Intelligent commodity multi-dimensional comparison method and system based on multi-modal analysis

Abstract

The invention relates to the technical field of intelligent commodity management, in particular to an intelligent commodity multi-dimensional comparison method and system based on multi-modal analysis. The method comprises the following steps: acquiring a text input by a user; performing similar commodity screening in a preset commodity database based on the user input text to generate similar commodity screening data; performing commodity display definition evaluation on the similar commodity screening data to generate commodity first-dimension evaluation data; obtaining commodity description parameters; carrying out commodity three-dimensional modeling on the multi-modal feature commodity by utilizing the commodity description parameters based on the image features of the commodity, and generating a commodity display three-dimensional model; and performing commodity display accuracy evaluation on the commodity display three-dimensional model to generate commodity second dimension evaluation data. According to the method, multi-dimensional evaluation, dynamic display adjustment and a visual report are introduced, so that the comprehensiveness of commodity comparison is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent commodity management, and in particular to a method and system for multi-dimensional comparison of intelligent commodities based on multimodal analysis. Background Art

[0002] Initially, product comparison methods mainly relied on traditional comparisons based on a single data source, such as static dimensions such as price and brand, which could not fully reflect the complex needs of consumers. With the advancement of information technology, especially the application of computer vision, natural language processing (NLP) and deep learning, more complex and intelligent multimodal analysis methods have emerged. Early multimodal analysis technologies focused on a single field, such as the combination of images and text, and assisted product comparison through basic image recognition and semantic analysis. As the technology matures, the analysis dimensions have gradually expanded to various forms of input, including voice, user behavior data, social media information, etc., forming an intelligent recommendation system for multimodal data fusion. This method is not limited to the comparison of surface features of products, but can also combine users' emotional preferences, historical purchasing behavior and real-time interactive data for comprehensive analysis. In recent years, innovations such as the multi-task learning framework and attention mechanism of deep learning have enabled intelligent product comparison to perform in-depth analysis in multiple dimensions and levels, providing more accurate and personalized comparison results. However, traditional product recommendation systems usually rely on static product information for screening, and fail to make dynamic adjustments based on user behavior and product display feedback. At the same time, product comparisons often focus on a single dimension, such as price or brand, and are unable to fully display the multi-dimensional characteristics of products, resulting in less comprehensive product comparisons. Summary of the invention

[0003] Based on this, it is necessary to provide a method and system for multi-dimensional comparison of smart products based on multimodal analysis to solve at least one of the above technical problems.

[0004] To achieve the above purpose, a method for multi-dimensional comparison of smart products based on multimodal analysis is provided, the method comprising the following steps: Step S1: obtaining user input text; screening similar products in a preset product database based on the user input text to generate similar product screening data; evaluating product display clarity on the similar product screening data to generate product first dimension evaluation data; Step S2: obtaining product description parameters; performing three-dimensional modeling of multimodal feature products using the product description parameters based on product image features to generate a three-dimensional product display model; evaluating the product display accuracy of the three-dimensional product display model to generate product second dimension evaluation data; Step S3: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; Step S4: Evaluate the product feedback dimension of the similar product screening data based on the product dynamic display adjustment data to generate product third-dimensional evaluation data; perform comprehensive dimensional comparison and visualization of the similar product screening data based on the product third-dimensional evaluation data, thereby generating a comprehensive dimensional comparison visualization report of similar products.

[0005] By obtaining the user's input text (such as search terms, preference descriptions, etc.), the system can intelligently understand the user's needs and accurately screen similar products, thereby providing personalized recommendations and avoiding the limitations of traditional methods of screening products by static conditions. Through dynamic screening, it is ensured that the displayed products match the user's needs, the relevance of product display is improved, and the time and energy of users browsing irrelevant products is reduced. Display clarity evaluation ensures that the displayed product images and description information are clear and accurate, so that users can quickly understand the characteristics and advantages and disadvantages of the products and improve the accuracy of purchase decisions. Converting products from two-dimensional plane displays to three-dimensional models can more intuitively present details such as the appearance, structure and size of the products, helping users to better understand the actual effects of the products, especially for complex products (such as electronic products, furniture, etc.). Product display accuracy evaluation ensures the consistency between the three-dimensional model and the actual product, avoids false or misleading displays, improves the user's trust in the product, and reduces returns and complaints caused by display distortion. Through three-dimensional modeling, users can perform interactive operations such as rotation and zooming on the product display page to enhance the shopping experience, making users feel as if they are experiencing the product in person and increase their desire to buy. Combining the evaluation of the first dimension (clarity) and the second dimension (accuracy), the system can conduct an in-depth analysis of the display of goods to ensure the authenticity and reliability of the displayed content and avoid false advertising or erroneous information affecting user judgment. According to the authenticity data of the goods, the system can dynamically adjust the display method of the goods (such as changing the product image, optimizing the description text, adjusting the display order, etc.), optimize the display effect according to real-time feedback, and ensure that the most suitable product display is continuously provided to users. By adjusting the product display in real time, it is possible to provide product information that better meets user expectations, reduce misleading and false displays, enhance user trust in the platform, and improve shopping conversion rates. By comprehensively evaluating the display effect, authenticity and user feedback of the goods, the system can comprehensively evaluate all dimensions of the goods, provide richer information, and help users make multi-angle decisions. The product feedback dimension evaluation combines user behavior data (such as click-through rate, purchase rate, evaluation, etc.) to reflect user feedback on product display in real time, further optimize product display strategy, and ensure that product display is more in line with user needs. The generated visualization report comprehensively compares various dimensions of products (such as clarity, accuracy, authenticity, feedback, etc.), intuitively presents the advantages and differences between products, helps users make comparisons easily, and improves shopping decision efficiency. Therefore, the present invention improves the comprehensiveness of product comparison by introducing multi-dimensional evaluation, dynamic display adjustment and visualization report.

[0006] Preferably, step S1 comprises the following steps: Step S11: Obtain user input text; Step S12: performing a product keyword search on the user input text to obtain product keyword search data; screening similar products in a preset product database based on the product keyword search data to generate similar product screening data; Step S13: extracting product display modal features from the similar product screening data to obtain product display modal feature data, wherein the product display modal features include text features, image features, and video features; Step S14: Evaluate the display clarity of similar products based on the product display modal feature data to generate product first dimension evaluation data.

[0007] By obtaining the user input text, the present invention can quickly identify user needs and provide accurate raw data for subsequent processing. The effectiveness of this step ensures that the system can receive the most direct user intention. By searching for commodity keywords, the commodity data related to user needs can be accurately captured. Combined with the screening of similar commodities in the commodity database, the system can quickly locate commodities that match user needs among a large number of commodities, thereby improving screening efficiency and accuracy. Extracting the text, image and video features of commodities helps to fully display the characteristics of commodities, which not only improves the multi-dimensional display effect of commodities, but also better meets the preferences of different users. For example, the impact of image and video features on visual experience, and text features help users understand the specific information of commodities. Evaluating the clarity of commodity display can optimize user experience. By evaluating the data in the first dimension, the commodity display can be further optimized according to indicators such as the visual effect of commodity display and the clarity of information transmission, ensuring that users obtain the best information experience.

[0008] Preferably, step S14 includes the following steps: Step S141: defining the clarity index of commodity display modality features, wherein the text feature clarity index includes information coverage, language simplicity and accuracy, the image feature clarity index includes resolution score, detail presentation score and color restoration score, and the video feature clarity index includes frame rate analysis, dynamic and static contrast and picture stability score; Step S142: Calculate the number of commodity display modes for the commodity display modal feature data to obtain the number of commodity display modal features; determine whether the number of commodity display modal features is greater than 1. If the number of commodity display modal features is greater than 1, mark the commodity as a multi-modal feature commodity; otherwise, mark the commodity as a single modal feature commodity and remove it; Step S143: performing modal score weighted fusion on multi-modal feature products according to the product display modal feature clarity index to generate multi-modal product display clarity; Step S144: Evaluate the display clarity of similar products by using multimodal product display clarity to generate product first dimension comparison data.

[0009] The present invention ensures a comprehensive evaluation of product display by defining clarity indexes for text, image and video features. Each modal feature is quantified from multiple dimensions, such as the language conciseness of the text, the resolution and detail presentation of the image, and the frame rate and stability of the video, which enables the system to accurately grasp various quality factors when displaying the product, thereby improving the user's understanding and recognition of the product display. By calculating the number of modal features of the product display and judging whether the product is a single modal feature product or a multi-modal feature product based on the number, it is helpful to classify the complexity of the product display. In the processing of multi-modal feature products, the system can combine the advantages of different display forms to provide a richer display experience, while single-modal products simplify the display content by eliminating them. This classification ensures that the system can perform targeted optimization according to the complexity of the modal features when processing the product. By weighted fusion of the scores of multi-modal products, a comprehensive score can be generated according to the clarity index of each modal feature. This weighted fusion ensures that the display of the product can find a balance between different modes, enhancing the globality and coordination of the product display. For example, high-quality display of images and videos can make up for the shortcomings of text display, and vice versa, thereby improving the overall user experience. By evaluating products based on the clarity of multimodal product display, it is possible to compare the clarity of similar products and provide users with more accurate product selection. Through this evaluation, the system can not only optimize the product display effect, but also improve the comparability between products, helping users make more rational purchasing decisions.

[0010] Preferably, step S2 comprises the following steps: Step S21: Obtain product description parameters; Step S22: Performing three-dimensional modeling of the multimodal feature product using the product description parameters based on the image features of the product to generate a three-dimensional model for product display; Step S23: optimizing the commodity display 3D model based on the video features of the commodity to generate a commodity display 3D optimized model; Step S24: Evaluate the accuracy of product display of the three-dimensional optimization model of product display according to the text features of the product, and generate evaluation data of the second dimension of the product.

[0011] By acquiring the commodity description parameters, the system can obtain basic information and detailed characteristics about the commodity, and provide the necessary data basis for subsequent multimodal processing. These parameters usually include key information such as size, material, function, etc., which can provide accurate guidance for three-dimensional modeling and ensure the true restoration of commodity display. Three-dimensional modeling based on commodity image features and description parameters can present the commodity as a three-dimensional visualization model. This step integrates the image and other attributes of the commodity (such as size and proportion) to ensure that the three-dimensional sense and details of the commodity can be displayed in the virtual display, thereby providing a more realistic and accurate display effect, which helps users better understand the appearance and function of the commodity when making a purchase decision. Optimizing the three-dimensional model based on the video features of the commodity can improve the dynamic effect of the commodity display. Video features such as frame rate, stability and dynamic and static contrast help the system optimize the sense of motion and dynamic performance in the commodity display, ensuring that the commodity is more smooth and natural in the visual display. This optimization can enhance the user's viewing experience and make the commodity display more attractive and interactive. By evaluating the accuracy of the 3D optimization model based on the product text features, the system can verify the accuracy of the product display content. This evaluation ensures that the product display is not only visually realistic, but also matches the product description. For example, whether the size, color, function, etc. of the product are consistent with the text description, thereby ensuring the authenticity and accuracy of the product display and improving users' trust in the display.

[0012] Preferably, step S23 includes the following steps: Step S231: performing three-dimensional spatial feature mapping on the video features of the product, thereby obtaining spatial mapping data; and reconstructing the model surface details of the three-dimensional model of the product display according to the spatial mapping data, thereby obtaining model detail data; Step S232: constructing a three-dimensional topological structure of the commodity display three-dimensional model through the model detail data, thereby obtaining model structure data; performing texture material mapping on the commodity display three-dimensional model according to the model structure data, thereby obtaining texture mapping data; Step S233: optimizing the lighting and shadow of the texture mapping data to obtain lighting and shadow effect data; performing dynamic effect rendering on the commodity display three-dimensional model according to the lighting and shadow effect data to obtain dynamic rendering data; Step S234: Utilize the dynamic rendering data to optimize the three-dimensional model of the commodity display to generate a three-dimensional optimized model of the commodity display.

[0013] The present invention performs three-dimensional spatial feature mapping on the video features of commodities, and the system can obtain the dynamic performance of commodities in three-dimensional space, which not only provides data support for the static display of commodities, but also can reconstruct the surface details of the three-dimensional model of commodity display according to these data, so as to present a more delicate and real commodity appearance. This step can enhance the visual layering of commodity display, making it easier for users to capture details when viewing commodities, and enhancing the attractiveness of display. By constructing the topological structure of the three-dimensional model of commodity display, the system can ensure the rationality and detailed performance of the structure of the commodity surface in three-dimensional space. At the same time, the texture material is mapped based on the model structure data, so that the appearance of the commodity is more in line with reality, reflecting the texture and sophistication of the material. This texture mapping optimization can significantly improve the authenticity of commodity display, making the appearance of the commodity more attractive and persuasive. By optimizing the lighting and shadow of the texture mapping data, the system can enhance the stereoscopic sense and depth of commodity display, so that the effect of commodity display is more layered and spatial. In addition, light and shadow optimization helps to present the real effect of commodities under different lighting conditions and improve the visual expressiveness. Based on the light and shadow effect data, the system performs dynamic effect rendering, making the product display not only exquisite in a static state but also smooth and natural in a dynamic display, thus bringing a more immersive experience to users. After the above steps are completed, the dynamic rendering data is used to optimize the three-dimensional model of product display. The system can further adjust and improve the various parameters of the model (such as lighting, texture, structure, etc.), thereby generating the final three-dimensional optimized model of product display. This optimized model ensures that the product can maintain the best performance in each display link, and can provide the best visual experience in both static and dynamic displays.

[0014] Preferably, step S24 includes the following steps: Step S241: performing part-of-speech classification on the text features of the product to generate product text part-of-speech classification data, wherein the product text part-of-speech classification data includes product noun part-of-speech and product adjective part-of-speech; Step S242: structurally matching the three-dimensional optimization model of commodity display by commodity noun parts of speech to generate a three-dimensional matching structure of commodity display; calculating actual structural parameters of the three-dimensional matching structure of commodity display according to commodity adjective parts of speech to obtain actual structural parameters; Step S243: using the actual structural parameters to traverse the three-dimensional matching structure of similar product display in the three-dimensional model of product display, thereby obtaining the actual structural parameter set of similar products; evaluating the product display accuracy of the actual structural parameter set of similar products through the actual structural parameters, and generating product second dimension evaluation data.

[0015] The present invention classifies the commodity text features by part of speech, and the system can extract key information such as nouns and adjectives of the commodity. This process helps to extract the core features of the commodity from the commodity description, such as the name, function, characteristics, etc. of the commodity. This information provides a basis for subsequent three-dimensional model matching and parameter calculation, which makes the commodity display process more accurate, closely combined with the actual description of the commodity, and improves the relevance of the display. By performing structural matching on the commodity display three-dimensional optimization model by commodity noun part of speech, the system can adjust the three-dimensional model according to the core features of the commodity to ensure that the structure of the model is consistent with the commodity description. Further, the actual structural parameter calculation of the model is performed according to the commodity adjective part of speech to ensure that the details of the commodity display (such as size, proportion, function, etc.) are accurate. In this way, the three-dimensional model of the commodity can perfectly reflect the actual characteristics of the commodity and improve the accuracy and reliability of the display. By traversing the three-dimensional matching structure of similar commodity displays and comparing it with the actual structural parameter set, the system can extract the most representative structural features from multiple similar commodities. This method helps the system to perform more extensive comparison and evaluation, ensuring that the three-dimensional model of each commodity display is not only accurate but also in line with the actual situation. By comparing these similar products, the generated product display accuracy evaluation data can provide users with more reliable product information and reduce the difference between the displayed and actual products.

[0016] Preferably, step S3 comprises the following steps: Step S31: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; Step S32: predicting the commodity purchase conversion rate based on the commodity authenticity data to generate commodity purchase conversion rate prediction data; Step S33: dynamically display and adjust the similar product screening data according to the product purchase conversion rate prediction data to generate product dynamic display adjustment data.

[0017] By combining the first dimension evaluation data of the product (such as display clarity, matching) and the second dimension evaluation data of the product (such as display accuracy, structural matching), the system can comprehensively calculate the authenticity of the product. This step ensures a high degree of consistency between the product display and the actual product, thereby improving the credibility of the product. The calculated product authenticity data helps the system to screen out the products that best meet the user's expectations, reduce the deviation between the display and the actual product, and increase the trust in the product display. Using the product authenticity data, the system can predict the purchase conversion rate of the product. This process can help the system understand which product display methods are more likely to prompt users to buy, and which factors (such as display clarity, accuracy, etc.) have a greater impact on the user's purchase decision. Through this prediction, merchants or platforms can understand the market performance of the product in advance, make targeted marketing strategy adjustments, and improve user conversion rates. According to the product purchase conversion rate prediction data, the system can dynamically adjust the display of similar products. By adjusting the display method (such as the display position, display mode, display order, etc. of the product), the system can optimize the product display according to the predicted conversion rate during the actual display process, ensuring that users get more attractive product recommendations during the shopping process, thereby enhancing users' purchasing interest and purchasing behavior.

[0018] Preferably, step S32 includes the following steps: Step S321: extracting product purchase features from similar product screening data based on product authenticity data to obtain product purchase feature data; dividing the product purchase feature data into data sets to generate a model training set and a model test set; Step S322: Perform model training on the model training set by using a regression analysis algorithm to generate a commodity purchase conversion rate prediction pre-model; perform model optimization iteration on the commodity purchase conversion rate prediction pre-model by using a model test set to generate a commodity purchase conversion rate prediction model; Step S323: Import the commodity authenticity data into the commodity purchase conversion rate prediction model to perform commodity purchase conversion rate prediction and generate commodity purchase conversion rate prediction data.

[0019] The present invention extracts purchase features from similar product screening data based on product authenticity data, and the system can capture key factors that affect user purchase decisions, such as the authenticity, display method, price, function, etc. of product display. The extracted product purchase feature data can help the model accurately identify which factors are related to purchase behavior, thereby improving the effect of the prediction model. Further, the system divides these feature data into training sets and test sets to ensure that the model can be trained and tested in different data environments, thereby effectively avoiding overfitting problems and ensuring the generalization ability of the prediction results. By training the model training set through a regression analysis algorithm, the system can generate a prediction pre-model for the product purchase conversion rate. The role of regression analysis in this step is to identify the relationship between purchase features and conversion rates and find out the most critical factors affecting purchase behavior. After the model training is completed, the pre-model is optimized and iterated using the model test set to gradually improve the accuracy and reliability of the model. Through repeated optimization, the system can generate a more accurate product purchase conversion rate prediction model, so that the prediction results are more in line with actual purchase behavior, thereby improving the accuracy of decision-making. Importing product authenticity data into the trained product purchase conversion rate prediction model to make specific purchase conversion rate predictions, the generated product purchase conversion rate prediction data can provide merchants or platforms with clear market expectations, help them understand the potential conversion capabilities of different products, and adjust display strategies or promotional activities based on the prediction results. Accurate purchase conversion rate predictions can not only help merchants optimize product display methods, but also provide users with more personalized product recommendations and increase purchase rates.

[0020] Preferably, step S4 comprises the following steps: Step S41: collecting similar product evaluations on similar product screening data according to the product dynamic display adjustment data to obtain similar product evaluation collection data; evaluating the product feedback dimension on the similar product evaluation collection data to generate product third dimension evaluation data; Step S42: Based on the third dimension evaluation data of the product, the similar product screening data is compared and visualized in a comprehensive dimension, so as to generate a comprehensive dimension comparison visualization report of similar products.

[0021] The present invention evaluates and collects similar product screening data according to the dynamic display adjustment data of the product. The system can obtain more information about product display from user feedback, including user preferences, product acceptance, effectiveness of display methods, etc. These evaluation data provide a practical basis for further evaluating the display effect of the product. Then, the system evaluates the product feedback dimension of these similar product evaluation data and generates the third dimension evaluation data of the product. The third dimension evaluation data helps the system understand the actual impact of product display on users, including the attractiveness of the product, user satisfaction, etc. This process makes the product display more in line with user needs, reduces invalid display, and improves the display effect. Based on the third dimension evaluation data of the product, the system can compare the similar product screening data in a comprehensive dimension and generate a comprehensive dimension comparison visualization report of similar products. This report clearly shows the comparison results of each product in multiple dimensions through charts, images or other visualization forms, such as display clarity, user feedback, purchase conversion rate, etc. This visualization report can help merchants, platforms or users intuitively understand the display effects of different products and identify the most popular or attractive products. Merchants or platforms can adjust product display strategies based on these visualization data to improve the accuracy of display.

[0022] In this specification, a system for multi-dimensional comparison of smart products based on multimodal analysis is provided, which is used to execute the above-mentioned method for multi-dimensional comparison of smart products based on multimodal analysis. The system for multi-dimensional comparison of smart products based on multimodal analysis includes: The display evaluation module is used to obtain the user input text; based on the user input text, similar products are screened in the preset product database to generate similar product screening data; the product display clarity evaluation is performed on the similar product screening data to generate the first dimension evaluation data of the product; The accuracy evaluation module is used to obtain product description parameters; perform three-dimensional modeling of multimodal feature products using product description parameters based on product image features to generate a three-dimensional model for product display; perform product display accuracy evaluation on the three-dimensional model for product display to generate product second-dimensional evaluation data; A display adjustment module is used to calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; The feedback evaluation module is used to evaluate the product feedback dimension of similar product screening data based on the product dynamic display adjustment data, and generate the product third dimension evaluation data; based on the product third dimension evaluation data, the similar product screening data is compared and visualized in a comprehensive dimension, thereby generating a similar product comprehensive dimension comparison visualization report.

[0023] The beneficial effect of the present invention is that based on the text input by the user, the system can quickly filter out related or similar commodities from the commodity database, thereby improving the relevance and accuracy of commodity display. The display clarity of similar commodities is evaluated, and the effect of commodity display is judged from multiple dimensions (such as text, image, video, etc.), providing data support for subsequent adjustments. This process can ensure that the display content is concise and clear, meets user needs, improves the display quality, and reduces the cognitive burden of users. Through three-dimensional modeling based on commodity image features and description parameters, the system can present detailed information of commodities in virtual space. Then, the display effect of the model is optimized through accuracy evaluation to ensure that the display of commodities is consistent with the actual product, which has a significant effect on improving the accuracy of commodity description and the authenticity of virtual display, and can effectively enhance the user's trust in the commodity, thereby increasing the willingness to buy. Through a comprehensive analysis of the commodity display effect (first dimension evaluation) and display accuracy (second dimension evaluation), the system can calculate the "authenticity" of the commodity, that is, the consistency of the display content with the actual commodity, and this data provides a quantitative basis for subsequent display adjustments. Dynamic adjustment of the display content based on commodity authenticity data can be optimized according to user feedback or behavior. This function ensures that products are always presented in a way that is most attractive to users and enhances purchasing intentions, thereby improving user experience. Based on the display adjustments of the products, the system will conduct dimensional evaluations of product feedback (such as user satisfaction, display effects, purchase intentions, etc.) to help merchants understand the impact of product display effects on user behavior, so as to make further optimizations. By visually displaying data in multiple dimensions, merchants or platforms can intuitively see the pros and cons of different products in each dimension and make optimization decisions accordingly. Visual reports help decision makers more clearly understand which display features are most attractive to users, further improving the overall display effect. Therefore, the present invention improves the comprehensiveness of product comparison by introducing multi-dimensional evaluation, dynamic display adjustment and visual reports. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of the steps of a method for multi-dimensional comparison of intelligent products based on multimodal analysis; Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart; Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG. The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0025] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0027] It should be understood that although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, and the term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0028] To achieve this, please refer to Figures 1 to 3 , a method for multi-dimensional comparison of smart commodities based on multimodal analysis, the method comprising the following steps: Step S1: obtaining user input text; screening similar products in a preset product database based on the user input text to generate similar product screening data; evaluating product display clarity on the similar product screening data to generate product first dimension evaluation data; Step S2: obtaining product description parameters; performing three-dimensional modeling of multimodal feature products using the product description parameters based on product image features to generate a three-dimensional product display model; evaluating the product display accuracy of the three-dimensional product display model to generate product second dimension evaluation data; Step S3: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; Step S4: Evaluate the product feedback dimension of the similar product screening data based on the product dynamic display adjustment data to generate product third-dimensional evaluation data; perform comprehensive dimensional comparison and visualization of the similar product screening data based on the product third-dimensional evaluation data, thereby generating a comprehensive dimensional comparison visualization report of similar products.

[0029] By obtaining the user's input text (such as search terms, preference descriptions, etc.), the system can intelligently understand the user's needs and accurately screen similar products, thereby providing personalized recommendations and avoiding the limitations of traditional methods of screening products by static conditions. Through dynamic screening, it is ensured that the displayed products match the user's needs, the relevance of product display is improved, and the time and energy of users browsing irrelevant products is reduced. Display clarity evaluation ensures that the displayed product images and description information are clear and accurate, so that users can quickly understand the characteristics and advantages and disadvantages of the products and improve the accuracy of purchase decisions. Converting products from two-dimensional plane displays to three-dimensional models can more intuitively present details such as the appearance, structure and size of the products, helping users to better understand the actual effects of the products, especially for complex products (such as electronic products, furniture, etc.). Product display accuracy evaluation ensures the consistency between the three-dimensional model and the actual product, avoids false or misleading displays, improves the user's trust in the product, and reduces returns and complaints caused by display distortion. Through three-dimensional modeling, users can perform interactive operations such as rotation and zooming on the product display page to enhance the shopping experience, making users feel as if they are experiencing the product in person and increase their desire to buy. Combining the evaluation of the first dimension (clarity) and the second dimension (accuracy), the system can conduct an in-depth analysis of the display of goods to ensure the authenticity and reliability of the displayed content and avoid false advertising or erroneous information affecting user judgment. According to the authenticity data of the goods, the system can dynamically adjust the display method of the goods (such as changing the product image, optimizing the description text, adjusting the display order, etc.), optimize the display effect according to real-time feedback, and ensure that the most suitable product display is continuously provided to users. By adjusting the product display in real time, it is possible to provide product information that better meets user expectations, reduce misleading and false displays, enhance user trust in the platform, and improve shopping conversion rates. By comprehensively evaluating the display effect, authenticity and user feedback of the goods, the system can comprehensively evaluate all dimensions of the goods, provide richer information, and help users make multi-angle decisions. The product feedback dimension evaluation combines user behavior data (such as click-through rate, purchase rate, evaluation, etc.) to reflect user feedback on product display in real time, further optimize product display strategy, and ensure that product display is more in line with user needs. The generated visualization report comprehensively compares various dimensions of products (such as clarity, accuracy, authenticity, feedback, etc.), intuitively presents the advantages and differences between products, helps users make comparisons easily, and improves shopping decision efficiency. Therefore, the present invention improves the comprehensiveness of product comparison by introducing multi-dimensional evaluation, dynamic display adjustment and visualization report.

[0030] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a schematic diagram of a method for multi-dimensional comparison of smart products based on multimodal analysis of the present invention. In this example, the method for multi-dimensional comparison of smart products based on multimodal analysis includes the following steps: Step S1: obtaining user input text; screening similar products in a preset product database based on the user input text to generate similar product screening data; evaluating product display clarity on the similar product screening data to generate product first dimension evaluation data; In an embodiment of the present invention, the user uses a natural language processing (NLP) toolkit (such as NLTK, spaCy, Hugging Face) to perform word segmentation, stop word removal, part-of-speech tagging, etc. on the user input text. Extract keywords or phrases from the text and convert them into vector form using algorithms such as TF-IDF and Word2Vec. The product description text is converted into a vector representation through an NLP model (such as BERT, Word2Vec), and the similarity between each product and the user input text is calculated. Select the product that is most similar to the user input for display. Use a convolutional neural network (CNN) to evaluate the clarity of the product image and automatically mark whether it meets the clarity standard. Analyze the completeness and accuracy of the product description text based on rules or deep learning models. The pre-trained text classification model can be used to classify the product description and mark whether there is missing information. Analyze the user evaluation data through sentiment analysis tools (such as VADER, TextBlob, etc.) to determine whether the product display is clear and credible. The evaluation results of multiple dimensions such as image quality, text integrity, and user evaluation are weighted and summarized to generate a product display clarity score (which can be a numerical value between 0-1 or other standardized score). Based on the score, decide whether to display the product at the front of the search results or whether the product display content needs to be optimized.

[0031] Step S2: obtaining product description parameters; performing three-dimensional modeling of multimodal feature products using the product description parameters based on product image features to generate a three-dimensional product display model; evaluating the product display accuracy of the three-dimensional product display model to generate product second dimension evaluation data; In the embodiment of the present invention, the description parameters of the commodity include the basic information, attributes (such as brand, type, color, size, material, etc.), function description, instructions for use, commodity specifications, etc. of the commodity. Detailed description information of these commodities can be obtained by crawling, database query or directly provided by commodity merchants. Commodity image data can extract features through computer vision technology. Common methods include: extracting features such as shape, texture, color distribution, object contour, etc. from commodity images using pre-trained CNN models (such as ResNet, VGG, Inception, etc.). More abstract visual features can be extracted from commodity images through automated deep learning models for subsequent three-dimensional modeling and display. By taking multiple photos of the commodity from different angles (such as front, back, left, right, top, etc.), these image data will be used as input for three-dimensional reconstruction. Noise removal and image enhancement (such as contrast adjustment, deblurring, etc.) are performed on the image to improve the quality of three-dimensional modeling. Using two or more commodity images with different perspectives, the three-dimensional structure of the commodity is restored through stereoscopic vision algorithms (such as SIFT, SURF, ORB and other feature matching methods). Use deep learning-based 3D reconstruction methods (such as multi-view neural network MVSNet) to automatically reconstruct the 3D model of the product from multi-view images. Combine product description parameters (such as size, scale, material, etc.) to accurately adjust and optimize the 3D model. For example, "This product is a plastic shell" in the product description will affect the material characteristics, and thus affect the surface material settings of the 3D model. Use computer graphics technology (such as OpenGL, Unity, Blender and other 3D modeling tools) to render and optimize the 3D data generated by stereo vision or deep learning algorithms to form a display 3D model of the product. Compare the generated 3D model with the actual product image and calculate the similarity between the model and the original image (for example, using indicators such as structural similarity index SSIM and peak signal-to-noise ratio PSNR). Check the integrity and accuracy of the 3D model by observing from different angles to ensure that the model details at each angle are clearly presented. Check whether the geometric shape of the 3D model is consistent with the actual size and shape of the product. This can be done by comparing it with the product's specification parameters (such as the size provided in the product description). Ensure that the surface texture of the 3D model is consistent with the actual texture in the product image, including color, material, etc., to determine whether it can accurately reflect the appearance of the product. Simulate the user's interactive experience with the 3D display model to ensure that the product model can be displayed smoothly without distortion or unreality when the user rotates, zooms, etc. Conduct user experience testing and evaluate the accuracy of the 3D model display through user feedback, including whether the model is easy to identify the details of the product. Based on the above evaluation method, comprehensively calculate the accuracy of the 3D model of the product display. A comprehensive score can be generated based on multiple dimensions such as geometric matching, image contrast, and user feedback (for example, a value of 0-1, the closer to 1, the higher the accuracy).Through weighted evaluation, factors such as image quality, geometric accuracy, and user feedback are integrated to ultimately arrive at a product display accuracy score.

[0032] Step S3: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; In an embodiment of the present invention, for each commodity, the first dimension evaluation data and the second dimension evaluation data of the commodity are used to perform weighted calculation to obtain the authenticity score of the commodity. This can be achieved by the following formula: Ri=α⋅Fi+β⋅Si; Ri is the authenticity score of commodity i, Fi is the first dimension evaluation data of commodity i, Si is the second dimension evaluation data of commodity i, and α and β are weight coefficients, reflecting the influence of each dimension. According to the authenticity score, similar commodities are screened out, and the authenticity scores of these commodities can be grouped according to a certain threshold. A higher authenticity score means that the commodity has a higher credibility among similar commodities. Based on the authenticity data of the commodity, the order, highlighting, etc. of similar commodities on the display interface can be dynamically adjusted. For example, commodities with higher authenticity scores can be ranked in front, or their exposure can be increased by highlighting, adding recommended tags, etc. According to the authenticity of the commodity and user behavior data (such as click volume, purchase conversion rate, etc.), the display mode of the commodity is adjusted through real-time feedback. The following algorithm can be used for adjustment: Di=γ⋅Ri+δ⋅Bi; Di is the priority of product i in the display, Ri is the authenticity score of the product, Bi is the user behavior indicator of the product (such as click-through rate), and γ and δ are the weight coefficients of dynamic display. Adjust the order of products on the display interface according to the value of Di. According to the authenticity score of the product, decide whether to display more detailed information or use more attractive visual effects (such as recommendation boxes, shining labels, etc.). Dynamically adjust the display content based on real-time user behavior data and product ratings.

[0033] Step S4: Evaluate the product feedback dimension of the similar product screening data based on the product dynamic display adjustment data to generate product third-dimensional evaluation data; perform comprehensive dimensional comparison and visualization of the similar product screening data based on the product third-dimensional evaluation data, thereby generating a comprehensive dimensional comparison visualization report of similar products.

[0034] In an embodiment of the present invention, the feedback information of the user on the product is evaluated, such as the evaluation after purchase, the return rate, the user rating, the comment content, etc. Based on these feedback information, the quality, service, and user experience of the product can be comprehensively scored to obtain the feedback dimension score of the product. The following formula can be used: Fi=α⋅Ratingi+β⋅ReturnRatei+γ⋅Review Counti; Fi is the feedback dimension score of product i, Ratingi is the score of product i (usually 1-5 stars), Return Ratei is the return rate of the product, Review Counti is the number of reviews of the product, and α, β, γ are weight coefficients, reflecting the importance of different feedback dimensions. The third dimension evaluation data of the product is generated according to the feedback dimension score Fi of the product, and it is combined with the first dimension and second dimension evaluation data of the product. Based on the first dimension evaluation data, the second dimension evaluation data, and the third dimension evaluation data of the product, a multi-dimensional comparative analysis is performed. For example, a radar chart, a stacked bar chart, a scatter plot, etc. can be used to show the comparison between the dimensions and highlight the advantages and disadvantages of different products. Based on the results of comprehensive dimensional analysis, a comprehensive dimensional comparison visualization report of similar products is generated through data visualization tools (such as Tableau, Power BI, etc.). The report content includes: visual comparison of the first dimension, second dimension and third dimension of the product to help users or managers quickly understand the advantages and disadvantages of different products. Through comprehensive dimensional analysis, the performance of each product in multiple dimensions is displayed, and which products perform well in multiple dimensions are recommended. The trend of product feedback dimensions (such as ratings, number of comments, etc.) changing over time is displayed to help users understand the changes in the product's reputation in the market.

[0035] Preferably, step S1 comprises the following steps: Step S11: Obtain user input text; Step S12: performing a product keyword search on the user input text to obtain product keyword search data; screening similar products in a preset product database based on the product keyword search data to generate similar product screening data; Step S13: extracting product display modal features from the similar product screening data to obtain product display modal feature data, wherein the product display modal features include text features, image features, and video features; Step S14: Evaluate the display clarity of similar products based on the product display modal feature data to generate product first dimension evaluation data.

[0036] In an embodiment of the present invention, text information related to a product is provided through an interface or voice input, and the input text may include product name, description, brand, purpose, and the like. The system obtains the text through a text input interface and prepares for further processing. Based on the text input by the user, the system uses a keyword extraction algorithm (such as TF-IDF, BERT, etc.) to extract important keywords from the text. The extracted keywords are used to search the product database to find related products. This step involves natural language processing technology, which combines the product information in the product database to return product data related to the keywords. Similar products are screened out according to the search results, and similar products are screened through a matching algorithm (such as cosine similarity or collaborative filtering) to generate similar product screening data. These products have high similarity with the text content input by the user and meet user needs. After obtaining the similar product screening data, the system extracts modal features from the display data of each product (such as product pictures, videos, etc.). The product display modal features include: extracting information such as text descriptions and titles of products, extracting keywords through natural language processing, sentiment analysis, etc. Use computer vision technology (such as convolutional neural network CNN) to process product images and extract features such as color, shape, texture, and pattern. Use video analysis algorithms (such as 3D convolutional neural network, action recognition, etc.) to extract key frames, dynamic behavior patterns, user interactions, and other information from product videos. Finally, integrate these data into a product display modality feature dataset through multimodal feature fusion. Perform readability analysis on the text description of the product, including sentence structure, information completeness, etc. Perform quality assessment on product images and videos, such as resolution, clarity, contrast, etc. Use image quality assessment models (such as NIQE, FSIM, etc.) to generate product first dimension evaluation data, indicating the score or evaluation result of each product in terms of clarity, for subsequent screening and recommendation.

[0037] Preferably, step S14 includes the following steps: Step S141: defining the clarity index of commodity display modality features, wherein the text feature clarity index includes information coverage, language simplicity and accuracy, the image feature clarity index includes resolution score, detail presentation score and color restoration score, and the video feature clarity index includes frame rate analysis, dynamic and static contrast and picture stability score; Step S142: Calculate the number of commodity display modes for the commodity display modal feature data to obtain the number of commodity display modal features; determine whether the number of commodity display modal features is greater than 1. If the number of commodity display modal features is greater than 1, mark the commodity as a multi-modal feature commodity; otherwise, mark the commodity as a single modal feature commodity and remove it; Step S143: performing modal score weighted fusion on multi-modal feature products according to the product display modal feature clarity index to generate multi-modal product display clarity; Step S144: Evaluate the display clarity of similar products by using multimodal product display clarity to generate product first dimension comparison data.

[0038] In an embodiment of the present invention, the coverage of the text is evaluated by comparing the matching degree between the key information provided in the product description and the user's needs. For example, whether the product description describes key information such as functions, features, specifications, etc. in detail. Evaluate the conciseness of the product description, avoid lengthy or overly complex sentences, and ensure that the information is clearly conveyed and easy to understand. Analyze the accuracy of grammar, spelling, and semantics in the product description to check whether there is misleading or ambiguous. Analyze the resolution of the product image to ensure that the clarity of the image is adapted to different devices and display platforms, and avoid blurry or low-quality images. Evaluate the presentation of important details in the image, such as the appearance, material texture, and color of the product. Detect the accuracy of the image color to ensure that the product color in the image matches the actual product color to avoid color difference affecting the shopping experience. Evaluate whether the frame rate of the product video is smooth to ensure that there is no freeze or delay during video playback. Analyze whether the switching between dynamic and static scenes in the video is clear and smooth, and whether it can effectively display the functions and features of the product. Through the stability analysis algorithm, evaluate whether the video is stable to avoid affecting the viewing experience due to jitter or blur. Count the number of modal feature data for product display to determine how many modalities (text, image, video, etc.) are used for the product display. If the number of display modal features of a product is greater than 1, mark the product as a "multimodal feature product"; if there is only a single modal feature, mark the product as a "single modal feature product". For single modal feature products, remove them from further display and evaluation processes. For multimodal feature products, proceed to the next step of weighted fusion of modal scores. For each multimodal feature product, score the clarity of the text features, image features, and video features respectively, and use the predefined clarity index to calculate the score of each modality. Fusion the clarity scores of each modality by weighted averaging. For example, you can set weights (such as text feature weight 0.4, image feature weight 0.3, video feature weight 0.3), and weighted sum the clarity scores of each modality to obtain the final multimodal product display clarity. Use the calculated multimodal product display clarity to evaluate the display clarity of the products in the similar product screening data. By comparing the display clarity of different products, we generate comparative data of the first dimension of the products, that is, the ranking or score of the products in terms of display effect. This data will be used in subsequent recommendation algorithms, search result sorting or other product screening functions to ensure that products with higher display quality are shown to users first.

[0039] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: Obtain product description parameters; Step S22: Performing three-dimensional modeling of the multimodal feature product using the product description parameters based on the image features of the product to generate a three-dimensional model for product display; Step S23: optimizing the commodity display 3D model based on the video features of the commodity to generate a commodity display 3D optimized model; Step S24: Evaluate the accuracy of product display of the three-dimensional optimization model of product display according to the text features of the product, and generate evaluation data of the second dimension of the product.

[0040] In the embodiment of the present invention, the description information of each commodity is extracted from the commodity database, including parameters such as text description, specifications, materials, functions, etc. The description of the commodity is analyzed, and the description of the commodity is converted into structured parameters using natural language processing (NLP) technology (such as entity recognition, information extraction, etc.), and these parameters will include: commodity size: such as length, width, height, weight, etc. Commodity material: such as metal, plastic, wood, etc. Functional characteristics: such as waterproof, earthquake-resistant, etc. Other description information: such as color, brand, purpose, etc. Feature extraction is performed on the commodity image, which can extract the key features of the commodity image, such as shape, texture, color distribution, etc., through a deep learning model (such as a convolutional neural network CNN). According to the image features and description parameters of the commodity, a three-dimensional model of the commodity is created through multi-view modeling technology (such as structured light scanning, stereo vision, or single image reconstruction based on deep learning). The model will consider: Commodity appearance: based on the shape and size of the image. Detailed display: such as the edge, surface texture, logo, etc. of the commodity. Functional parameters: such as embedded buttons, switches, interface positions, etc. The three-dimensional model of the product generated through the above steps will fully demonstrate the appearance and function of the product. This model will be used for subsequent optimization and evaluation. Extract important visual features from the product video, including the performance of the product in different scenes, such as the appearance changes of the product under different angles and lighting, or the dynamic effects of product operations. Use video processing technology (such as optical flow analysis, key frame extraction, action recognition, etc.) to extract these features. Based on the actual lighting, reflection effects, dynamic textures, etc. in the video, optimize the surface texture of the three-dimensional model to ensure consistency with the actual product. According to the dynamic changes in the video (such as the folding, unfolding, rotation, etc. of the product), adjust the dynamic form of the three-dimensional model so that it has good performance under various viewing angles. By comparing the details in the video with the original three-dimensional model, optimize the edges, contours and details of the object. The optimized three-dimensional model is more realistic and accurate, and can reflect the real appearance and dynamic display effect of the product. Extract keywords in the product text description, such as product function, size, material, etc., and analyze the text information of the product through NLP technology (such as BERT, Word2Vec, etc.). Evaluate the display effect of the model by comparing the accuracy of the three-dimensional optimized model of the product with the text description. Specific evaluation indicators include: Compare the product parameters in the text description (such as size, weight, function, etc.) with the parameters in the 3D model and calculate the matching degree. Analyze whether the appearance of the 3D model is consistent with the text description of the product, and ensure that the model matches the appearance, color, texture and other features in the description. Verify whether the 3D model accurately displays the functional characteristics of the product (such as buttons, interfaces, operation methods, etc.). Generate a product display accuracy score based on the accuracy evaluation results. This data will be used to compare the display effects of different products to ensure that the final displayed product 3D model has high accuracy.

[0041] Preferably, step S23 includes the following steps: Step S231: performing three-dimensional spatial feature mapping on the video features of the product, thereby obtaining spatial mapping data; and reconstructing the model surface details of the three-dimensional model of the product display according to the spatial mapping data, thereby obtaining model detail data; Step S232: constructing a three-dimensional topological structure of the commodity display three-dimensional model through the model detail data, thereby obtaining model structure data; performing texture material mapping on the commodity display three-dimensional model according to the model structure data, thereby obtaining texture mapping data; Step S233: optimizing the lighting and shadow of the texture mapping data to obtain lighting and shadow effect data; performing dynamic effect rendering on the commodity display three-dimensional model according to the lighting and shadow effect data to obtain dynamic rendering data; Step S234: Utilize the dynamic rendering data to optimize the three-dimensional model of the commodity display to generate a three-dimensional optimized model of the commodity display.

[0042] In an embodiment of the present invention, by extracting dynamic features such as motion, rotation, and scaling from the video of the product, computer vision technology (such as optical flow analysis, motion detection, etc.) is used to analyze the product in the video frame by frame to obtain its spatial position, angle, size, and other information. The spatial features in the video (such as the motion trajectory and relative position of the product in three-dimensional space) are mapped to the three-dimensional model. Multi-view geometric reconstruction or stereoscopic vision technology can be used, combined with the depth information in the video, to convert the dynamic spatial features of the product into spatial mapping data through a three-dimensional reconstruction algorithm. This spatial mapping data will describe the morphological changes of the product in three-dimensional space and can reflect the spatial position and posture of the product under different viewing angles. According to the spatial mapping data, the surface details of the three-dimensional model of the product are reconstructed. This step will add the detailed features of the product, such as texture, surface gloss, cracks, scratches, etc., to the three-dimensional model. Image-based texture mapping technology (such as UV mapping) can be used to fuse the detail information captured in the video to the three-dimensional surface to ensure that the details of the product are presented more realistically. This step will obtain reconstruction data containing the surface details of the product as the basis for further optimization of the model. Based on the model detail data, the topological structure of the three-dimensional model of the product is constructed. Topological structure refers to the connection relationship and shape of each part in the 3D model, ensuring that the model has the correct geometric structure. The 3D model is meshed through algorithms (such as Marching Cubes or Poisson Surface Reconstruction), and the surface details of the product are encrypted and refined in the process to optimize the surface quality. According to the topological structure of the model, texture material mapping is performed. In this process, each face of the model will be textured according to its geometric shape so that realistic material effects can be presented in the subsequent rendering process. Texture mapping data will include information such as UV coordinates, reflectivity, transparency, glossiness, etc. of the image texture, which can help accurately reproduce the material and surface details of the product in the 3D model. Map the texture and material information to the 3D model of the product to obtain complete texture mapping data, preparing for subsequent lighting optimization and dynamic effect rendering. Use lighting models (such as Phong model, Blinn-Phong model or PBR (Physical Based Rendering) model) to perform lighting calculations on texture mapping data. Lighting calculations include the effects of different light sources such as ambient light, scattered light, and specular light on the model. Based on the lighting calculations, the shadow effect is optimized through algorithms. Shadow mapping or ray tracing technology can be used to calculate the occlusion effect between objects and light sources, thereby generating a realistic shadow effect and improving the visual effect of the model. The above steps generate light and shadow effect data, which includes the performance of the products in the scene under different lighting conditions, such as shadow depth, light reflection degree, etc. Based on the light and shadow effect data, the three-dimensional model of the product is dynamically rendered.During this process, the rotation, scaling and other dynamic changes of the product are taken into account, and each frame of the image is rendered so that the model presents natural light and shadow effects in dynamic scenes. By rendering the dynamic effects, dynamic rendering data is obtained, which will show the appearance of the product under different lighting and dynamic states. According to the dynamic rendering data, the three-dimensional model of the product display is optimized. The optimization content includes: according to the rendering effect, further improving the detail accuracy of the model, such as adding surface texture, improving shadow quality, etc. Optimizing the computational efficiency in the rendering process to ensure that the rendering can be displayed in real time under multiple perspectives, suitable for interactive experience. Through the above optimization steps, the final three-dimensional optimized model of product display is generated. This model has high-precision detail presentation, realistic light and shadow effects, and smooth dynamic effects, which is suitable for a variety of application scenarios such as display, interaction and virtual trial.

[0043] Preferably, step S24 includes the following steps: Step S241: performing part-of-speech classification on the text features of the product to generate product text part-of-speech classification data, wherein the product text part-of-speech classification data includes product noun part-of-speech and product adjective part-of-speech; Step S242: structurally matching the three-dimensional optimization model of commodity display by commodity noun parts of speech to generate a three-dimensional matching structure of commodity display; calculating actual structural parameters of the three-dimensional matching structure of commodity display according to commodity adjective parts of speech to obtain actual structural parameters; Step S243: using the actual structural parameters to traverse the three-dimensional matching structure of similar product display in the three-dimensional model of product display, thereby obtaining the actual structural parameter set of similar products; evaluating the product display accuracy of the actual structural parameter set of similar products through the actual structural parameters, and generating product second dimension evaluation data.

[0044] In an embodiment of the present invention, relevant feature information is extracted from the description text of the commodity. The commodity description text generally includes information such as the commodity name, characteristics, use, material, and brand. The text is analyzed by natural language processing (NLP) technology, and each word in the text is marked as its grammatical category (such as noun, verb, adjective, etc.) by using part-of-speech tagging. According to the results of part-of-speech tagging, the nouns (such as "mobile phone", "television") and adjectives (such as "smart", "high-definition") in the commodity description are classified to generate commodity text part-of-speech classification data, including: core nouns indicating commodities (such as "mobile phone case", "laptop computer", etc.), which are used to describe the basic type of commodities. Indicates the characteristics or attributes of commodities (such as "durable", "high performance", etc.), which are used to describe the characteristics and functions of commodities. Based on the noun part-of-speech data of the commodity, the commodity display three-dimensional optimization model is structurally matched. For example, the overall structure and appearance characteristics (such as the outline, size, and design of the shoe) of the commodity display three-dimensional model are determined by commodity nouns (such as "shoes"). The commodity display three-dimensional optimization model is matched with the structure of the actual commodity through a model matching algorithm (such as geometric matching, model alignment, etc.) to ensure that the display model is consistent with the appearance of the real commodity. Through the process of structural matching, a three-dimensional matching structure of product display is generated, which contains the comparison information between the three-dimensional model and the actual product structure, such as the shape, size, and proportion of the model. The actual structural parameters of the three-dimensional matching structure of product display are calculated using the adjective parts of speech (such as "exquisite" and "lightweight") in the product description. This step calculates the actual parameters related to the product model through the description of the adjective parts of speech, such as the weight, material, and appearance texture of the object. Using the actual structural parameters, the three-dimensional matching structures of similar product displays in the three-dimensional model library of product display are traversed. This process finds the closest product model by comparing the structural similarities (such as shape, size, texture, etc.) between different products. Through the traversal process, all three-dimensional matching structures of product displays similar to the target product display three-dimensional model are collected, and the actual structural parameter sets of similar products are generated based on the actual structural parameters. These parameter sets contain the structural data of all products similar to the target product display three-dimensional optimization model. The actual structural parameter sets of similar products are used to evaluate the accuracy of product display. The evaluation criteria include the matching degree between the product model and the actual product, the accuracy of details, and the realism of the material. The evaluation results can be obtained through error analysis, matching calculation or other similarity evaluation algorithms (such as Euclidean distance, cosine similarity, etc.). Based on the evaluation results, the second dimension evaluation data of the product is generated, which reflects the accuracy and realism of the three-dimensional model of the product display. The evaluation results can provide data support for the optimization of product display, ensuring that the product display is more realistic and accurate.

[0045] As an example of the present invention, refer to Figure 3As shown, in this example, step S3 includes: Step S31: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; Step S32: predicting the commodity purchase conversion rate based on the commodity authenticity data to generate commodity purchase conversion rate prediction data; Step S33: dynamically display and adjust the similar product screening data according to the product purchase conversion rate prediction data to generate product dynamic display adjustment data.

[0046] In an embodiment of the present invention, the first dimension evaluation data of the product in step S14 (i.e., the product display clarity evaluation data) and the second dimension evaluation data of the product in step S24 (i.e., the product display accuracy evaluation data) are integrated. The two represent different dimensions of product display. The first dimension focuses on the clarity of display and the multimodal fusion effect, while the second dimension focuses on the authenticity and precise matching of the product. A product authenticity calculation model is constructed, which combines the evaluation data of the two dimensions and calculates the authenticity of the product through weighted average or machine learning algorithms (such as regression analysis, decision trees, or neural networks). The goal of the authenticity calculation is to measure the matching degree and simulation degree between the displayed product and the actual product. The formula can be used: product authenticity = w1 ⋅ product display clarity + w2 ⋅ product display accuracy; wherein w1 and w2 are weights set according to actual needs, reflecting the different importance of display clarity and display accuracy to product authenticity. Based on the above calculation, the product authenticity data of each product is obtained, which measures the similarity between the product display and the actual product. The product authenticity data provides an important basis for the next step of purchase conversion rate prediction. Construct a product purchase conversion rate prediction model, which predicts the purchase conversion rate of the product based on the product authenticity data, product display effect and historical purchase data. The conversion rate prediction model includes multiple features, such as product authenticity, display clarity, brand influence, market demand, etc. Logistic regression, support vector machine (SVM) or deep learning model can be used for prediction. The input of the model includes product authenticity data and other related features, and the output is the expected purchase conversion rate of the product. In the prediction model, the authenticity data needs to be standardized or normalized to ensure the range consistency of the data. At the same time, other features that affect the purchase conversion rate, such as product category, user behavior data, promotional information, etc., need to be considered. Using historical purchase data can help the training model better understand which product display features have a strong correlation with purchase conversion. Based on the above model, the purchase conversion rate is predicted for each product to generate product purchase conversion rate prediction data. This data shows the expected conversion rate for each product, which is usually a value between 0 and 1, indicating the probability of the product being purchased. According to the product purchase conversion rate prediction data, a priority is assigned to each product in the similar product screening data. Products with higher conversion rates are more likely to be purchased by users, so they can be displayed first. Build a dynamic display adjustment algorithm that uses the purchase conversion rate prediction data of products to adjust the order and display of products in the display according to the conversion rate. For example, products with high conversion rates can be placed in a prominent position on the page (such as the homepage, top, or recommendation column), while products with low conversion rates can be hidden or displayed in less user-attracted areas.Based on the above adjustment algorithm, the order and display style (such as pictures, texts, videos, etc.) of product display are dynamically adjusted to generate dynamic product display adjustment data, which contains the real-time adjustment results of product display, including the adjusted display order, display method, display location, etc. The display adjustment can be continuously optimized based on real-time user feedback (such as click volume, browsing time, purchase behavior, etc.), so as to display products that better meet users' needs in the future.

[0047] Preferably, step S32 includes the following steps: Step S321: extracting product purchase features from similar product screening data based on product authenticity data to obtain product purchase feature data; dividing the product purchase feature data into data sets to generate a model training set and a model test set; Step S322: Perform model training on the model training set by using a regression analysis algorithm to generate a commodity purchase conversion rate prediction pre-model; perform model optimization iteration on the commodity purchase conversion rate prediction pre-model by using a model test set to generate a commodity purchase conversion rate prediction model; Step S323: Import the commodity authenticity data into the commodity purchase conversion rate prediction model to perform commodity purchase conversion rate prediction and generate commodity purchase conversion rate prediction data.

[0048] In an embodiment of the present invention, similar commodities are screened based on commodity authenticity data. These commodities can be selected by various similarity indicators (such as price range, brand category, commodity attributes, etc.) to ensure that the selected commodities are representative. Relevant purchase feature data, such as purchase frequency, user evaluation, commodity image quality, whether the description is complete, participation in promotional activities, etc., are extracted from the selected commodities. These feature data will directly affect the purchase conversion rate of the commodities. All extracted commodity purchase feature data are divided into two parts according to a certain ratio (such as 70%:30%): a training set and a test set. The training set is used for model training, and the test set is used for later model optimization and verification. The training set data is trained using a regression analysis algorithm (such as linear regression, logistic regression, decision tree regression, etc.) to construct a commodity purchase conversion rate prediction model. This model predicts the conversion rate based on the input commodity purchase feature data. The model test set is used to verify the initially trained prediction model, and the error between the prediction result and the actual purchase conversion rate is adjusted and optimized. The optimization process involves parameter adjustment, algorithm selection (such as gradient boosting regression, support vector machine regression, etc.), and feature selection and reconstruction. The model performance is further optimized through cross-validation and other methods to ensure that the model has good generalization ability and avoid overfitting. The new product authenticity data is input into the trained product purchase conversion rate prediction model. The data set includes the latest authenticity score, description update, promotion information, etc. of the product. The model predicts based on the input product feature data and outputs the purchase conversion rate prediction data of the product. This prediction data can provide valuable decision support for marketers and sales personnel and optimize product promotion strategies.

[0049] Preferably, step S4 comprises the following steps: Step S41: collecting similar product evaluations on similar product screening data according to the product dynamic display adjustment data to obtain similar product evaluation collection data; evaluating the product feedback dimension on the similar product evaluation collection data to generate product third dimension evaluation data; Step S42: Based on the third dimension evaluation data of the product, the similar product screening data is compared and visualized in a comprehensive dimension, so as to generate a comprehensive dimension comparison visualization report of similar products.

[0050] In the embodiment of the present invention, in the process of displaying goods, based on the dynamic display adjustment data of goods, the system should collect the user's interaction and evaluation data on the goods. These data may include: whether the user clicks on the displayed goods, the time the user stays on the product display page, reflecting the attractiveness of the product display, whether the user purchases according to the displayed goods, and the user's feedback on the goods, including text comments and star ratings, such as whether operations such as adding to the shopping cart, collecting or sharing are performed. These evaluation data are collected from multiple channels (such as website click behavior, social media sharing, user ratings, etc.), and matched with the display data of similar goods to form similar goods evaluation collection data. The collected similar goods evaluation collection data are analyzed and evaluated to obtain product feedback dimension evaluation data. The evaluation dimension includes: analyzing according to the ratings and comments to reflect the satisfaction of the goods, and evaluating the attractiveness of the goods by comparing the click-through rate and purchase conversion, including user evaluation, comment quality and interaction frequency. The comments can be classified using sentiment analysis technology, and the proportion of positive feedback and negative feedback can be further determined to quantify the feedback quality of the goods. Based on the above evaluation data, the third dimension evaluation data of the goods is generated, which quantifies the performance of the goods in user interaction. Based on the third dimension evaluation data of the product, a comprehensive comparison is conducted on the screening data of similar products. The dimensions include: the display effect of the product (provided by the first and second dimension evaluation data), user feedback (assessing the satisfaction and interaction quality of the product through the third dimension evaluation data), purchase conversion rate, and by comparing the performance of each product in multiple dimensions, the comprehensive score of each product is calculated to reflect its comprehensive performance in all dimensions. The comprehensive dimension comparison results of the products are presented through a visual report.

[0051] In this specification, a system for multi-dimensional comparison of smart products based on multimodal analysis is provided, which is used to execute the above-mentioned method for multi-dimensional comparison of smart products based on multimodal analysis. The system for multi-dimensional comparison of smart products based on multimodal analysis includes: The display evaluation module is used to obtain the user input text; based on the user input text, similar products are screened in the preset product database to generate similar product screening data; the product display clarity evaluation is performed on the similar product screening data to generate the first dimension evaluation data of the product; The accuracy evaluation module is used to obtain product description parameters; perform three-dimensional modeling of multimodal feature products using product description parameters based on product image features to generate a three-dimensional model for product display; perform product display accuracy evaluation on the three-dimensional model for product display to generate product second-dimensional evaluation data; A display adjustment module is used to calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; The feedback evaluation module is used to evaluate the product feedback dimension of similar product screening data based on the product dynamic display adjustment data, and generate the product third dimension evaluation data; based on the product third dimension evaluation data, the similar product screening data is compared and visualized in a comprehensive dimension, thereby generating a similar product comprehensive dimension comparison visualization report.

[0052] The beneficial effect of the present invention is that based on the text input by the user, the system can quickly filter out related or similar commodities from the commodity database, thereby improving the relevance and accuracy of commodity display. The display clarity of similar commodities is evaluated, and the effect of commodity display is judged from multiple dimensions (such as text, image, video, etc.), providing data support for subsequent adjustments. This process can ensure that the display content is concise and clear, meets user needs, improves the display quality, and reduces the cognitive burden of users. Through three-dimensional modeling based on commodity image features and description parameters, the system can present detailed information of commodities in virtual space. Then, the display effect of the model is optimized through accuracy evaluation to ensure that the display of commodities is consistent with the actual product, which has a significant effect on improving the accuracy of commodity description and the authenticity of virtual display, and can effectively enhance the user's trust in the commodity, thereby increasing the willingness to buy. Through a comprehensive analysis of the commodity display effect (first dimension evaluation) and display accuracy (second dimension evaluation), the system can calculate the "authenticity" of the commodity, that is, the consistency of the display content with the actual commodity, and this data provides a quantitative basis for subsequent display adjustments. Dynamic adjustment of the display content based on commodity authenticity data can be optimized according to user feedback or behavior. This function ensures that products are always presented in a way that is most attractive to users and enhances purchasing intentions, thereby improving user experience. Based on the display adjustments of the products, the system will conduct dimensional evaluations of product feedback (such as user satisfaction, display effects, purchase intentions, etc.) to help merchants understand the impact of product display effects on user behavior, so as to make further optimizations. By visually displaying data in multiple dimensions, merchants or platforms can intuitively see the pros and cons of different products in each dimension and make optimization decisions accordingly. Visual reports help decision makers more clearly understand which display features are most attractive to users, further improving the overall display effect. Therefore, the present invention improves the comprehensiveness of product comparison by introducing multi-dimensional evaluation, dynamic display adjustment and visual reports.

[0053] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0054] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may 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 the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for multi-dimensional comparison of smart products based on multimodal analysis, characterized in that: The following steps are involved: Step S1: Get user input text; Based on the user input text, similar products are screened in the preset product database to generate similar product screening data; the product display clarity is evaluated on the similar product screening data to generate the first dimension evaluation data of the product; Step S2: obtaining product description parameters; performing three-dimensional modeling of multimodal feature products using the product description parameters based on product image features to generate a three-dimensional product display model; evaluating the product display accuracy of the three-dimensional product display model to generate product second dimension evaluation data; Step S3: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; Step S4: Evaluate the product feedback dimension of the similar product screening data based on the product dynamic display adjustment data to generate product third-dimensional evaluation data; perform comprehensive dimensional comparison and visualization of the similar product screening data based on the product third-dimensional evaluation data, thereby generating a comprehensive dimensional comparison visualization report of similar products.

2. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: Obtain user input text; Step S12: performing a product keyword search on the user input text to obtain product keyword search data; screening similar products in a preset product database based on the product keyword search data to generate similar product screening data; Step S13: extracting product display modal features from the similar product screening data to obtain product display modal feature data, wherein the product display modal features include text features, image features, and video features; Step S14: Evaluate the display clarity of similar products based on the product display modal feature data to generate product first dimension evaluation data.

3. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 1 is characterized in that: Step S14 includes the following steps: Step S141: defining the clarity index of commodity display modality features, wherein the text feature clarity index includes information coverage, language simplicity and accuracy, the image feature clarity index includes resolution score, detail presentation score and color restoration score, and the video feature clarity index includes frame rate analysis, dynamic and static contrast and picture stability score; Step S142: Calculate the number of commodity display modes for the commodity display modal feature data to obtain the number of commodity display modal features; determine whether the number of commodity display modal features is greater than 1. If the number of commodity display modal features is greater than 1, mark the commodity as a multi-modal feature commodity; otherwise, mark the commodity as a single modal feature commodity and remove it; Step S143: performing modal score weighted fusion on multi-modal feature products according to the product display modal feature clarity index to generate multi-modal product display clarity; Step S144: Evaluate the display clarity of similar products by using multimodal product display clarity to generate product first dimension comparison data.

4. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: Obtain product description parameters; Step S22: Performing three-dimensional modeling of the multimodal feature product using the product description parameters based on the image features of the product to generate a three-dimensional model for product display; Step S23: optimizing the commodity display 3D model based on the video features of the commodity to generate a commodity display 3D optimized model; Step S24: Evaluate the accuracy of product display of the three-dimensional optimization model of product display according to the text features of the product, and generate evaluation data of the second dimension of the product.

5. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 4 is characterized in that: Step S23 includes the following steps: Step S231: performing three-dimensional spatial feature mapping on the video features of the product, thereby obtaining spatial mapping data; and reconstructing the model surface details of the three-dimensional model of the product display according to the spatial mapping data, thereby obtaining model detail data; Step S232: constructing a three-dimensional topological structure of the commodity display three-dimensional model through the model detail data, thereby obtaining model structure data; and performing texture material mapping on the commodity display three-dimensional model according to the model structure data, thereby obtaining texture mapping data; Step S233: optimizing the lighting and shadow of the texture mapping data to obtain lighting and shadow effect data; performing dynamic effect rendering on the commodity display three-dimensional model according to the lighting and shadow effect data to obtain dynamic rendering data; Step S234: Utilize the dynamic rendering data to optimize the three-dimensional model of the commodity display to generate a three-dimensional optimized model of the commodity display.

6. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 4 is characterized in that: Step S24 includes the following steps: Step S241: performing part-of-speech classification on the text features of the product to generate product text part-of-speech classification data, wherein the product text part-of-speech classification data includes product noun part-of-speech and product adjective part-of-speech; Step S242: structurally matching the three-dimensional optimization model of commodity display by commodity noun parts of speech to generate a three-dimensional matching structure of commodity display; calculating actual structural parameters of the three-dimensional matching structure of commodity display according to commodity adjective parts of speech to obtain actual structural parameters; Step S243: using the actual structural parameters to traverse the three-dimensional matching structure of similar product display in the three-dimensional model of product display, thereby obtaining the actual structural parameter set of similar products; evaluating the product display accuracy of the actual structural parameter set of similar products through the actual structural parameters, and generating product second dimension evaluation data.

7. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: Calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; Step S32: predicting the commodity purchase conversion rate based on the commodity authenticity data to generate commodity purchase conversion rate prediction data; Step S33: dynamically display and adjust the similar product screening data according to the product purchase conversion rate prediction data to generate product dynamic display adjustment data.

8. The method for multi-dimensional comparison of intelligent commodities based on multimodal analysis according to claim 7 is characterized in that: Step S32 includes the following steps: Step S321: extracting product purchase features from similar product screening data based on product authenticity data to obtain product purchase feature data; dividing the product purchase feature data into data sets to generate a model training set and a model test set; Step S322: Perform model training on the model training set by using a regression analysis algorithm to generate a commodity purchase conversion rate prediction pre-model; perform model optimization iteration on the commodity purchase conversion rate prediction pre-model by using a model test set, thereby generating a commodity purchase conversion rate prediction model; Step S323: Import the commodity authenticity data into the commodity purchase conversion rate prediction model to perform commodity purchase conversion rate prediction and generate commodity purchase conversion rate prediction data.

9. The method for multi-dimensional comparison of smart commodities based on multimodal analysis according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: collecting similar product evaluations on similar product screening data according to the product dynamic display adjustment data to obtain similar product evaluation collection data; evaluating the product feedback dimension on the similar product evaluation collection data to generate product third dimension evaluation data; Step S42: Based on the third dimension evaluation data of the product, the similar product screening data is compared and visualized in a comprehensive dimension, so as to generate a comprehensive dimension comparison visualization report of similar products.

10. A system for multi-dimensional comparison of intelligent commodities based on multimodal analysis, characterized in that: A method for performing multi-dimensional comparison of smart products based on multi-modal analysis as claimed in claim 1, wherein the system for multi-dimensional comparison of smart products based on multi-modal analysis comprises: The display evaluation module is used to obtain the user input text; based on the user input text, similar products are screened in the preset product database to generate similar product screening data; the product display clarity evaluation is performed on the similar product screening data to generate the first dimension evaluation data of the product; The accuracy evaluation module is used to obtain product description parameters; based on the image features of the product, the product description parameters are used to perform three-dimensional modeling of the multimodal feature product to generate a three-dimensional model for product display; the product display accuracy is evaluated on the three-dimensional model for product display to generate evaluation data for the second dimension of the product; A display adjustment module is used to calculate the authenticity of similar product screening data based on the product first dimension evaluation data and the product second dimension evaluation data to generate product authenticity data; dynamically display and adjust the similar product screening data according to the product authenticity data to generate product dynamic display adjustment data; The feedback evaluation module is used to evaluate the product feedback dimension of similar product screening data based on the product dynamic display adjustment data, and generate the product third dimension evaluation data; based on the product third dimension evaluation data, the similar product screening data is compared and visualized in a comprehensive dimension, thereby generating a similar product comprehensive dimension comparison visualization report.

Images