide personalized recommendations and services; By analyzing and mining user historical behavior data, a user portrait can be constructed, that is, a description and feature summary of users, which helps to better understand users' needs and provide more accurate product recommendations and personalized experiences for them; By obtaining product image data, the description and data of products can be enriched. Product image data can provide a more intuitive and vivid product display to help users better understand and evaluate products; By extracting features from product image data, a visual feature representation of the product can be obtained, which helps to extract visual features such as the texture, color, and shape of the product for subsequent product comparison, similarity calculation, and recommendation; Associating product image feature data with product attribute data can comprehensively consider the visual features and other attribute features of products, which helps to construct a more comprehensive and multi-dimensional product feature vector and provide the input data required for more accurate similarity calculation and recommendation algorithms; By calculating the similarity between user portrait data and product feature vector data, the matching degree between users and products can be evaluated, which helps to find potential interest areas of users and recommend products with high similarity to users; Based on users' historical behavior data and similarity calculation data, personalized product recommendations can be realized. By analyzing users' behaviors and preferences and combining the similarity evaluation of products, products that meet users' interests and preferences can be recommended to them, improving user satisfaction and shopping experience. Therefore, the present invention provides a product management method and system based on an intelligent e-commerce platform, which realizes personalized push of product images to users by constructing user portraits and accurately identifying and associating product image features. Description of the Drawings

[0086] Figure 1 It is a schematic flowchart of the steps of a product management method based on an intelligent e-commerce platform.

[0087] Figure 2 is Figure 1 A detailed implementation step flowchart of step S1 in

[0088] Figure 3 A detailed implementation step flowchart of generating feature fixed-point scaling data.

[0089] Figure 4 A detailed implementation step flowchart of generating product attribute data.

[0090] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0091] The technical method of the present invention patent will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0092] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0093] It should be understood that although the terms "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed related items.

[0094] To achieve the above object, please refer to Figures 1 to 4 , a product management method based on a smart e-commerce platform, comprising the following steps:

[0095] Step S1: Obtain user historical behavior data; perform compensation for the consumption desire factor on the user historical behavior data to generate user portrait data;

[0096] Step S2: Obtain product image data; perform product image feature extraction on the product image data to generate product image feature data; perform hyperparameter image recognition on the product image data to generate product attribute data;

[0097] Step S3: Perform data association on the product image feature data and the product attribute data to generate product feature vector data;

[0098] Step S4: Calculate the similarity between the user portrait data and the product feature vector data to generate similarity data;

[0099] Step S5: Perform personalized recommendation based on the user's historical behavior data and the similarity calculation data, so as to generate product recommendation data.

[0100] The beneficial effects of this application are as follows: By analyzing the user's historical behavior data, the user's preferences, hobbies, and behavior patterns can be understood, including purchase records, click behaviors, and browsing habits, which helps to reveal the user's interests and needs and provide personalized recommendations and services; By analyzing and mining the user's historical behavior data, a user portrait can be constructed, that is, a description and feature summary of the user, which helps to better understand the user's needs and provide more accurate product recommendations and personalized experiences for them; By obtaining product image data, the description and data of the product can be enriched. The product image data can provide a more intuitive and vivid product display to help users better understand and evaluate the product; By extracting the features of the product image data, the visual feature representation of the product can be obtained, which helps to extract the visual features of the texture, color, and shape of the product for subsequent product comparison, similarity calculation, and recommendation; Associating the product image feature data with the product attribute data can comprehensively consider the visual features and other attribute features of the product, which helps to construct a more comprehensive and multi-dimensional product feature vector and provide the input data required for a more accurate similarity calculation and recommendation algorithm; By calculating the similarity between the user portrait data and the product feature vector data, the matching degree between the user and the product can be evaluated, which helps to find the potential interest areas of the user and recommend products with high similarity to the user; Based on the user's historical behavior data and the similarity calculation data, personalized product recommendations can be realized. By analyzing the user's behaviors and preferences and combining the similarity evaluation of the products, products that meet the user's interests and preferences can be recommended to them, improving user satisfaction and shopping experience. Therefore, the present invention provides a product management method and system based on an intelligent e-commerce platform, which realizes personalized push of product images to users by constructing a user portrait and accurately identifying and associating product image features.

[0101] In the embodiment of the present invention, refer to Figure 1 As described, it is a schematic diagram of the step flow of the product management method based on the intelligent e-commerce platform of the present invention. In this example, the product management method based on the intelligent e-commerce platform includes the following steps:

[0102] Step S1: Obtain the user's historical behavior data; perform compensation for the consumption desire factor on the user's historical behavior data, so as to generate user portrait data;

[0103] In an embodiment of the present invention, crawler technology is used to collect historical behavior data of users; the historical behavior data of users is processed and analyzed, and machine learning algorithms are used to construct user portraits; during the construction of user portraits, characteristic data of users, such as gender, age, hobbies, and purchase preferences, are extracted to generate user portrait data, representing the personal characteristics and behavior preferences of users.

[0104] Step S2: Obtain product image data; extract product image feature data from the product image data, thereby generating product image feature data; perform hyperparameter image recognition on the product image data, thereby generating product attribute data;

[0105] In an embodiment of the present invention, product image data is obtained through product inventory; the product image data is processed and analyzed, and computer vision technology is used to extract the visual features of the product; hyperparameter image recognition is performed to identify the attribute data of the product, generating product attribute data, representing the visual features and attribute data of the product.

[0106] Step S3: Correlate the product image feature data and the product attribute data, thereby generating product feature vector data;

[0107] In an embodiment of the present invention, the product image feature data and the product attribute data are correlated and matched according to the unique identifier of the product; the product image features and attribute data are integrated to construct the feature vector data of the product, and the feature vector contains the visual features and attribute data of the product, which is used to represent the feature representation of the product.

[0108] Step S4: Calculate the similarity between the user portrait data and the product feature vector data, thereby generating similarity data;

[0109] In an embodiment of the present invention, the cosine similarity is used to calculate the similarity between the user portrait data and the product feature vector data; according to the similarity calculation result, similarity data is generated, representing the similarity relationship between the user and the product, and a higher similarity indicates that the user has a higher preference for the product.

[0110] Step S5: Perform personalized recommendation based on the user's historical behavior data and the similarity calculation data, thereby generating product recommendation data;

[0111] In an embodiment of the present invention, based on the user's historical behavior data and the similarity calculation data, a recommendation algorithm is used for personalized recommendation; for a specific user, according to their historical behavior data and the similarity calculation result, products that match their interests and preferences are recommended; product recommendation data is generated, including relevant data of the recommended products.

[0112] Preferably, step S1 includes the following steps:

[0113] Step S11: Perform time-frequency conversion on the user's historical behavior data to generate frequency-domain representation data;

[0114] Step S12: Draw a spectrogram for the frequency-domain representation data to generate spectrogram data;

[0115] Step S13: Perform peak statistics on the spectrogram data to generate periodic component data;

[0116] Step S14: Judge the periodic trend of the periodic component data to generate user habit data;

[0117] Step S15: Conduct multi-factor consumption mining on the user's historical behavior data to generate data on the to-be-consumed themes;

[0118] Step S16: Construct a user portrait model based on the user habit data and the user text theme data to generate user portrait model data;

[0119] Step S17: Use the user portrait model data to perform model prediction on the user's historical behavior data to generate user portrait data.

[0120] Through time-frequency conversion, the present invention can convert the user's historical behavior data into frequency-domain representation data, thereby extracting the characteristics of its frequency components, which helps to discover the periodic changes or important frequency components in the data and provides a basis for subsequent spectrum analysis and periodic detection; by drawing a spectrogram, the frequency-domain representation data can be visually visualized as spectrogram data, and the spectrogram shows the energy distribution of different frequency components, helping us observe and analyze the frequency-domain characteristics in the data; by performing peak statistics on the spectrogram data, the periodic components corresponding to the frequency peaks can be identified, which helps to extract the periodic change components in the data and provides a basis for subsequent periodic trend judgment; through periodic trend judgment, it can be analyzed whether the periodic component data shows an obvious periodic change trend, which helps to identify and discover the periodic behavior habits of users; by integrating the user habit data and the to-be-consumed theme data, a user portrait model can be constructed to help the system better understand the user and provide personalized product recommendations and services; based on the constructed user portrait model, the user's historical behavior data can be predicted and inferred, which helps to predict the user's future behavior and interest evolution.

[0121] As an example of the present invention, refer to Figure 2 As shown, in this example, step S1 includes:

[0122] Step S11: Perform time-frequency conversion on the user's historical behavior data to generate frequency-domain representation data;

[0123] In the embodiments of the present invention, Fourier transform is applied to perform time-frequency conversion on the user's historical behavior data, converting the time-domain data into frequency-domain representation data; the frequency-domain representation data is obtained, which represents the changes of the user behavior data at different frequencies.

[0124] Step S12: Draw a spectrogram for the frequency-domain representation data, thereby generating spectrogram data;

[0125] In the embodiments of the present invention, the frequency-domain representation data is used to draw a spectrogram with the frequency as the horizontal axis and the amplitude or power as the vertical axis.

[0126] Step S13: Perform peak statistics on the spectrogram data, thereby generating periodic component data;

[0127] In the embodiments of the present invention, the spectrogram data is analyzed to identify peak points, that is, the data points with the highest amplitude or power in the frequency-domain data; the positions and amplitude data of the peak points are statistically analyzed, and periodic component data is obtained, which represents the components with obvious periodicity in the frequency-domain data.

[0128] Step S14: Judge the periodic trend of the periodic component data, thereby generating user habit data;

[0129] In the embodiments of the present invention, a periodic analysis method is used to perform periodic trend analysis on the periodic component data to detect the periodic change trend; according to the analysis result of the periodic trend, user habit data is generated, which represents the changes in the user's behavior habits or preferences in different time periods or different cycles.

[0130] Step S15: Perform multi-factor consumption mining on the user's historical behavior data, thereby generating data of the to-be-consumed theme;

[0131] In the embodiments of the present invention, the user's historical behavior data is decomposed into fluctuation cycles and analyzed for regional latitude changes. Finally, product change loss parameters and product theme data are obtained, and then consumption desire scoring factors are generated.

[0132] Step S16: Construct a user portrait model based on the user habit data and the data of the to-be-consumed theme, thereby generating user portrait model data;

[0133] In the embodiments of the present invention, the user habit data and the data of the to-be-consumed theme are combined to construct a user portrait model; the user portrait model will comprehensively consider the user's behavior habits, emotional tendencies and interest preferences, and generate user portrait model data, which represents the comprehensive characteristics and portrait data of the user.

[0134] Step S17: Use the user portrait model data to perform model prediction on the user's historical behavior data, thereby generating user portrait data;

[0135] In an embodiment of the present invention, a prediction model is used to perform model prediction on user historical behavior data to predict the user's future behavior preferences or interests; user portrait data is generated to represent the user's possible future behavior tendencies and interest characteristics.

[0136] Preferably, step S15 includes the following steps:

[0137] Step S151: Perform wave cycle decomposition on the user historical behavior data to generate wave cycle decomposition data;

[0138] Step S152: Identify periodic behavior changes in the wave cycle decomposition data to generate user behavior change data;

[0139] Step S153: Extract product change loss parameters from the user behavior change data to generate product change loss parameters;

[0140] Step S154: Analyze the regional latitude changes in the user historical behavior data to generate local climate environment data;

[0141] Step S155: Screen product themes based on the local climate environment data to generate product theme data;

[0142] Step S156: Use the product change loss parameters to correct the product themes in the product theme data to generate user target products;

[0143] Step S157: Infer the user's mood based on the local climate environment data and accumulate the consumption desire scores to generate a consumption desire scoring factor;

[0144] Step S158: Use the consumption desire scoring factor to compensate the user target products to generate the to-be-consumed theme data.

[0145] Through the decomposition of the fluctuation cycle of the user's historical behavior data, the present invention can reveal potential repetitive patterns and periodic behaviors, which helps to understand the periodic changes in user behavior and provides a basis for subsequent analysis. By analyzing and identifying the decomposed data of the fluctuation cycle, the changing trends of user behavior can be captured, and identifying these behavior changes can help to understand user preferences. By extracting the product change and loss parameters, the impact degree of user behavior changes on the product can be quantified, and these parameters can help to optimize product features, reduce user dissatisfaction with the product, and better meet user needs. The analysis of the regional latitude changes can help to understand the differences in the behavior patterns and preferences of users in different regions. Based on the local climate environment data, the consumption behavior of users can be associated with the geographical environment, providing useful information for customized recommendations and market positioning. By screening product themes based on the local climate environment data, customized product recommendations can be made according to the differences in the environments where users are located, which helps to improve the relevance of products and user satisfaction, and increase users' interest in and willingness to purchase products. By using the product change and loss parameters, the product theme data can be corrected to adapt to the changes in user behavior and preferences, which helps to generate products that meet user goals and needs and improve user acceptance and satisfaction with the products. By associating emotion speculation with the consumption desire, a consumption desire scoring factor can be generated, thereby understanding the changes in user tendencies and purchase willingness and providing guidance for personalized recommendations and marketing. By using the consumption desire scoring factor to compensate the target products of users, the product recommendations can be adjusted according to users' desires and tendencies.

[0146] In the embodiments of the present invention, historical behavior data of users is collected, including purchase records, browsing histories, and geographical location data. The original data is decomposed into different fluctuating cycle components to obtain the periodic change data of user behavior. Time series analysis technology is used to identify the periodic behavior changes in the decomposed data of the fluctuating cycle, such as data on the increase in the purchase volume or the decrease in the browsing volume of users within a specific time period. By comparing the sales data and user feedback of products in different time periods, the degree of loss caused by the changes is evaluated, and the change values of the user churn rate and the purchase conversion rate are extracted. Geographical data related to users, such as the cities and regions where they are located, is obtained, and user groups in different regions are identified based on the geographical data. According to the geographical climate environment data, regions related to specific product themes are screened. For cold drink products, regions with higher temperatures are selected as the themes suitable for sales. Combining the user behavior data and the geographical climate environment data, the preference degrees of users in different regions for specific product themes are analyzed. Combining the product theme data and the product change loss parameters, the relative advantages and disadvantages of different product themes under user behavior changes are analyzed. According to the analysis results, the product theme data is corrected, and the characteristics of different product themes are adjusted to meet the user's goals and needs. According to the geographical climate environment data, the user's emotional state is speculated, and the user's emotional state is associated with the geographical climate environment data to accumulate the scores of the consumption desire. Combining the user's target products and the consumption desire scoring factors, the products are compensated to generate the desired consumption theme data, which reflects the user's consumption tendency and preference for product themes in different emotional states.

[0147] Preferably, step S2 includes the following steps:

[0148] Step S21: Extract the picture quality conditions from the product image data to generate quality condition data;

[0149] Step S22: Use the quality condition data to perform adaptive image enhancement on the product image data to generate image enhancement data;

[0150] Step S23: Perform weighted pooling calculation on the image enhancement data to generate product image feature data;

[0151] Step S24: Perform image marking on the image enhancement data to generate marked product image data;

[0152] Step S25: Perform attribute label analysis on the marked product image data to generate attribute label analysis data;

[0153] Step S26: Perform attribute coding on the attribute label analysis data to generate image attribute label data;

[0154] Step S27: Perform correlation positioning on the product image feature data to generate feature positioning scaling data;

[0155] Step S28: Construct an image recognition model based on the feature fixed-point scaling data and the image attribute label data, so as to generate image recognition model data;

[0156] Step S29: Use the image recognition model data to perform hyperparameter probability judgment on the product image feature data, so as to generate product attribute data.

[0157] The present invention can evaluate the quality of the product image through picture quality condition extraction, including factors such as image clarity, exposure, and noise; adaptive image enhancement based on the quality condition data can improve the visual effect of the image, enhance the clarity of the image, and contribute to enhancing the visual attractiveness, recognizability, and visual quality of the product image; by performing weighted pooling calculation on the image enhancement data, key product image features can be extracted, and these features can capture the visual attributes of the edges, textures, and shapes of the image. The extracted product image feature data can be used for subsequent image classification, similarity matching, retrieval, and recognition tasks; labeling the product image data helps to establish the connection between the picture and the product data, improving the searchability, understandability, and usability of the image; analyzing the attribute labels of the labeled product image data can identify and extract key product attributes, which helps to understand the data of the product's features, specifications, functions, and uses; through attribute encoding, the attribute data can be converted into a digital encoding form, facilitating subsequent machine learning and data analysis tasks; the feature fixed-point scaling data can be used to optimize the distribution of the feature space and improve the performance of subsequent machine learning algorithms; the image recognition model data can be used for tasks such as product automatic classification, recognition, and detection.

[0158] In the embodiment of the present invention, the product image is analyzed and evaluated to extract features related to the quality conditions. By analyzing these features, quality condition data is generated; according to the quality condition data, adaptive image enhancement processing is performed on the product image data; pooling operations are performed on the image enhancement data to extract key image features, and dimensionality reduction processing is performed on the image enhancement data to obtain a set of data representing the image features; the image enhancement data is labeled to add semantic data to the image, marking different objects in the image and adding corresponding labels to each object; each label in the image is analyzed and recognized to extract the attribute labels of the objects; one-hot encoding is used to convert the attribute labels into a representation form; a feature fixed-point scaling data and attribute label data are used to train an image classifier to identify different categories of product images; by inputting the product image feature data into the image recognition model, the model is used to predict the attribute probability of the image.

[0159] Preferably, the adaptive image enhancement processing is performed through an adaptive image enhancement calculation formula, and the specific adaptive image enhancement calculation formula is:

[0160]

[0161] In the formula, I(x, y) represents the pixel value of the enhanced image, (x, y) represents the pixel coordinates on the image, N represents the number of samples of the quality condition data, i represents the index value of the quality condition data, ω(i) represents the weight factor of the i-th quality condition data, n represents a number approaching infinity, f(i, x, y) represents the change rate of the i-th quality condition data at the position (x, y), g(i, x, y) represents the gradient mean value of the i-th quality condition data, θ(i, x, y) represents the function angle value of the i-th quality condition data, ρ(i, x, y) represents the adaptive factor of the i-th quality condition data at (x, y), and α represents the adaptive image enhancement error correction amount.

[0162] Among them, the derivation process of the adaptive image enhancement calculation formula is as follows:

[0163] Step 1: Define the quality condition data

[0164] Suppose there are N samples of quality condition data with the index i. Each sample has the following attributes:

[0165] ω(i): The weight factor of the i-th quality condition data.

[0166] f(i, x, y): The change rate of the i-th quality condition data at the position (x, y).

[0167] g(i, x, y): The gradient mean value of the i-th quality condition data.

[0168] θ(i, x, y): The function angle value of the i-th quality condition data.

[0169] ρ(i, x, y): The adaptive factor of the i-th quality condition data at the position (x, y).

[0170] Step 2: Derivation process

[0171] To understand the derivation process more clearly, we decompose the above formula into the following steps:

[0172] Step 2.1: Normalize the change rate

[0173] First, we divide the change rate f(i, x, y) by the gradient mean value

[0174] Step 2.2: Take the limit

[0175] Then, we take the limit This represents the value of the change rate divided by the gradient mean value when n approaches infinity.

[0176] Step 2.3: Multiply by sin(θ(i, x, y)) and

[0177] Next, we multiply the result of Step 2.2 by the sine of the function angle value sin(θ(i, x, y)) and the square root of the adaptive factor

[0178] Step 2.4: Weighted average

[0179] Then, we perform a weighted average on the results of all quality condition data. Each result is multiplied by the corresponding weight factor ω(i), then summed and divided by the number of samples N:

[0180]

[0181] Step 2.5: Add the error correction amount

[0182] Finally, we add the adaptive image enhancement error correction amount α to the result of Step 2.4:

[0183]

[0184] In this way, the derivation process of the calculation formula for adaptive image enhancement is completed.

[0185] The adaptive image enhancement calculation formula of the present invention is used for adaptive image enhancement of product image data; in the formula Taking the number of samples N of the quality condition data as the normalization factor for calculating the average value. By taking the average value processing, the influence of sample data on the result can be reduced, ensuring the integrity and consistency of the enhanced image; ω(i) is used for weighted processing of different quality condition data. By adjusting the weight factor, the importance of different quality condition data can be adjusted, so that in the enhancement process, important quality condition data has a greater influence, thereby improving the adaptability and flexibility of the enhancement effect; The part is for normalizing the change rate and gradient mean of the quality condition data. By calculating the logarithm of the ratio of the change rate to the gradient mean and taking the limit, the significantly changing area can be magnified, enhancing the contrast and detail data of the image, which helps to highlight the edge and texture features in the image; sin(θ(i, x, y)) can adjust the weight according to the feature changes in different directions by introducing angle data, thereby changing the brightness and contrast of the image; As a weight value, used to adjust the weights of quality condition data at different pixel positions, the adaptive factor can be adaptively adjusted according to the characteristic changes in the local area to improve the adaptability and locality of image enhancement; α can finely adjust the enhanced image, correct possible errors and biases, and further improve the accuracy and quality of image enhancement.

[0186] Preferably, step S23 includes the following steps:

[0187] Step S231: Scale the size of the model image of the image enhancement data to generate image input data;

[0188] Step S232: Perform maximum convolution calculation, average convolution calculation, and minimum convolution calculation on the image input data using a preset pre-trained model to generate first initial convolution data, second initial convolution data, and third initial convolution data;

[0189] Step S233: Batch-normalize and activate the first initial convolution data, second initial convolution data, and third initial convolution data to generate first convolution data, second convolution data, and third convolution data;

[0190] Step S234: Perform maximum pooling calculation on the first convolution data to generate first pooling data; perform average pooling calculation on the second convolution data to generate second pooling data; perform minimum pooling calculation on the third convolution data to generate third pooling data;

[0191] Step S235: Perform corresponding preset weight calculations on the first pooling data, second pooling data, and third pooling data respectively to obtain first confidence data, second confidence data, and third confidence data;

[0192] Step S236: Conduct result voting based on the first confidence data, second confidence data, and third confidence data to generate feature representation confidence data;

[0193] Step S237: Generate product image feature data based on the feature representation confidence data.

[0194] The present invention adjusts images of different sizes to the same size, ensuring that the model can adapt to various input images, increasing the generalization ability of the model, reducing the computational amount and memory consumption; by finding the maximum value in the image, important features of edges and details can be highlighted, by calculating the average value of pixels in the image, the image can be smoothed and noise can be reduced, by finding the minimum value in the image, details of dark parts and shadows can be highlighted; by normalizing the data of each batch, the training process can be accelerated, the problems of gradient disappearance and gradient explosion can be reduced, the stability and convergence speed of the network can be improved, applying an activation function to the convolutional data can introduce non-linear characteristics, enhancing the expression ability of the model, enabling it to learn more complex functions; by selecting the maximum value in the pooling window, the main features in the image can be retained, by calculating the average value of pixels in the pooling window, the size of the feature map can be reduced and the features can be smoothed, by selecting the minimum value in the pooling window, details of dark parts and shadows in the image can be retained; according to different weight calculation methods, different features can be adaptively selected and emphasized, improving the expression ability of the model for input data; by integrating the voting results of different confidence data, a more comprehensive and representative feature representation can be obtained, enhancing the description ability of the image content and reducing the misjudgment risk brought by individual data; according to the confidence data of the feature representation, the product image features most relevant to solving a specific task can be selected and extracted, reducing redundant and noisy data and improving the effectiveness of the features.

[0195] In an embodiment of the present invention, the resize function of the OpenCV library is used to perform image scaling operations; the scaled image data is input into a pre-trained model, and according to the design of the model, maximum convolution, average convolution, and minimum convolution calculations are performed to obtain first initial convolution data, second initial convolution data, and third initial convolution data; batch normalization operations are performed on the first initial convolution data, second initial convolution data, and third initial convolution data, and then the normalized data is activated through an activation function (such as ReLU) to obtain first convolution data, second convolution data, and third convolution data; Max Pooling is used to perform maximum pooling operations on the first convolution data, Average Pooling is used to perform average pooling operations on the second convolution data, and Min Pooling is used to perform minimum pooling operations on the third convolution data; corresponding preset weight calculations are respectively performed on the first pooled data, second pooled data, and third pooled data, so as to obtain first confidence data, second confidence data, and third confidence data; the majority voting algorithm is used to vote on the first confidence data, second confidence data, and third confidence data to determine the final confidence data of the feature representation.

[0196] Preferably, step S27 includes the following steps:

[0197] Step S271: Perform feature standardization on the product image feature data to generate standardized feature data;

[0198] Step S272: Calculate the balanced rank for the standardized feature data to generate balanced rank data;

[0199] Step S273: Calculate the rank correlation coefficient for the balanced rank data to generate rank correlation coefficient data;

[0200] Step S274: Conduct correlation verification based on the rank correlation coefficient data to generate correlation verification data;

[0201] Step S275: Judge the correlation strength of the correlation verification data to generate correlation feature data;

[0202] Step S276: Perform feature fixed-point scaling on the correlation feature data to generate feature fixed-point scaling data.

[0203] In the present invention, by standardizing the feature data, the value ranges of different features can be unified to the same scale, avoiding the influence of value differences between different features on subsequent processing; by calculating the balanced rank data, the feature values can be sorted and transformed into a rank form, making the data have strong comparability and interpretability; the rank correlation coefficient can measure the correlation degree between different features, thus helping us understand the correlation and mutual influence degree between features; through correlation verification, we can determine which features actually have significant correlations, thus excluding irrelevant features or features that may introduce noise, and improving the accuracy of modeling and prediction; by judging the correlation verification data, the strength of the correlation can be expressed in numerical form, helping us distinguish strong correlation, weak correlation, and non-correlation feature relationships; by feature fixed-point scaling, the weights of features can be adjusted, making features that have a greater impact on the result have higher importance, improving the performance of the model on training and test data, and enhancing the generalization ability and prediction accuracy of the model.

[0204] As an example of the present invention, refer to Figure 3 As shown, in this example, step S27 includes:

[0205] Step S271: Perform feature standardization on the product image feature data to generate standardized feature data;

[0206] In the embodiment of the present invention, the value of each feature is converted into its deviation from the average value and divided by the standard deviation to generate standardized feature data.

[0207] Step S272: Calculate the balanced rank for the standardized feature data to generate balanced rank data;

[0208] In the embodiments of the present invention, the numerical value of each feature is converted into its sorting position in the feature vector; for n samples, the minimum value is ranked in the first position, the maximum value is ranked in the nth position, and the rankings of the remaining values are determined according to the size; in this way, balanced rank data is generated.

[0209] Step S273: Calculate the rank correlation coefficient for the balanced rank data, thereby generating rank correlation coefficient data;

[0210] In the embodiments of the present invention, the Spearman rank correlation coefficient is used to perform correlation analysis on the balanced rank data; the rank correlation coefficient measures the degree of correlation of the rank relationship between two variables; by calculating the rank correlation coefficient, rank correlation coefficient data is obtained, which is used to measure the correlation between features.

[0211] Step S274: Perform correlation verification based on the rank correlation coefficient data, thereby generating correlation verification data;

[0212] In the embodiments of the present invention, a threshold of 0.5 is set to determine the existence of correlation; if the rank correlation coefficient is greater than the threshold, it indicates that there is a correlation relationship between features; if the rank correlation coefficient is less than or equal to the threshold, it indicates that there is no correlation between features; through this correlation verification process, correlation verification data is generated.

[0213] Step S275: Judge the correlation strength of the correlation verification data, thereby generating correlation feature data;

[0214] In the embodiments of the present invention, the correlation is divided into strong correlation, medium correlation and weak correlation levels, and the correlation verification data is analyzed. Through this judgment process, correlation feature data is generated.

[0215] Step S276: Perform feature fixed-point scaling on the correlation feature data, thereby generating feature fixed-point scaling data.

[0216] In the embodiments of the present invention, the correlation feature data is mapped to a range between 0 and 100; the feature fixed-point scaling process is performed on the correlation feature data; through this feature fixed-point scaling process, feature fixed-point scaling data is generated.

[0217] Preferably, step S28 includes the following steps:

[0218] Step S281: Divide the feature fixed-point scaling data and the image attribute label data into datasets, thereby generating training set data and test set data;

[0219] Step S282: Use the support vector machine technology to construct a model, thereby generating first recognition model data;

[0220] Step S283: Conduct the first hyperparameter search on the first recognition model data to generate hyperparameter configuration data;

[0221] Step S284: Dynamically adjust the validation set for the training set data according to the hyperparameter configuration data to generate dynamically adjusted data;

[0222] Step S285: Conduct the second hyperparameter search on the first recognition model data using the dynamically adjusted data to generate hyperparameter data;

[0223] Step S286: Optimize the hyperparameters of the first recognition model data using the hyperparameter data to generate the second recognition model data;

[0224] Step S287: Evaluate the second recognition model data using the test set data to generate the second model evaluation data;

[0225] Step S288: Optimize the second recognition model data using the second model evaluation data to generate the image recognition model data.

[0226] By partitioning the dataset, the present invention can reduce the overfitting phenomenon of the model on the training set, verify the effect of the model on unseen test data, and improve the robustness and reliability of the model; the support vector machine model can provide an explanation for the classification decision, helping us understand the data distribution and decision boundary behind the model; the first hyperparameter search helps us determine a suitable set of hyperparameters for the support vector machine model, and the second hyperparameter search can search for the best combination in a limited hyperparameter space, reducing the complexity of parameter search and accelerating the model optimization process; by adjusting the partitioning method of the validation set according to the hyperparameter configuration data, the performance of the model under different hyperparameter combinations can be better evaluated, which helps to select the best hyperparameters; by using the dynamically adjusted data for the second hyperparameter search, the performance of the model can be further optimized, improving the accuracy and generalization ability of the model; using the hyperparameter data to optimize the first recognition model can improve the performance of the model, making the model fit the data more accurately and improving the robustness of the model; evaluating the second recognition model using the test set data can obtain the accuracy of the model on unseen data, verifying the generalization ability and prediction ability of the model; through the second model evaluation data, the deficiencies of the model can be identified, and corresponding optimization and improvement can be carried out to improve the performance and prediction accuracy of the model. After model optimization, the generated image recognition model data can be used in actual application scenarios for accurate image recognition and classification tasks.

[0227] In the embodiments of the present invention, 70% of the feature fixed-point scaling data and image attribute label data is used for training, and 30% of the data is used for testing; the support vector machine technology is used to construct a model for the training set data, an SVM is used to construct a classifier, and the image attributes are mapped to the corresponding labels. Through this step, the first recognition model data is generated; the hyperparameters of the first recognition model data are searched for the first time. By trying different hyperparameter configurations, the best hyperparameter configuration is found; then, techniques such as cross-validation are used to evaluate the performance of each hyperparameter configuration; according to the hyperparameter configuration data, the training set data is further divided into a training set and a validation set, and a dynamic adjustment method is used to adjust the hyperparameter configuration of the model according to the performance of the validation set; by trying different hyperparameter configurations and using the validation set to evaluate the performance, the best hyperparameter configuration is found; according to the found best hyperparameter configuration, the model is retrained to obtain the second recognition model data; the performance of the model is evaluated by calculating the difference between the prediction result and the true label or using other evaluation metrics; according to the evaluation result, optimization operations such as model adjustment, feature selection, and sample adjustment are performed.

[0228] Preferably, step S29 includes the following steps:

[0229] Step S291: Classify the product image feature data by using the image recognition model data to generate feature classification data;

[0230] Step S292: Calculate the hyperparameter weights for the feature classification data to generate feature calculation data;

[0231] Step S293: Predict the feature probability for the feature calculation data to generate feature probability prediction data;

[0232] Step S294: Sort the probability gradients for the feature probability prediction data to generate probability gradient sorting data;

[0233] Step S295: Perform category attribute association according to the preset category attribute relationship data and the probability gradient sorting data to generate product attribute data.

[0234] Through the image recognition model, the present invention classifies the product image feature data, which can effectively classify different features, identify various features existing in the product image, extract the key features in the product image, and thus help to understand and describe the attributes and characteristics of the product; by calculating the hyperparameter weights for the feature classification data, the importance and contribution degree of different features can be determined, which is helpful for further optimizing the feature selection and modeling process; using the feature calculation data for feature probability prediction can obtain the probability of each feature appearing in the product image, helping us understand the feature distribution of the product image; by performing probability gradient sorting on the feature probability prediction data, the most representative features in each product image can be determined, providing more accurate and effective feature descriptions; through the preset category attribute relationship data and probability gradient sorting data, the features of the product can be associated with their corresponding attributes, further enriching and improving the product description and attribute data.

[0235] As an example of the present invention, referring to Figure 4 as shown, in this example, the step S29 includes:

[0236] Step S291: Use the image recognition model data to classify the product image feature data, thereby generating feature classification data;

[0237] In the embodiment of the present invention, the product image is input into the model, and the model will classify the features according to the learned feature and label data, and assign each product image to the corresponding feature category; through this step, the feature classification data is generated.

[0238] Step S292: Calculate the hyperparameter weights for the feature classification data, thereby generating feature calculation data;

[0239] In the embodiment of the present invention, according to the importance of the feature in the classification task, the corresponding weight is adjusted, and the feature calculation data is obtained by calculating the product of the feature classification data and the hyperparameter weight.

[0240] Step S293: Perform feature probability prediction on the feature calculation data, thereby generating feature probability prediction data;

[0241] In the embodiment of the present invention, by applying the probability model, the probability prediction result of each feature category is obtained, representing the probability value that the product image belongs to the feature category; through this step, the feature probability prediction data is generated.

[0242] Step S294: Perform probability gradient sorting on the feature probability prediction data, thereby generating probability gradient sorting data;

[0243] In the embodiments of the present invention, the probability prediction values of each product image in different feature categories are compared, and the feature category with the highest probability is selected as the main feature category of the product image; through this step, probability gradient sorting data is generated.

[0244] Step S295: Perform category attribute association according to the preset category attribute relationship data and the probability gradient sorting data, so as to generate product attribute data;

[0245] In the embodiments of the present invention, according to the preset category attribute relationship data, the probability gradient sorting data is associated with the corresponding attributes to generate product attribute data; for example, if the probability gradient sorting data determines that a product belongs to a specific feature category, according to the preset relationship, the attributes corresponding to the feature category can be assigned to the product; through this step, the final product attribute data is generated.

[0246] Preferably, the present invention also provides a product management system based on an intelligent e-commerce platform, which is used to execute the above-mentioned product management method based on an intelligent e-commerce platform, including:

[0247] A user portrait construction module, which is used to obtain user historical behavior data; construct a user portrait for the user historical behavior data, so as to generate user portrait data;

[0248] A product image feature analysis module, which is used to obtain product image data; extract product image features from the product image data, so as to generate product image feature data; perform hyperparameter image recognition on the product image data, so as to generate product attribute data;

[0249] A product feature vector generation module, which is used to perform data association on the product image feature data and the product attribute data, so as to generate product feature vector data;

[0250] A similarity calculation module, which is used to calculate the similarity between the user portrait data and the product feature vector data, so as to generate similarity data;

[0251] A personalized recommendation module, which is used to perform personalized recommendation according to the user historical behavior data and the similarity calculation data, so as to generate product recommendation data.

[0252] The beneficial effects of this application are as follows. By analyzing the user's historical behavior data, it is possible to understand the user's preferences, likes, and behavior patterns, including purchase records, click behaviors, browsing habits, etc., which helps to reveal the user's interests and needs and provide personalized recommendations and services. By analyzing and mining the user's historical behavior data, a user profile can be constructed, that is, a description and feature summary of the user, which helps to better understand the user's needs and provide more accurate product recommendations and personalized experiences. By obtaining product image data, the description and data of the product can be enriched. The product image data can provide a more intuitive and vivid product display to help users better understand and evaluate the product. By extracting features from the product image data, a visual feature representation of the product can be obtained, which helps to extract visual features such as the texture, color, and shape of the product for subsequent product comparison, similarity calculation, and recommendation. By associating the product image feature data with the product attribute data, the visual features and other attribute features of the product can be comprehensively considered, which helps to construct a more comprehensive and multi-dimensional product feature vector and provide the input data required for more accurate similarity calculation and recommendation algorithms. By calculating the similarity between the user profile data and the product feature vector data, the matching degree between the user and the product can be evaluated, which helps to find the user's potential interest areas and recommend products with high similarity to the user. Based on the user's historical behavior data and similarity calculation data, personalized product recommendations can be realized. By analyzing the user's behaviors and preferences and combining the similarity evaluation of the products, products that meet the user's interests and preferences can be recommended to them, improving user satisfaction and shopping experience. Therefore, the present invention provides a product management method and system based on an intelligent e-commerce platform, which realizes personalized product push for users by constructing a user profile and accurately identifying and associating product image features and attributes.

[0253] Therefore, in any aspect, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, it is intended to cover all changes falling within the meaning and scope of the equivalent elements of the application documents within the present invention.

[0254] The above are only the specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A product management method based on an intelligent e-commerce platform, characterized in that, It includes the following steps: Step S1: Obtain the user's historical behavior data; perform compensation for the desired consumption factors on the user's historical behavior data to generate user portrait data; Step S2: Obtain the product image data; extract the product image features from the product image data to generate product image feature data; perform hyperparameter image recognition on the product image data to generate product attribute data, where Step S2 specifically includes: Step S21: Extract the picture quality conditions from the product image data to generate quality condition data; Step S22: Use the quality condition data to perform adaptive image enhancement on the product image data to generate image enhancement data, and the adaptive image enhancement process is processed through the adaptive image enhancement calculation formula, where the adaptive image enhancement calculation formula is specifically: ； Wherein, represents the pixel value of the enhanced image, represents the pixel coordinates on the image, represents the number of samples of the quality condition data, represents the index value of the quality condition data, represents the weight factor of the th quality condition data, represents a number approaching infinity, represents the th quality condition data at the position represents the th gradient mean value of the quality condition data, represents the th function angle value of the quality condition data, represents the th quality condition data at the adaptive factor at, represents the adaptive image enhancement error correction amount; Step S23: Perform weighted pooling calculation on the image enhancement data to generate product image feature data; Step S24: Perform image marking on the image enhancement data to generate marked product image data; Step S25: Perform attribute label analysis on the marked product image data to generate attribute label analysis data; Step S26: Perform attribute coding on the attribute label analysis data to generate image attribute label data; Step S27: Perform correlation pinpointing on the product image feature data to generate feature pinpointing and scaling data; Step S28: Construct an image recognition model for the feature pinpointing and scaling data and the image attribute label data to generate image recognition model data; Step S29: Use the image recognition model data to perform hyperparameter probability judgment on the product image feature data to generate product attribute data; Step S3: Perform data association on the product image feature data and the product attribute data to generate product feature vector data; Step S4: Calculate the similarity between the user portrait data and the product feature vector data to generate similarity data; Step S5: Perform personalized recommendation based on the user's historical behavior data and the similarity calculation data to generate product recommendation data.

2. The product management method based on the intelligent e-commerce platform according to claim 1, wherein Step S1 includes the following steps: Step S11: Perform time-frequency conversion on the user's historical behavior data to generate frequency-domain representation data; Step S12: Draw a spectrogram for the frequency-domain representation data to generate spectrogram data; Step S13: Perform peak statistics on the spectrogram data to generate periodic component data; Step S14: Judge the periodic trend of the periodic component data to generate user habit data; Step S15: Perform multi-factor consumption mining on the user's historical behavior data to generate desired consumption theme data; Step S16: Construct a user portrait model for the user habit data and the desired consumption theme data to generate user portrait model data; Step S17: Use the user portrait model data to perform model prediction on the user's historical behavior data to generate user portrait data.

3. The product management method based on the intelligent e-commerce platform according to claim 2, wherein, Step S15 includes the following steps: Step S151: Perform wavelet cycle decomposition on the user's historical behavior data to generate wavelet cycle decomposition data; Step S152: Identify the periodic behavior changes in the decomposed data of the fluctuation period, so as to generate user behavior change data; Step S153: Extract the product change loss parameters from the user behavior change data, so as to generate product change loss parameters; Step S154: Analyze the regional latitude changes in the user's historical behavior data, so as to generate location climate environment data; Step S155: Screen the product themes according to the location climate environment data, so as to generate product theme data; Step S156: Use the product change loss parameters to correct the product changes in the product theme data, so as to generate the user's target product; Step S157: Infer the user's mood based on the location climate environment data and accumulate the consumption desire scores, so as to generate a consumption desire scoring factor; Step S158: Use the consumption desire scoring factor to compensate the user's target product, so as to generate the desired consumption theme data.

4. The product management method based on the intelligent e-commerce platform according to claim 1, wherein, Step S23 includes the following steps: Step S231: Scale the size of the model image for the image enhancement data, so as to generate image input data; Step S232: Use a preset pre-trained model to perform maximum convolution calculation, average convolution calculation, and minimum convolution calculation on the image input data, so as to generate first initial convolution data, second initial convolution data, and third initial convolution data; Step S233: Batch normalize and activate the first initial convolution data, second initial convolution data, and third initial convolution data, so as to generate first convolution data, second convolution data, and third convolution data; Step S234: Perform maximum pooling calculation on the first convolution data to generate first pooling data; perform average pooling calculation on the second convolution data to generate second pooling data; perform minimum pooling calculation on the third convolution data to generate third pooling data; Step S235: Perform corresponding preset weight calculations on the first pooling data, second pooling data, and third pooling data respectively, so as to obtain first confidence data, second confidence data, and third confidence data; Step S236: Conduct a result vote based on the first confidence data, second confidence data, and third confidence data, so as to generate feature representation confidence data; Step S237: Generate product image feature data based on the feature representation confidence data.

5. The product management method based on the intelligent e-commerce platform according to claim 1, characterized in that, Step S27 includes the following steps: Step S271: Standardize the features of the product image feature data to generate standardized feature data; Step S272: Calculate the balanced rank for the standardized feature data to generate balanced rank data; Step S273: Calculate the rank correlation coefficient for the balanced rank data to generate rank correlation coefficient data; Step S274: Conduct a correlation verification based on the rank correlation coefficient data to generate correlation verification data; Step S275: Judge the correlation strength of the correlation verification data to generate correlation feature data; Step S276: Perform feature fixed-point scaling on the correlation feature data to generate feature fixed-point scaling data.

6. The product management method based on the intelligent e-commerce platform according to claim 1, wherein, Step S28 includes the following steps: Step S281: Divide the feature fixed-point scaling data and the image attribute label data into datasets, thereby generating training set data and test set data; Step S282: Use the support vector machine technology to construct a model, thereby generating the first recognition model data; Step S283: Conduct the first search for hyperparameters on the first recognition model data, thereby generating hyperparameter configuration data; Step S284: Dynamically adjust the validation set for the training set data according to the hyperparameter configuration data, thereby generating dynamically adjusted data; Step S285: Conduct a second search for hyperparameters on the first recognition model data using the dynamically adjusted data, thereby generating hyperparameter data; Step S286: Optimize the hyperparameters of the first recognition model data using the hyperparameter data, thereby generating the second recognition model data; Step S287: Evaluate the second recognition model data using the test set data, thereby generating the second model evaluation data; Step S288: Optimize the second recognition model data using the second model evaluation data, thereby generating the image recognition model data.

7. The product management method based on the intelligent e-commerce platform according to claim 1, characterized in that, Step S29 includes the following steps: Step S291: Classify the product image feature data using the image recognition model data, thereby generating feature classification data; Step S292: Calculate the hyperparameter weights for the feature classification data, thereby generating feature calculation data; Step S293: Predict the feature probabilities for the feature calculation data, thereby generating feature probability prediction data; Step S294: Sort the probability gradients for the feature probability prediction data, thereby generating probability gradient sorting data; Step S295: Perform category attribute association based on the preset category attribute relationship data and the probability gradient sorting data, thereby generating product attribute data.

8. A product management system based on an intelligent e-commerce platform, characterized in that, For implementing the product management method based on the intelligent e-commerce platform as described in claim 1, the product management system based on the intelligent e-commerce platform includes: A user portrait construction module, configured to obtain user historical behavior data; construct a user portrait for the user historical behavior data, thereby generating user portrait data; A product image feature analysis module, configured to obtain product image data; extract product image features from the product image data, thereby generating product image feature data; perform hyperparameter image recognition on the product image data, thereby generating product attribute data; A product feature vector generation module, configured to associate the product image feature data and the product attribute data, thereby generating product feature vector data; A similarity calculation module, configured to calculate the similarity between the user portrait data and the product feature vector data, thereby generating similarity data; A personalized recommendation module, configured to perform personalized recommendation based on the user historical behavior data and the similarity calculation data, thereby generating product recommendation data.

Images