Virtual clothing collocation recommendation method and system based on knowledge graph

Abstract

The invention relates to a virtual costume matching recommendation method and system based on a knowledge graph, and the method comprises the steps: obtaining virtual image data of a user, extracting image feature data of the user from the virtual image data, and constructing a dynamic user portrait in combination with the historical purchase record, social behavior data and preference setting of the user; constructing a costume knowledge graph, wherein the costume knowledge graph comprises costume attribute nodes and costume relation edges; based on a semantic matching algorithm of graph embedding, performing embedded representation on the dynamic user portrait and the clothing knowledge graph, calculating semantic similarity between the dynamic user portrait and the clothing knowledge graph, and generating a semantic matching result; based on a logical reasoning technology, according to scene requirements set by the user, reasoning to generate a clothing collocation combination of the user, and generating a logical reasoning result; and according to a semantic matching result and a logical reasoning result, generating a costume matching suggestion, and supporting a user to carry out interactive adjustment and feedback. The method and the device have the effect of improving the clothing recommendation accuracy.

Description

Technical Field

[0001] This application relates to the technical field of clothing recommendation, and in particular to a virtual clothing matching recommendation method and system based on knowledge graph. Background Technology

[0002] Currently, with the rapid development of e-commerce platforms and social media, virtual clothing recommendation has become one of the key technologies for improving user experience and sales conversion rates. Most existing virtual clothing recommendation systems rely on users' historical purchase records, browsing data, and social behavior data for personalized recommendations. Common methods include collaborative filtering, content-based recommendation, and hybrid recommendation, which generate recommendations by analyzing the relationship between users and clothing. However, these traditional methods typically face problems such as low recommendation accuracy, rapidly changing user needs, and poor scenario adaptability.

[0003] Existing apparel recommendation methods based on graph embedding technology can capture the relationships between clothing items and user preferences relatively well, but most of them ignore the complex interaction between the user's virtual avatar and the clothing. In traditional recommendation systems, user profiles are usually statically constructed, while user interests, needs, and scenarios often change dynamically, resulting in inaccurate and unpersonalized recommendations. Furthermore, most recommendation methods rely solely on historical data for reasoning, failing to fully consider the impact of real-time scenario requirements and fashion trends.

[0004] The existing technical solutions mentioned above have the following drawbacks: they fail to fully integrate user virtual avatar data with clothing knowledge graphs, lack refined analysis of users' dynamic needs, resulting in poor adaptability and real-time performance of recommendation results, and thus cannot truly meet the needs of personalized recommendations. Therefore, there is room for improvement. Summary of the Invention

[0005] To improve the accuracy of clothing recommendations, this application provides a virtual clothing matching recommendation method and system based on knowledge graphs.

[0006] The above-mentioned objective of this application is achieved through the following technical solution: A knowledge graph-based virtual clothing matching recommendation method, comprising: Acquire user virtual avatar data, extract user avatar feature data from the virtual avatar data, and combine it with the user's historical purchase records, social behavior data and preference settings to construct a dynamic user profile; Construct a clothing knowledge graph, which includes clothing attribute nodes and clothing relationship edges; A semantic matching algorithm based on graph embedding is used to embed the dynamic user profile and the clothing knowledge graph, calculate the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generate semantic matching results. Based on logical reasoning technology, the system generates clothing combination combinations for users according to the scenario requirements set by the users, and generates logical reasoning results. Based on the semantic matching results and logical reasoning results, clothing matching suggestions are generated, and users can make interactive adjustments and provide feedback.

[0007] By employing the aforementioned technical solutions, and by acquiring user virtual avatar data and extracting user avatar feature data from it, combined with users' historical purchase records, social behavior data, and preference settings, a dynamic user profile can be constructed. This allows for a more comprehensive and accurate capture of users' personal preferences and needs, thereby generating a more personalized and dynamic user profile and improving the accuracy and real-time performance of recommendations. Furthermore, by constructing a clothing knowledge graph, including clothing attribute nodes and clothing relationship edges, various attribute relationships between clothing items can be systematically represented, enabling the recommendation system to understand the complex connections between clothing items and thus providing users with more multi-dimensional clothing recommendation criteria. Finally, through a semantic matching algorithm based on graph embedding, the dynamic user profile is linked to clothing... By embedding knowledge graphs and calculating their semantic similarity, user preferences can be efficiently matched with clothing characteristics, improving the accuracy of the recommendation system and making the recommendations more in line with user needs. Through logical reasoning technology, clothing combinations can be generated based on user-defined scenario requirements, ensuring that recommendations meet actual user needs in different scenarios and improving the practical applicability and scenario adaptability of the recommendations. By generating clothing matching suggestions based on semantic matching and logical reasoning results, and supporting interactive adjustments and feedback from users, flexible and dynamic clothing recommendation services can be provided, further enhancing the user experience and enabling the recommendation system to adjust and optimize recommended content in real time.

[0008] In one example, this application can be further configured as follows: obtaining user virtual avatar data, extracting user avatar feature data from the virtual avatar data, and constructing a dynamic user profile by combining the user's historical purchase records, social behavior data, and preference settings includes: The virtual avatar data uploaded by the user is analyzed using computer vision technology to identify and extract the avatar feature data, which includes body shape features, skin color features, hairstyle features, and facial expressions. Based on the user's historical purchase records, the user's purchase preference information is analyzed and extracted, including clothing style, color, material and brand; Analyze the social behavior data, which includes user interactions, likes, and comments on social platforms, to obtain information on changes in user interests; By combining the purchase preference information, the interest change information, the preference settings, and the image feature data, the dynamic user profile is generated, and the dynamic user profile is dynamically updated based on user interaction feedback.

[0009] By employing the aforementioned technical solutions, analyzing user-uploaded virtual avatar data using computer vision technology and extracting avatar feature data, a comprehensive understanding of the user's appearance can be achieved from a visual perspective, thereby providing personalized clothing recommendations that match the user's image. By analyzing and extracting user purchase preference information based on historical purchase records, it is possible to deeply understand users' long-term consumption habits and preferences, ensuring that recommended clothing matches the user's actual needs and preferences, thus improving the relevance of recommendations. Analyzing social behavior data allows for real-time acquisition of information on changes in user interests, providing insights into user feedback on current fashion trends and preferences, and offering a basis for personalized adjustments to the recommendation system. By combining purchase preference information, interest change information, preference settings, and avatar feature data to generate dynamic user profiles, it is possible to dynamically capture changes in user needs, ensuring that user profiles are up-to-date and that recommendation results more accurately adapt to changing user demands. By dynamically updating user profiles based on user interaction feedback, the accuracy of user profiles can be continuously optimized, enabling the recommendation system to adapt to changes in user preferences in real time, thereby providing more accurate and personalized clothing recommendations.

[0010] In one example, this application can be further configured such that the construction of the clothing knowledge graph includes: Collect and integrate multiple apparel data sources, including apparel styles, colors, materials, brands, applicable occasions, seasonal characteristics, user reviews, and fashion trends; Based on the clothing data source, determine the attribute categories of the clothing. The attribute categories include the appearance attributes, functional attributes, and association attributes with user characteristics of the clothing. Then, construct the clothing attribute nodes based on the attribute categories of the clothing. Define the relationship edges between the clothing attribute nodes to obtain the clothing relationship edges, which represent the association between clothing attribute nodes; The clothing knowledge graph is constructed using graph database technology, and the clothing attribute nodes and clothing relationship edges are stored and indexed. The system acquires external fashion trends, expert recommendations, and user-generated content within a preset time period. Based on these external fashion trends, expert recommendations, and user-generated content, it supplements and optimizes the clothing knowledge graph through data fusion and knowledge extraction techniques.

[0011] By adopting the above technical solution and collecting and integrating multiple clothing data sources, including clothing styles, colors, materials, brands, applicable occasions, seasonal characteristics, user reviews, and fashion trends, a comprehensive multi-dimensional information on clothing can be covered. This provides rich clothing attribute data for the recommendation system, making the recommendation results more accurate and diverse. By determining the attribute categories of clothing based on the clothing data sources and constructing clothing attribute nodes, clothing information can be structured and classified and represented according to different attribute categories. This allows various features of clothing to be clearly mapped to the knowledge graph, improving the hierarchy and operability of the graph. By defining the relationship edges between clothing attribute nodes, the correlation between various clothing attributes can be clarified. Such relationship edges help capture the mutual influence and matching rules between clothing attributes, thereby... This approach enables the recommendation system to more accurately understand the inherent relationships between clothing items and provide matching recommendations that meet actual needs. By utilizing graph database technology to construct a clothing knowledge graph, storing and indexing clothing attribute nodes and clothing relationship edges, it can effectively manage and query massive amounts of clothing information data, improving the system's ability to process large-scale clothing data and ensuring that the recommendation system can respond to user needs quickly and accurately. By acquiring external fashion trends, expert recommendations, and user-generated content within a preset time period, and supplementing and optimizing the clothing knowledge graph through data fusion and knowledge extraction techniques, it can ensure that the recommendation system can continuously update and optimize the clothing knowledge graph based on the latest fashion trends and user feedback, making the recommendation results more in line with current fashion trends and enhancing the timeliness and accuracy of the user experience.

[0012] In one example, this application can be further configured as follows: the graph embedding-based semantic matching algorithm embeds the dynamic user profile and the clothing knowledge graph, calculates the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generates semantic matching results including: The image feature data in the dynamic user profile and the clothing attributes in the clothing knowledge graph are vectorized, and graph embedding technology is used to map the image feature data and clothing attributes to a low-dimensional vector space. Based on a pre-trained graph embedding model, the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph is calculated to obtain the semantic similarity. Based on the semantic similarity, each clothing attribute node in the clothing knowledge graph is sorted, and a preliminary recommended clothing list is generated. Based on the preliminary recommended clothing list and the preference settings, the semantic matching result is generated.

[0013] By employing the above technical solution, and by vectorizing the image feature data in the dynamic user profile and the clothing attributes in the clothing knowledge graph, complex user and clothing information can be transformed into easily computed vector representations, improving computational efficiency and the accuracy of similarity matching. By calculating the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph based on a pre-trained graph embedding model, the matching degree between user needs and clothing attributes can be efficiently measured, ensuring that the recommendation system can accurately identify the association between users and clothing, thereby providing personalized recommendations. By generating a preliminary recommended clothing list based on the semantic similarity, clothing can be prioritized according to the similarity between the user profile and clothing attributes, ensuring that the clothing in the recommended list best matches the user's needs and preferences, improving the relevance and accuracy of the recommendations. By generating the semantic matching result based on the preliminary recommended clothing list and the preference settings, the recommendation result can be further optimized according to the user's preference settings, ensuring that the final recommended clothing not only matches the semantic matching result but can also be further adjusted according to the user's personalized needs, making the recommendation result more accurate and in line with user expectations.

[0014] In one example, this application can be further configured as follows: based on a pre-trained graph embedding model, the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph is calculated, and the semantic similarity is obtained by: The image feature data in the dynamic user profile is transformed into the embedded representation of user nodes to generate user embedding vectors; the clothing attribute nodes in the clothing knowledge graph are transformed into low-dimensional vector representations of clothing attribute nodes to generate clothing attribute node embedding vectors. Through the formula: The semantic similarity is calculated, where U is the user embedding vector, A is the clothing attribute node embedding vector, U·A represents the dot product of vector U and vector A, ||U|| represents the magnitude of vector U, and ||A|| represents the magnitude of vector A.

[0015] By adopting the above technical solutions, and transforming the image feature data in the dynamic user profile into embedded representations of user nodes to generate user embedding vectors, complex user appearance features can be transformed into digital vector forms, making them processable by computer systems and providing structured input data for subsequent recommendation calculations. Furthermore, by transforming clothing attribute nodes in the clothing knowledge graph into low-dimensional vector representations of clothing attribute nodes to generate clothing attribute node embedding vectors, various clothing attributes can be converted into low-dimensional vectors, facilitating efficient calculation and comparison. This allows the recommendation system to quickly process and identify similarities between clothing items. Finally, by calculating the semantic similarity using a formula, the similarity between users and clothing can be effectively measured, providing the recommendation system with a quantitative and accurate matching degree metric, ensuring that the recommendation results meet the user's interests and needs.

[0016] In one example, this application can be further configured as follows: based on logical reasoning technology, according to the user-defined scenario requirements, the application infers and generates the user's clothing combination, and the generated logical reasoning results include: The scenario requirements are mapped to the clothing attributes in the clothing knowledge graph to generate a preliminary clothing candidate list; based on the relationship edges between the preliminary clothing candidate list and the clothing attributes in the clothing knowledge graph, and combined with the scenario requirements, a preliminary clothing matching combination is deduced. Based on user feedback, the initial clothing combinations are optimized to obtain the logical reasoning result.

[0017] By adopting the above technical solution, and mapping the scenario requirements to the clothing attributes in the clothing knowledge graph, the recommendation system can ensure that it can filter out clothing that meets the requirements of the scenario set by the user, providing preliminary candidate options for subsequent clothing matching. Utilizing the inherent relationships between clothing attributes, reasonable clothing combinations are generated based on meeting the scenario requirements, thereby providing matching schemes that meet the user's needs. By combining user feedback information, the preliminary clothing matching combinations are optimized to obtain the logical reasoning result. The matching combinations can be dynamically adjusted based on real-time user feedback, ensuring that the recommendation results better meet the user's personalized needs and improving the accuracy and satisfaction of the recommendations.

[0018] In one example, this application can be further configured such that generating clothing matching suggestions based on the semantic matching result and logical reasoning result includes: Based on the semantic matching results, the required clothing items are selected, and combined with the logical reasoning results, the order and combination of clothing are adjusted to generate preliminary clothing matching suggestions. Based on the trend nodes in the clothing knowledge graph, fashion trend prediction is performed on clothing matching to obtain clothing fashion trend prediction results. The preliminary clothing matching suggestions and the clothing fashion trend prediction results are integrated to generate the clothing matching suggestions, which include complete clothing matching, individual item recommendations, and fashion trend predictions.

[0019] By employing the aforementioned technical solutions, and filtering out desired clothing items based on semantic matching results, and adjusting the order and combination of clothing items based on logical reasoning results to generate preliminary clothing matching suggestions, we can ensure that the recommended clothing combinations not only meet the user's needs but also consider reasonable combinations between clothing items, making the recommendations more accurate and in line with the user's actual needs in the scenario. By predicting fashion trends based on the trend nodes in the clothing knowledge graph, we can capture changes in clothing trends and update the matching suggestions of the recommendation system in real time, making the recommended content more fashionable and forward-looking, thereby improving the timeliness and market adaptability of the recommendations. By integrating the preliminary clothing matching suggestions and the clothing fashion trend prediction results to generate final clothing matching suggestions, we can combine the user's personalized needs with fashion trends, providing more comprehensive and personalized clothing recommendations, including complete clothing combinations, single item recommendations, and fashion trends, enhancing the accuracy and attractiveness of the recommendation system.

[0020] The second objective of this invention is achieved through the following technical solution: A knowledge graph-based virtual clothing matching recommendation system, comprising: The feature extraction module is used to acquire user virtual avatar data, extract user avatar feature data from the virtual avatar data, and construct a dynamic user profile by combining the user's historical purchase records, social behavior data and preference settings. A graph construction module is used to construct a clothing knowledge graph, which includes clothing attribute nodes and clothing relationship edges; a semantic matching module is used to embed the dynamic user profile and the clothing knowledge graph into a semantic matching algorithm based on graph embedding, calculate the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generate semantic matching results. The logical reasoning module is used to generate clothing matching combinations for users based on the user's set scenario requirements, using logical reasoning technology, and to generate logical reasoning results. The suggestion generation module is used to generate clothing matching suggestions based on the semantic matching results and logical reasoning results, and supports users to make interactive adjustments and provide feedback.

[0021] By employing the aforementioned technical solutions, and by acquiring user virtual avatar data and extracting user avatar feature data from it, combined with users' historical purchase records, social behavior data, and preference settings, a dynamic user profile can be constructed. This allows for a more comprehensive and accurate capture of users' personal preferences and needs, thereby generating a more personalized and dynamic user profile and improving the accuracy and real-time performance of recommendations. Furthermore, by constructing a clothing knowledge graph, including clothing attribute nodes and clothing relationship edges, various attribute relationships between clothing items can be systematically represented, enabling the recommendation system to understand the complex connections between clothing items and thus providing users with more multi-dimensional clothing recommendation criteria. Finally, through a semantic matching algorithm based on graph embedding, the dynamic user profile is linked to clothing... By embedding knowledge graphs and calculating their semantic similarity, user preferences can be efficiently matched with clothing characteristics, improving the accuracy of the recommendation system and making the recommendations more in line with user needs. Through logical reasoning technology, clothing combinations can be generated based on user-defined scenario requirements, ensuring that recommendations meet actual user needs in different scenarios and improving the practical applicability and scenario adaptability of the recommendations. By generating clothing matching suggestions based on semantic matching and logical reasoning results, and supporting interactive adjustments and feedback from users, flexible and dynamic clothing recommendation services can be provided, further enhancing the user experience and enabling the recommendation system to adjust and optimize recommended content in real time.

[0022] In summary, this application includes the following beneficial technical effects: 1. By acquiring user virtual avatar data and extracting user avatar feature data, combined with users' historical purchase records, social behavior data, and preference settings, a dynamic user profile can be constructed. This allows for a more comprehensive and accurate capture of users' personal preferences and needs, resulting in a more personalized and dynamic user profile and improving the accuracy and real-time performance of recommendations. 2. By constructing a clothing knowledge graph, including clothing attribute nodes and clothing relationship edges, various attribute relationships between clothing items can be systematically represented, enabling the recommendation system to understand the complex connections between clothing items and providing users with more multi-dimensional clothing recommendation criteria. 3. Through a semantic matching algorithm based on graph embedding, the dynamic user profile and the clothing knowledge graph are embedded and their semantic similarity is calculated. This allows for efficient matching of user preferences with clothing characteristics, improving the accuracy of the recommendation system and making the recommendation results more in line with user needs. 2. By using logical reasoning technology to generate clothing combinations based on user-defined scenario requirements, the system ensures that the recommended results meet the user's actual needs in different scenarios, improving the practical applicability and scenario adaptability of the recommendations. By generating clothing matching suggestions based on semantic matching results and logical reasoning results, and supporting interactive adjustments and feedback from users, the system can provide users with flexible and dynamic clothing recommendation services, further enhancing the user experience and enabling the recommendation system to adjust and optimize the recommended content in real time. Attached Figure Description

[0023] Figure 1 This is a flowchart of a virtual clothing matching recommendation method based on knowledge graph in one embodiment of this application; Figure 2 This is a flowchart illustrating the implementation of step S10 in a knowledge graph-based virtual clothing matching recommendation method according to an embodiment of this application. Figure 3 This is a flowchart illustrating the implementation of step S20 in a knowledge graph-based virtual clothing matching recommendation method according to an embodiment of this application. Figure 4 This is a flowchart illustrating the implementation of step S30 in a knowledge graph-based virtual clothing matching recommendation method according to an embodiment of this application. Figure 5 This is a flowchart illustrating the implementation of step S32 in a knowledge graph-based virtual clothing matching recommendation method according to an embodiment of this application. Figure 6 This is a flowchart illustrating the implementation of step S40 in a knowledge graph-based virtual clothing matching recommendation method according to an embodiment of this application. Figure 7 This is a flowchart illustrating the implementation of step S50 in a knowledge graph-based virtual clothing matching recommendation method according to an embodiment of this application. Figure 8 This is a principle block diagram of a virtual clothing matching recommendation system based on knowledge graphs in one embodiment of this application; Detailed Implementation

[0024] The present application will be further described in detail below with reference to the accompanying drawings.

[0025] In one embodiment, such as Figure 1 As shown, this application discloses a virtual clothing matching recommendation method based on knowledge graphs, which specifically includes the following steps: S10: Obtain user virtual avatar data, extract user image feature data from the virtual avatar data, and combine it with the user's historical purchase records, social behavior data and preference settings to build a dynamic user profile.

[0026] Specifically, by acquiring user avatar data, the system first extracts image feature data such as body shape, skin tone, hairstyle, and facial expressions from the user-uploaded avatar. This data accurately reflects the user's appearance characteristics, thus providing a basis for personalized clothing matching for the recommendation system. For example, through computer vision technology, user-uploaded avatar photos can be analyzed, and image processing and feature extraction algorithms (such as convolutional neural networks) can be used to identify facial contours, hairstyles, skin tone, and body shape information, thereby obtaining these image feature data. These features will serve as an important component of the user's personalized needs. In addition to image features, dynamic user profiles also combine users' historical purchase records to analyze past purchasing behavior, including the clothing styles, brands, and materials chosen by the user. Based on this information, user preference features are extracted. Furthermore, based on users' social behavior data such as interactions, likes, and comments on social platforms, dynamic changes in user interests are captured. By integrating this multi-dimensional data, a comprehensive and dynamically changing user profile is generated.

[0027] S20: Construct a clothing knowledge graph, which includes clothing attribute nodes and clothing relationship edges.

[0028] Specifically, by constructing a clothing knowledge graph, the various attributes of clothing (such as style, color, material, brand, etc.) are first classified, and attribute nodes are created for each attribute. Furthermore, by analyzing the relationships between clothing items (such as matching relationships, applicable occasions, seasonal characteristics, etc.), relationship edges are defined between these nodes. The clothing knowledge graph can then accurately reflect the inherent connections between different clothing attributes. For example, the relationship edge between the clothing style node and the color node may represent which colors are more commonly paired with which styles, and the edge between the applicable occasion node and the style node represents common clothing styles in different occasions. Through this structure, the knowledge graph can effectively represent the correlations between clothing attributes and provide rich graph-based data for subsequent recommendations.

[0029] S30: A semantic matching algorithm based on graph embedding embeds dynamic user profiles and clothing knowledge graphs, calculates the semantic similarity between the dynamic user profiles and clothing knowledge graphs, and generates semantic matching results.

[0030] Specifically, using a graph embedding-based semantic matching algorithm, the user profile and clothing knowledge graph are first graph-embedded, transforming the user's image feature data and clothing attribute nodes into low-dimensional vector representations. Graph embedding technology maps the originally high-dimensional and complex user and clothing data to a low-dimensional space, facilitating effective similarity calculation. Specifically, the semantic similarity between the user embedding vector and the clothing attribute embedding vector is calculated using common similarity calculation methods (such as cosine similarity or dot product methods) to measure the matching degree between user needs and clothing attributes. For example, the similarity between a user's body shape features and a certain clothing style is obtained by calculating the dot product of the user and clothing node vectors; clothing with higher similarity is prioritized for recommendation to the user.

[0031] S40: Based on logical reasoning technology, it infers and generates clothing matching combinations for the user according to the user's set scenario requirements, and generates logical reasoning results.

[0032] Specifically, using logical reasoning technology, the system first identifies the user-defined scenario (such as business or casual occasions) based on the user's specific needs. Then, it infers clothing combinations that match the scenario based on the relationships within the clothing knowledge graph. For example, in a business setting, the recommendation system might infer combinations like suits, ties, and dress shoes, determining the order of these combinations based on the relationships between clothing attribute nodes in the graph. Logical reasoning not only relies on preset relationships in the clothing knowledge graph but can also further optimize the inference results based on the user's personalized needs and preferences, generating preliminary clothing combinations that meet the user's requirements.

[0033] S50: Based on semantic matching results and logical reasoning results, generate clothing matching suggestions and support users to make interactive adjustments and provide feedback.

[0034] Specifically, clothing matching suggestions are generated based on semantic matching and logical reasoning results. First, clothing items matching the user's needs are selected based on semantic matching results between the user profile and the clothing knowledge graph. Then, the matching order and combination of clothing items are optimized by combining the matching results generated through logical reasoning. For example, an initial outfit might consist of multiple garments; by adjusting the order (e.g., placing the top first and accessories last), the recommended outfits better suit the user's dressing habits in specific scenarios. Furthermore, users can interactively adjust and provide feedback on the recommendations. Real-time user feedback further refines and optimizes the recommendations, ensuring that the results always align with user needs, thereby improving overall recommendation accuracy and user satisfaction.

[0035] By employing the aforementioned technical solutions, and by acquiring user virtual avatar data and extracting user avatar feature data from it, combined with users' historical purchase records, social behavior data, and preference settings, a dynamic user profile can be constructed. This allows for a more comprehensive and accurate capture of users' personal preferences and needs, thereby generating a more personalized and dynamic user profile and improving the accuracy and real-time performance of recommendations. Furthermore, by constructing a clothing knowledge graph, including clothing attribute nodes and clothing relationship edges, various attribute relationships between clothing items can be systematically represented, enabling the recommendation system to understand the complex connections between clothing items and thus providing users with more multi-dimensional clothing recommendation criteria. Finally, through a semantic matching algorithm based on graph embedding, the dynamic user profile is linked to clothing... By embedding knowledge graphs and calculating their semantic similarity, user preferences can be efficiently matched with clothing characteristics, improving the accuracy of the recommendation system and making the recommendations more in line with user needs. Through logical reasoning technology, clothing combinations can be generated based on user-defined scenario requirements, ensuring that recommendations meet actual user needs in different scenarios and improving the practical applicability and scenario adaptability of the recommendations. By generating clothing matching suggestions based on semantic matching and logical reasoning results, and supporting interactive adjustments and feedback from users, flexible and dynamic clothing recommendation services can be provided, further enhancing the user experience and enabling the recommendation system to adjust and optimize recommended content in real time.

[0036] In one embodiment, such as Figure 2 As shown, in step S10, user virtual avatar data is acquired, user avatar feature data is extracted from the virtual avatar data, and combined with the user's historical purchase records, social behavior data, and preference settings to construct a dynamic user profile, specifically including: S11: Analyze the virtual avatar data uploaded by users using computer vision technology, identify and extract avatar feature data, including body shape features, skin color features, hairstyle features and facial expressions.

[0037] Specifically, a deep learning-based convolutional neural network (CNN) model is used to perform multi-task learning on the user-uploaded virtual avatar images. For example, a pre-trained ResNet-50 model is used as the base network, and multiple branch output layers are added to predict body features such as height, shoulder width, and waist circumference. Skin color features are obtained by analyzing the average hue and saturation of the skin region through the HSV color space. Hairstyle features are obtained by segmenting the hair region using the semantic segmentation model U-Net and extracting attributes such as hair length, hair color, and curl. Facial expressions are obtained by detecting facial key points through the OpenFace framework and combining them with an expression classification model to identify the user's expression type, such as happy or neutral. All feature extraction processes are optimized through end-to-end training to ensure the correlation and accuracy between features.

[0038] S12: Based on the user's historical purchase records, analyze and extract the user's purchase preference information, including clothing style, color, material and brand.

[0039] Specifically, natural language processing techniques (such as the BERT model) are used to perform semantic parsing on product description text in historical purchase records to extract keywords (such as "loose T-shirt", "pure cotton", "sports brand"). Collaborative filtering algorithms are then used to perform cluster analysis on user purchase behavior. For example, matrix factorization is used to decompose the user-product interaction matrix into latent factor vectors, thereby uncovering implicit patterns of users' preferences for specific styles (such as casual wear and formal wear), color preferences (such as cool colors and warm colors), material preferences (such as cotton and silk), and brand loyalty. At the same time, time series analysis (such as the LSTM model) is combined to capture the dynamic trends of user preferences.

[0040] S13: Analyze social behavior data, which includes user interactions, likes, and comments on social platforms, to obtain information on changes in user interests.

[0041] Specifically, web scraping technology is used to acquire users' public behavioral data on social media platforms, such as Weibo interactions, Instagram likes, and Douyin comments. Sentiment analysis models (such as VADER) are used to classify the sentiment polarity of the comments (positive, neutral, negative), and topic modeling techniques (such as LDA) are combined to extract clothing-related interest topics (such as "retro style" and "street fashion brands") from the text. At the same time, a user-content interaction graph is constructed using graph neural networks (GNN) to analyze the interest propagation path in users' likes and reposts. For example, if a user frequently likes a fashion blogger's "minimalist style" outfit content, it is inferred that their interest is shifting towards minimalism. The interest weights are dynamically adjusted using a time decay factor to ensure the real-time nature of interest changes. In addition, combining users' geographical location information (such as IP address positioning) and social circle analysis (such as friends' interest preferences) further enriches the dimensions and accuracy of interest change information.

[0042] S14: Combine purchase preference information, interest change information, preference settings and image characteristic data to generate dynamic user profiles, and dynamically update the dynamic user profiles based on user interaction feedback.

[0043] Specifically, a dynamic user profile is generated by combining the aforementioned purchase preference information, interest change information, preference settings, and image feature data. First, by comprehensively analyzing the extracted user preference and interest change data, a multi-dimensional user profile is constructed, including the user's external features such as body shape, skin color, hairstyle, and facial expressions. Simultaneously, it incorporates their historical purchase behavior, social interactions, and emotional preferences to comprehensively present the user's personalized needs. This user profile is a dynamic, real-time updated model. As user purchase behavior, social behavior, and user feedback change, the user profile can be continuously updated and adjusted, thus maintaining a high degree of alignment with user needs. Through this process, recommended content can be flexibly adjusted according to the user's changing needs, ensuring that each recommendation meets the current user preferences as much as possible.

[0044] In one embodiment, such as Figure 3 As shown, in step S20, which involves constructing a clothing knowledge graph, the specific steps include: S21: Collect and integrate multiple apparel data sources, including apparel styles, colors, materials, brands, applicable occasions, seasonal characteristics, user reviews, and fashion trends.

[0045] Specifically, by collecting and integrating multiple apparel data sources, the process first categorizes and integrates information such as style, color, material, brand, applicable occasions, seasonal characteristics, user reviews, and fashion trends. For example, by analyzing product descriptions, user reviews, fashion blogs, and social media content from numerous e-commerce platforms, basic apparel attributes are extracted, such as the color, material, applicable occasions, and seasonal characteristics of a suit. This information is then structured to form apparel data sources.

[0046] S22: Based on the clothing data source, determine the attribute categories of the clothing. The attribute categories include the appearance attributes, functional attributes, and related attributes of the clothing to user characteristics. Then, construct clothing attribute nodes based on the attribute categories of the clothing.

[0047] Specifically, based on the clothing data source, the different attributes of the clothing are first classified. These attribute categories include the appearance attributes (such as style, color, and material), functional attributes (such as applicable occasions, comfort, and durability), and attributes related to user characteristics (such as user body shape, skin tone, and preferences). For example, the appearance attributes of a suit may include style (such as single-breasted or double-breasted), color (such as black or dark blue), and material (such as wool or cotton); its functional attributes may include suitability for formal occasions and providing a high-quality and comfortable wearing experience; and its attributes related to user characteristics may include suitability for a slimmer body type or suitability for users with darker skin tones.

[0048] S23: Define the relationship edges between clothing attribute nodes to obtain clothing relationship edges, which represent the association between clothing attribute nodes.

[0049] Specifically, relationship edges are defined between the clothing attribute nodes to represent the associations between them. For example, in a clothing knowledge graph, there might be a relationship edge between the style node and the color node of a suit, indicating that a particular style of suit is more suitable for a specific color. Similarly, there might be a relationship edge between suit style and applicable occasion, indicating the suit style suitable for formal occasions. Relationship edges are not limited to connections between attributes; new association rules can be set based on factors such as fashion trends, such as the association between a certain color and seasonal demand, or the demand changes for a particular suit in a specific season. Through these relationship edges, the graph can more accurately describe the inherent connections between various clothing attributes.

[0050] S24: Construct a clothing knowledge graph using graph database technology, storing and indexing clothing attribute nodes and clothing relationship edges.

[0051] Specifically, the clothing knowledge graph is constructed using graph database technology. Graph databases can effectively store a large number of clothing attribute nodes and relational edges, and support efficient querying and analysis. First, various attribute nodes of clothing (such as style, color, material, brand, etc.) are stored as nodes, and the relationships between nodes are represented by relational edges and stored in the graph database. The advantage of graph databases is that they can quickly obtain the relationship information between clothing nodes and other related nodes through query statements, and can efficiently perform graph traversal and path search, thereby enabling recommendation systems based on graph embedding algorithms to quickly process and match user needs with clothing features.

[0052] S25: Obtain external fashion trends, expert recommendations, and user-generated content within a preset time period. Based on these external fashion trends, expert recommendations, and user-generated content, supplement and optimize the apparel knowledge graph through data fusion and knowledge extraction technologies.

[0053] Specifically, the process involves acquiring external fashion trends, expert recommendations, and user-generated content within a predetermined timeframe. First, it involves scraping the latest trend information from fashion blogs, social media, and e-commerce platforms, combining this data with expert fashion predictions and user-generated content (such as comments and likes) to obtain external fashion trend data. For example, by analyzing popular hashtags on social media, it's possible to identify which styles and colors are gaining widespread attention in a particular season or period. Combining this with expert recommendations and user-generated content (such as positive user reviews of specific clothing styles), it's possible to predict which clothing items will be popular in the future. Then, using data fusion and knowledge extraction techniques, this external fashion trend data, expert recommendations, and user-generated content are integrated to supplement and optimize the clothing knowledge graph, ensuring that the fashion trend information within the graph remains aligned with market demand and user interests.

[0054] In one embodiment, such as Figure 4 As shown, in step S30, which is the semantic matching algorithm based on graph embedding, the dynamic user profile and the clothing knowledge graph are embedded and represented, the semantic similarity between the dynamic user profile and the clothing knowledge graph is calculated, and the semantic matching result is generated, specifically including: S31: Vectorize the image feature data in the dynamic user profile and the clothing attributes in the clothing knowledge graph, and use graph embedding technology to map the image feature data and clothing attributes to a low-dimensional vector space.

[0055] Specifically, by vectorizing the image feature data in the dynamic user profile and the clothing attributes in the clothing knowledge graph, the user's image features such as body shape, skin color, and hairstyle, as well as clothing attributes such as style, color, and material, are first converted into low-dimensional vectors using graph embedding technology. This process involves training a graph embedding model based on the GraphSAGE model, mapping each user's image features and each clothing attribute node to a shared low-dimensional space. In the graph embedding model, the user's image features and clothing attribute nodes are represented as fixed-dimensional vectors. These vectors represent the relationships and similarities between the features. In this way, complex high-dimensional data can be transformed into low-dimensional data suitable for computer processing, making clothing recommendations more efficient and able to capture deeper feature relationships. The graph embedding process of GraphSAGE is as follows: in, Let N(v) be the embedding of node v at layer k, and let N(v) represent the neighboring nodes of node v. It is the embedding of the neighbor node u in the previous layer. This refers to the embedding of node v in the previous layer. The Aggregator is an aggregation function used to summarize the embedding information of neighboring nodes. In this way, the embedding of node v is updated based on the embedding information of its neighboring nodes, thereby capturing the node's contextual information.

[0056] S32: Based on a pre-trained graph embedding model, calculate the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph to obtain the semantic similarity.

[0057] Specifically, based on a pre-trained graph embedding model (built using the GraphSAGE model), the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph is calculated. First, common similarity calculation methods such as dot product or cosine similarity are used to evaluate the matching degree between each user feature and clothing feature. For example, the dot product of the user's skin color feature vector and the clothing color vector is calculated; the magnitude of the dot product indicates the degree of similarity between the two. The larger the result, the better the clothing color matches the user's skin color, and thus the greater the likelihood of recommending the clothing. Through this calculation method, the semantic similarity between the dynamic user profile and each clothing attribute can be obtained, measuring the fit between clothing and user preferences.

[0058] S33: Based on semantic similarity, sort each clothing attribute node in the clothing knowledge graph and generate a preliminary recommended clothing list.

[0059] Specifically, the calculated semantic similarity scores are first sorted, and then the clothing attribute nodes that best match the user profile are selected based on the magnitude of the similarity. For example, for a particular user, the system may calculate a series of clothing attribute nodes based on semantic similarity. These nodes are arranged according to their matching degree, with the most relevant clothing items appearing first, forming a preliminary recommendation list. This sorting result reflects the degree of matching between each clothing attribute and the user. The clothing items in the recommendation list should best meet the user's needs, and the provided recommendation list should be highly targeted and relevant.

[0060] S34: Based on the initial recommended clothing list and combined with preference settings, generate semantic matching results.

[0061] Specifically, the system first adjusts the initial clothing recommendation list based on the user's personal preferences, which may include brand preferences, style preferences, or preferences for certain colors and materials. In this way, the system further optimizes the recommended clothing list to ensure that the recommendations are not only based on the similarity between clothing and user characteristics, but also accurately reflect the user's specific needs and preferences, thus providing more precise and personalized clothing matching suggestions. For example, if a user indicates a preference for a particular brand or style in their settings, the system will consider this brand or style preference even if a particular garment matches the user's other characteristics highly, optimizing the recommended outfits and ultimately generating the outfits the user is most likely to like. Furthermore, the user's preferences can be incorporated into the recommendation model using a weighted formula. Assuming the user preference vector is P, the final garment matching degree can be optimized using the following weighted formula: S(U,A)=α·cosine_similarity(U,A)+β·P, where α and β are weight parameters that can be pre-set according to requirements, cosine_similarity(U,A) is the semantic similarity between the user and the garment node, P represents the matching degree of the user's preferences, and the final S(U,A) value serves as a comprehensive score used to generate the optimized recommendation results. By adjusting the weight parameters, the recommendation results can be flexibly adjusted according to the user's preferences, thereby achieving more personalized and accurate garment recommendations.

[0062] In one embodiment, such as Figure 5 As shown, in step S32, based on the pre-trained graph embedding model, the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph is calculated to obtain the semantic similarity, which specifically includes: S321: Transform the image feature data in the dynamic user profile into the embedded representation of user nodes, and generate user embedding vectors.

[0063] Specifically, the image feature data in the dynamic user profile is transformed into embedded representations of user nodes. First, the user's body shape, skin color, hairstyle, facial expression, and other image features are converted into low-dimensional vector representations using a pre-trained graph embedding model. During this process, image features are extracted using a deep learning model (such as a convolutional neural network or graph convolutional network) and mapped to a low-dimensional space. For example, a user's body shape feature might be transformed into a vector containing values ​​such as height, weight, and body type; skin color feature might be transformed into a vector representing skin tone depth; and hairstyle and facial expression might be transformed into vectors reflecting hairstyle type and facial expression, respectively. In the graph embedding process, these user image features are embedded into a low-dimensional vector space, represented as user embedding vectors. These vectors serve as the basis for subsequent matching with clothing attribute nodes.

[0064] S322: Transform clothing attribute nodes in the clothing knowledge graph into low-dimensional vector representations of clothing attribute nodes, and generate clothing attribute node embedding vectors.

[0065] Specifically, the clothing attribute nodes in the clothing knowledge graph are transformed into low-dimensional vector representations of clothing attribute nodes. First, graph embedding technology is used to encode various clothing attributes (such as style, color, material, etc.). Each clothing attribute, such as color, style, brand, etc., will be converted into a corresponding vector representation. For example, the color of clothing may be mapped to a multi-dimensional vector, with different dimensions of the vector corresponding to different color features. The style of clothing may be represented by a vector, such as suit, T-shirt, skirt, etc. In this process, the graph embedding model used maps clothing attribute nodes to a low-dimensional space, ensuring that the relationships between attributes can be represented by geometric vector relationships, thus facilitating subsequent similarity calculations.

[0066] S323: By formula: The semantic similarity is calculated, where U is the user embedding vector, A is the clothing attribute node embedding vector, U·A represents the dot product of vector U and vector A, ||U|| represents the magnitude of vector U, and ||A|| represents the magnitude of vector A.

[0067] Specifically, the semantic similarity is calculated using a formula. First, the similarity between the user embedding vector U and the clothing attribute node embedding vector A is calculated. This similarity can be calculated using cosine similarity or dot product. The formula is as follows: Let U be the user embedding vector and A be the clothing attribute node embedding vector. U·A represents the dot product of vectors U and A, ||U|| represents the magnitude of vector U, and ||A|| represents the magnitude of vector A. The similarity between the user and clothing attribute nodes is calculated using the dot product; a higher value indicates a better match. For example, if a user's skin tone and the color of a certain garment highly match, their similarity value is high, meaning that the garment might be more suitable for recommendation to that user. In this way, the semantic similarity between user profiles and clothing attributes can be quantified, thus providing more accurate clothing recommendations.

[0068] In one embodiment, such as Figure 6 As shown, in step S40, based on logical reasoning technology, according to the user's set scenario requirements, the system infers and generates the user's clothing combination, generating a logical reasoning result, specifically including: S41: Map the scenario requirements to the clothing attributes in the clothing knowledge graph to generate a preliminary clothing candidate list.

[0069] Specifically, by mapping the scenario requirements to clothing attributes in the clothing knowledge graph, the user's defined scenario requirements are first clarified. For example, if a user needs to dress for a formal business occasion, the scenario requirements will be defined with keywords such as "formal" and "professional." Based on these scenario requirements, clothing attribute nodes that meet the requirements of "formal" occasions are filtered from the clothing knowledge graph. For example, clothing attribute nodes such as business suits, ties, and dress shoes will be mapped as essential elements for formal occasions. By filtering the attributes of these nodes, a preliminary clothing candidate list is generated. This list includes clothing attributes that highly match the scenario requirements. These clothing attribute nodes form a basic recommended candidate set, which contains all clothing choices related to the scenario requirements.

[0070] S42: Based on the preliminary clothing candidate list and the relationship edges between clothing attributes in the clothing knowledge graph, and combined with the scenario requirements, a preliminary clothing matching combination is deduced.

[0071] Specifically, the system first understands the matching rules between clothing attributes based on the relationships between them. For example, suits with ties and dress shoes are common matching relationships, while casual wear with sneakers also forms a similar relationship. Through inference algorithms, combined with the "formal" requirement of the scenario, possible preliminary clothing combinations are generated. For instance, in a formal setting, the system might infer a dark suit with a red tie and black dress shoes, while for female users, it might infer a dark blue skirt with a white shirt and black high heels. This process infers from the known relationships between clothing attributes and ensures that each combination meets the requirements of the scenario and the matching degree of the clothing attributes, thus providing preliminary clothing combinations that meet the needs.

[0072] S43: Based on user feedback, optimize the initial clothing combinations to obtain logical reasoning results.

[0073] Specifically, the first step is to obtain user feedback on the recommended outfits. This feedback can include the user's choice of a particular combination, their level of liking for it, their rating, or other interactive information. For example, if a user dislikes a certain color or style in the initial recommended outfits, their feedback will be used as a basis for adjustment. Based on user feedback, the optimization algorithm will adjust the outfit combinations according to the user's preferences. Specifically, if a user prefers a gray suit to a black suit, the system will prioritize recommending gray suits in the generated outfits. The optimized outfit combinations will better meet the user's personalized needs, resulting in more accurate logical reasoning and ensuring that the final outfit suggestions meet the user's scenario requirements and personal preferences.

[0074] In one embodiment, such as Figure 7As shown, in step S50, based on the semantic matching results and logical reasoning results, clothing matching suggestions are generated, specifically including: S51: Based on the semantic matching results, filter out the required clothing items, and combine the results of logical reasoning to adjust the order and combination of clothing items and generate preliminary clothing matching suggestions.

[0075] Specifically, based on the semantic matching results, clothing items that meet the user's needs are first selected from the recommended clothing candidates. These needs may include style, color, material, etc., and are sorted and filtered according to the user's preferences and the degree of matching with the scenario requirements. Assuming the user has set two key scenario requirements: "casual" and "lightweight," the style, material, and comfort of the clothing will be used as the basis for filtering during the matching process. For example, from the initial candidate clothing, suits and formal shirts have a low matching degree and will be filtered out, while sweatpants and T-shirts have a high matching degree and will be retained. Based on this, according to the logical reasoning results, the order and combination of clothing are adjusted. For example, sneakers are placed first, with T-shirts and sweatpants as a pairing, ensuring that the recommended clothing combinations meet the scenario requirements and user preferences, generating preliminary clothing combination suggestions, and displaying them to the user for further interaction.

[0076] S52: Based on the trend nodes in the clothing knowledge graph, predict the fashion trends of clothing matching and obtain the clothing fashion trend prediction results.

[0077] Specifically, by analyzing historical fashion trends and current market dynamics, the popularity of clothing items within a given period is identified. For example, assuming that data analysis from the past season reveals that certain types of sneakers, minimalist T-shirts, and high-waisted jeans were popular in summer, trend nodes will mark these items as hot fashion elements. This fashion trend prediction is based on continuously updated trend data in a clothing knowledge graph and recommendations from fashion experts. It reflects market changes, combines user needs, and predicts potentially popular clothing combinations for the next or current season, thus influencing recommended clothing combinations.

[0078] S53: Integrate preliminary clothing matching suggestions and clothing fashion trend prediction results to generate clothing matching suggestions, including complete clothing matching, individual item recommendations, and fashion trend predictions.

[0079] Specifically, the system merges fashion trend predictions with initial outfit suggestions, adjusting the types of clothing in the outfits to ensure that the recommendations not only meet the user's individual needs but also align with current fashion trends. For example, if the initial recommended outfit is a casual T-shirt and sweatpants, but current trends show a more popular "minimalist style," the system will adjust the recommendations to a minimalist style sweatpants and jeans. Alternatively, based on trends, it may add clothing of a certain color or style to the recommendations to make them fashionable. During the integration process, the final outfit suggestions not only include complete outfit combinations (such as sneakers, T-shirts, and sweatpants) but also provide individual item recommendations (such as recommending specific styles of sneakers or T-shirts). Combined with fashion trend predictions, the system provides users with updated fashion outfit suggestions, ensuring the timeliness, personalization, and alignment with current fashion trends of the recommendations.

[0080] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0081] In one embodiment, a knowledge graph-based virtual clothing matching recommendation system is provided, which corresponds one-to-one with the knowledge graph-based virtual clothing matching recommendation method described in the previous embodiment. For example... Figure 8 As shown, this virtual clothing matching recommendation system based on knowledge graph includes a feature extraction module, a knowledge graph construction module, a semantic matching module, a logical reasoning module, and a suggestion generation module. The detailed descriptions of each functional module are as follows: The feature extraction module is used to acquire user virtual avatar data, extract user avatar feature data from the virtual avatar data, and combine it with the user's historical purchase records, social behavior data, and preference settings to construct a dynamic user profile; The graph construction module is used to construct a clothing knowledge graph, which includes clothing attribute nodes and clothing relationship edges; the semantic matching module is used for a semantic matching algorithm based on graph embedding, which embeds the dynamic user profile and the clothing knowledge graph into a representation, calculates the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generates semantic matching results. The logical reasoning module is used to generate clothing matching combinations for users based on the user's set scenario requirements, using logical reasoning technology, and to generate logical reasoning results. The suggestion generation module generates clothing matching suggestions based on semantic matching results and logical reasoning results, and supports interactive adjustments and feedback from users.

[0082] Optionally, the feature extraction module includes: The image analysis submodule is used to analyze the virtual image data uploaded by users through computer vision technology, and to identify and extract image feature data, including body shape features, skin color features, hairstyle features and facial expressions. The preference analysis submodule is used to analyze and extract users' purchase preference information based on their historical purchase records. Purchase preference information includes clothing style, color, material and brand. The interest analysis submodule is used to analyze social behavior data, which includes users' interactions, likes, and comments on social platforms, to obtain information on changes in users' interests. The profile generation submodule is used to combine purchase preference information, interest change information, preference settings and image feature data to generate dynamic user profiles, and dynamically update the dynamic user profiles based on user interaction feedback.

[0083] Optionally, the map construction module includes: The data source submodule is used to collect and integrate multiple apparel data sources, including apparel styles, colors, materials, brands, applicable occasions, seasonal characteristics, user reviews, and fashion trends. The category area module is used to determine the attribute categories of clothing based on the clothing data source. The attribute categories include the appearance attributes, functional attributes, and association attributes with user characteristics of clothing, and to construct clothing attribute nodes based on the attribute categories of clothing. Define the edge submodule to define the relationship edges between clothing attribute nodes, and obtain the clothing relationship edges, which represent the association between clothing attribute nodes; The construction submodule is used to build a clothing knowledge graph using graph database technology, storing and indexing clothing attribute nodes and clothing relationship edges; The optimization submodule is used to acquire external fashion trends, expert recommendations, and user-generated content within a preset time period. Based on these external fashion trends, expert recommendations, and user-generated content, the apparel knowledge graph is supplemented and optimized through data fusion and knowledge extraction technologies.

[0084] Optionally, the semantic matching module includes: The vectorization submodule is used to vectorize the image feature data in the dynamic user profile and the clothing attributes in the clothing knowledge graph. It uses graph embedding technology to map the image feature data and clothing attributes to a low-dimensional vector space. The similarity calculation submodule is used to calculate the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph based on the pre-trained graph embedding model, and obtain the semantic similarity. The sorting submodule is used to sort each clothing attribute node in the clothing knowledge graph according to semantic similarity and generate a preliminary recommended clothing list. The semantic matching submodule is used to generate semantic matching results based on the initial recommended clothing list and preference settings.

[0085] Optionally, the similarity calculation submodule includes: The user embedding unit is used to transform the image feature data in the dynamic user profile into the embedded representation of the user node and generate the user embedding vector. The attribute embedding unit is used to transform clothing attribute nodes in the clothing knowledge graph into low-dimensional vector representations of clothing attribute nodes, and generate clothing attribute node embedding vectors. Calculation unit, used to calculate using the formula: The semantic similarity is calculated, where U is the user embedding vector, A is the clothing attribute node embedding vector, U·A represents the dot product of vector U and vector A, ||U|| represents the magnitude of vector U, and ||A|| represents the magnitude of vector A.

[0086] Optionally, the logical reasoning module includes: The mapping submodule is used to map scenario requirements to clothing attributes in the clothing knowledge graph, generating a preliminary clothing candidate list; The preliminary reasoning submodule is used to infer preliminary clothing matching combinations based on the relationship edges between clothing attributes in the preliminary clothing candidate list and clothing knowledge graph, combined with scenario requirements. The feedback submodule is used to optimize the initial clothing combination based on user feedback and obtain logical reasoning results.

[0087] Optionally, the suggestion generation module includes: The preliminary suggestion submodule is used to filter out the required clothing items based on the semantic matching results, and combine the results of logical reasoning to adjust the order and combination of clothing to generate preliminary clothing matching suggestions. The fashion prediction submodule is used to predict fashion trends for clothing combinations based on the popular trend nodes in the clothing knowledge graph, and obtain the clothing fashion trend prediction results. The "Generate Outfit Suggestions" submodule integrates preliminary outfit suggestions and fashion trend predictions to generate outfit suggestions, including complete outfit combinations, individual item recommendations, and fashion trend predictions.

[0088] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0089] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A virtual clothing matching recommendation method based on knowledge graph, characterized in that, The knowledge graph-based virtual clothing matching recommendation method includes: Acquire user virtual avatar data, extract user avatar feature data from the virtual avatar data, and combine it with the user's historical purchase records, social behavior data and preference settings to construct a dynamic user profile; Construct a clothing knowledge graph, which includes clothing attribute nodes and clothing relationship edges; A semantic matching algorithm based on graph embedding is used to embed the dynamic user profile and the clothing knowledge graph, calculate the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generate semantic matching results. Based on logical reasoning technology, the system generates clothing combination combinations for users according to the scenario requirements set by the users, and generates logical reasoning results. Based on the semantic matching results and logical reasoning results, clothing matching suggestions are generated, and users can make interactive adjustments and provide feedback.

2. The virtual clothing matching recommendation method based on knowledge graph according to claim 1, characterized in that, The process of acquiring user virtual avatar data, extracting user avatar feature data from the virtual avatar data, and constructing a dynamic user profile by combining the user's historical purchase records, social behavior data, and preference settings includes: The virtual avatar data uploaded by the user is analyzed using computer vision technology to identify and extract the avatar feature data, which includes body shape features, skin color features, hairstyle features, and facial expressions. Based on the user's historical purchase records, the user's purchase preference information is analyzed and extracted, including clothing style, color, material and brand; Analyze the social behavior data, which includes user interactions, likes, and comments on social platforms, to obtain information on changes in user interests; By combining the purchase preference information, the interest change information, the preference settings, and the image feature data, the dynamic user profile is generated, and the dynamic user profile is dynamically updated based on user interaction feedback.

3. The virtual clothing matching recommendation method based on knowledge graph according to claim 1, characterized in that, The construction of the clothing knowledge graph includes: Collect and integrate multiple apparel data sources, including apparel styles, colors, materials, brands, applicable occasions, seasonal characteristics, user reviews, and fashion trends; Based on the clothing data source, determine the attribute categories of the clothing. The attribute categories include the appearance attributes, functional attributes, and association attributes with user characteristics of the clothing. Then, construct the clothing attribute nodes based on the attribute categories of the clothing. Define the relationship edges between the clothing attribute nodes to obtain the clothing relationship edges, which represent the association between clothing attribute nodes; The clothing knowledge graph is constructed using graph database technology, and the clothing attribute nodes and clothing relationship edges are stored and indexed. The system acquires external fashion trends, expert recommendations, and user-generated content within a preset time period. Based on these external fashion trends, expert recommendations, and user-generated content, it supplements and optimizes the clothing knowledge graph through data fusion and knowledge extraction techniques.

4. The virtual clothing matching recommendation method based on knowledge graph according to claim 1, characterized in that, The graph embedding-based semantic matching algorithm embeds the dynamic user profile and the clothing knowledge graph into a representation, calculates the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generates semantic matching results including: The image feature data in the dynamic user profile and the clothing attributes in the clothing knowledge graph are vectorized, and graph embedding technology is used to map the image feature data and clothing attributes to a low-dimensional vector space. Based on a pre-trained graph embedding model, the semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph is calculated to obtain the semantic similarity. Based on the semantic similarity, each clothing attribute node in the clothing knowledge graph is sorted, and a preliminary recommended clothing list is generated. Based on the preliminary recommended clothing list and the preference settings, the semantic matching result is generated.

5. The virtual clothing matching recommendation method based on knowledge graph according to claim 4, characterized in that, The semantic similarity between the dynamic user profile and each clothing attribute node in the clothing knowledge graph is calculated based on the pre-trained graph embedding model, and the semantic similarity includes: The image feature data in the dynamic user profile is transformed into an embedded representation of the user node, generating a user embedding vector; The clothing attribute nodes in the clothing knowledge graph are transformed into low-dimensional vector representations of clothing attribute nodes, generating clothing attribute node embedding vectors. Through the formula: The semantic similarity is calculated, where U is the user embedding vector, A is the clothing attribute node embedding vector, U·A represents the dot product of vector U and vector A, ||U|| represents the magnitude of vector U, and ||A|| represents the magnitude of vector A.

6. The virtual clothing matching recommendation method based on knowledge graph according to claim 1, characterized in that, The aforementioned logical reasoning technology, based on the user's defined scenario requirements, generates clothing combination combinations for the user, and the generated logical reasoning results include: The scenario requirements are mapped to the clothing attributes in the clothing knowledge graph to generate a preliminary clothing candidate list; Based on the relationship edges between the preliminary clothing candidate list and the clothing attributes in the clothing knowledge graph, and combined with the scenario requirements, preliminary clothing matching combinations are deduced. Based on user feedback, the initial clothing combinations are optimized to obtain the logical reasoning result.

7. The virtual clothing matching recommendation method based on knowledge graph according to claim 1, characterized in that, The process of generating clothing matching suggestions based on the semantic matching results and logical reasoning results includes: Based on the semantic matching results, the required clothing items are selected, and combined with the logical reasoning results, the order and combination of clothing are adjusted to generate preliminary clothing matching suggestions. Based on the trend nodes in the clothing knowledge graph, fashion trend prediction is performed on clothing matching to obtain clothing fashion trend prediction results. The preliminary clothing matching suggestions and the clothing fashion trend prediction results are integrated to generate the clothing matching suggestions, which include complete clothing matching, individual item recommendations, and fashion trend predictions.

8. A virtual clothing matching recommendation system based on knowledge graphs, characterized in that, The knowledge graph-based virtual clothing matching recommendation system includes: The feature extraction module is used to acquire user virtual avatar data, extract user avatar feature data from the virtual avatar data, and construct a dynamic user profile by combining the user's historical purchase records, social behavior data and preference settings. A graph construction module is used to construct a clothing knowledge graph, which includes clothing attribute nodes and clothing relationship edges; a semantic matching module is used to embed the dynamic user profile and the clothing knowledge graph into a semantic matching algorithm based on graph embedding, calculate the semantic similarity between the dynamic user profile and the clothing knowledge graph, and generate semantic matching results. The logical reasoning module is used to generate clothing matching combinations for users based on the user's set scenario requirements, using logical reasoning technology, and to generate logical reasoning results. The suggestion generation module is used to generate clothing matching suggestions based on the semantic matching results and logical reasoning results, and supports users to make interactive adjustments and provide feedback.

Images