Brand oriented security administrator engine in item list system

Abstract

Methods, systems, and computer storage media for providing brand oriented security management in an item listing system are described. By using a brand-oriented machine learning model, a brand-oriented security administrator engine supports the generation of different integrated reports and analyses of brand-related security metrics, including brand compliance, infringement events, law enforcement actions taken, and overall item list system integrity. The brand-oriented security administrator engine also supports visualization of brand-oriented security analysis result data in a manner that enhances availability and decision making by presenting complex brand-oriented security analysis result data in a clear, operable format. In operation, brand oriented security data is accessed. Brand-oriented security data is analyzed using a brand-oriented machine learning model trained based on a multi-dimensional authenticity analysis dataset. And generating brand-oriented security analysis result data. The brand-oriented security analysis result data is transmitted such that one or more visualizations are displayed based on the brand-oriented security analysis result data.

Description

TECHNICAL FIELD

[0001] Various aspects of the technology described herein relate generally to systems, methods, and computer storage media for providing brand-oriented artificial intelligence management, etc. using a brand-oriented artificial intelligence system associated with an item listing system. BACKGROUND

[0002] Users utilize item listing systems to browse product listings, make purchases, and sometimes also sell items. When a seller lists an item, particularly a branded item, they must adhere to guidelines set by the platform and ensure the authenticity and accuracy of their listing. For users purchasing items, branded items often enjoy a reputation of quality and authenticity, which influences purchase decisions. However, if an item is listed with counterfeit or false claims, it can damage the trust between the buyer and the item listing system. In this context, item listing providers execute strategies to prevent fraudulent listings, protect brand integrity, and maintain a positive user experience to uphold their reputation and legal obligations. For example, brand safety tools can include trademark monitoring, anti-counterfeiting measures, digital brand protection, intellectual property enforcement, and market surveillance. SUMMARY

[0003] A brand-oriented artificial intelligence system provides brand-oriented artificial intelligence functionality to support different applications and services in an item listing system. The brand-oriented artificial intelligence system includes a brand-oriented machine learning model training engine, a brand-oriented safety administrator engine, and a brand-oriented listing management engine. In this way, the brand-oriented artificial intelligence system supports the item listing system by automating brand-related tasks such as machine learning training, safety administration and analysis, and item listing quality enforcement.

[0004] A brand-oriented machine learning model training engine supports training of a brand-oriented machine learning model that predicts a brand for an item listing that does not include brand information. Model training includes brand data processing, model selection, and optimization algorithms. Specifically, brand-oriented machine learning model training is the process of teaching a machine learning model (using a multi-dimensional authenticity analysis dataset) to recognize patterns and make predictions or decisions based on input data. This training can be based on novel training techniques and training features in data associated with the multi-dimensional authenticity analysis dataset (i.e., brand protection and verification data and item listing system data) to generate a brand-oriented machine learning model. For example, the training is based on brand protection and verification data (e.g., verified rights owner (VeRO) data and post-sale authentication (PSA) data) and item listing system data (e.g., item listing text and / or images) that support identifying highly relevant brand features for machine learning to support brand prediction for item listings.

[0005] The brand-oriented machine learning model can be a multi-modal model that includes text-based models and image recognition models that can be selectively implemented to support functionality (e.g., applications in the item listing system). The brand-oriented machine learning model can be deployed to operate with and in association with applications and services of the brand-oriented security administrator engine and the brand-oriented item listing management engine.

[0006] The brand-oriented security administrator engine supports the generation of comprehensive reports and analysis of brand-related security metrics, including brand compliance rates, infringement events, enforcement actions taken, and overall item listing system integrity. The brand-oriented security administrator engine also supports the visualization of brand-oriented security analysis result data in a manner that enhances usability and decision making by presenting complex brand-oriented security analysis result data in a clear, actionable format.

[0007] The brand-oriented item listing management engine supports listing accuracy compliance with item listing marketplace standards and optimization of search visibility and customer experience. The brand-oriented listing management engine also supports the identification of product listing guidelines that are intentionally bypassed or violated by the item listing system and the taking of corrective actions to improve listing quality. Further, the brand-oriented listing management engine improves customer satisfaction by improving CSAT scores through improvements; reduces instances of bad purchase experiences, thereby minimizing potential losses and attracting more customers to the item listing platform; and facilitates growth in total merchandise transaction value within the item listing system platform.

[0008] In this manner, the brand-oriented machine learning model training engine trains and deploys brand-oriented machine learning models; the brand-oriented security administrator tool leverages the novel brand-oriented machine learning models to provide novel functionality associated with brand monitoring and detection, dashboards, real-time alerts and notifications, and trend analysis graphs; and the brand-oriented listing management tool leverages the novel brand-oriented machine learning models to provide novel functionality associated with listing management, including seller listing processes and visibility optimization based on listing quality signals.

[0009] This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the DETAILED DESCRIPTION. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used to determine the scope of the claimed subject matter. BRIEF DESCRIPTION OF DRAWINGS

[0010] The technology described herein is described in detail below with reference to the attached drawing figures, wherein:

[0011] FIG. 1A and FIG. 1Bis a block diagram of a brand-oriented artificial intelligence system for providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0012] FIGS. 2A-2C is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0013] FIG. 3 is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0014] FIG. 4 is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0015] FIG. 5 is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0016] FIG. 6 is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0017] FIGS. 7A-7C is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0018] FIG. 8 is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0019] FIGS. 9A-9B is a schematic and interface associated with providing brand-oriented artificial intelligence system management in a project listing system in accordance with aspects of the technology described herein;

[0020] FIG. 10 is provided a first exemplary method of providing brand-oriented machine learning model training engine functionality in a project listing system in accordance with aspects of the technology described herein;

[0021] FIG. 11 is provided a second exemplary method of providing brand-oriented machine learning model training engine functionality in a project listing system in accordance with aspects of the technology described herein;

[0022] FIG. 12A third example method of providing brand-oriented machine learning model training engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0023] FIG. 13 A first example method of providing brand-oriented security administrator engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0024] FIG. 14 A second example method of providing brand-oriented security administrator engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0025] FIG. 15 A third example method of providing brand-oriented security administrator engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0026] FIG. 16 A first example method of providing brand-oriented listing management engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0027] FIG. 17 A second example method of providing brand-oriented listing management engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0028] FIG. 18 A third example method of providing brand-oriented listing management engine functionality in a project listing system according to aspects of the technology described herein is provided;

[0029] FIG. 19 A block diagram of an example project listing system computing environment suitable for implementing aspects of the technology described herein is provided;

[0030] FIG. 20 A block diagram of an example distributed computing environment suitable for implementing aspects of the technology described herein is provided; and

[0031] FIG. 21 A block diagram of an example computing environment suitable for implementing aspects of the technology described herein. DETAILED DESCRIPTION

[0032] SUMMARY

[0033] An item listing system and platform supports storing items (products or assets) in an item database and provides a search system for receiving queries and identifying search result items based on the queries. An item (e.g., a physical item or a digital item) refers to a product or asset offered for listing on an item listing platform. The search system supports identifying result items from the item database for a received query. The item database can be dedicated for a content platform or an item listing platform, such as the EBAY content platform developed by EBAY Inc. of San Jose, California. The item listing system can also provide AI- supported applications (“AI applications”) that utilize AI models (e.g., natural language processing models (NLP) and image recognition models) to perform computational tasks. The applications use AI to perform various tasks across different domains, including product recommendations, search optimization, sentiment analysis, product categorization, customer support, and spam and fraud detection.

[0034] In the fast-paced world of e-commerce, ensuring that product listings are complete and accurate is important for e-commerce platform sellers, buyers, and suppliers. For example, Jane, an experienced online seller, is preparing to list a new set of wireless earbuds on her preferred item listing platform. However, due to her haste, she forgets to fill in the “brand” field before submitting the listing. This seemingly minor oversight can lead to a range of technical and user experience issues that can impact everything from search functionality to customer trust. When brand information is omitted from an item listing on an item listing platform, it presents significant technical challenges for combating counterfeiting. Automated detection systems struggle to verify product authenticity without brand data, while manual reviews are more labor-intensive and prone to error. Incomplete brand information hinders trend analysis and reporting, leading to data gaps that impact insights into counterfeiting activity. This data deficiency erodes user trust, complicates compliance with legal regulations, and reduces the effectiveness of user reporting mechanisms. To address these issues, platforms must enforce the brand field, enhance verification processes, educate sellers, and work closely with brands.

[0035] Item listing system providers implement strategies to prevent fraudulent listings, protect brand integrity, and maintain a positive user experience to uphold their reputation and legal obligations. Brand safety can refer to measures taken within an online marketplace to protect a brand’s identity, reputation, and intellectual property. It involves safeguarding against various threats, including counterfeiting, unauthorized reselling, trademark infringement, brand impersonation, and brand dilution. Brand safety can include trademark protection, counterfeiting prevention, authorized seller management, brand monitoring, brand identity protection, cybersecurity, and legal enforcement.

[0036] Conventionally, item listing systems are not configured with comprehensive logic and infrastructure to effectively provide brand-oriented artificial intelligence functionality for item listing systems. These systems are unable to provide machine learning model training that sufficiently leverages artificial intelligence to provide brand-oriented secure machine learning models, secure administrator tools, and listing management tools.

[0037] Item listing systems lack machine learning engines, techniques, and data to sufficiently train machine learning models to effectively and comprehensively support item listing systems. Machine learning model training can be based on limited and biased data, and thus the model can not sufficiently generalize to a wide variety of brands. Further, without sufficient contextual understanding (e.g., understanding a brand based on knowledge of the industry, market trends, and seller and buyer behavior), a trained model can misinterpret brand-related signals or fail to capture subtle brand associations.

[0038] Item listing systems also lack sufficiently trained machine learning models that support providing brand-oriented security. For example, brand security refers to measures taken within an item listing system to protect a brand’s identity, reputation, and intellectual property. Item listing systems can collect data based on item categories and other data that focuses on brand characteristics. However, without proper brand security data, brand monitoring tools, a security administrator is unable to visualize and identify fraud trends to secure the brands associated with the item listing system.

[0039] Item listing systems also typically lack sufficiently trained machine learning models to effectively support item listing functionality and features. For example, item listing platforms can encounter challenges in balancing the tradeoff between a seamless user experience and accurate listing information. On one hand, there can be a tendency to require users to input all necessary data during the item listing process, while on the other hand, there can be a desire to simplify the seller listing process without requiring all item listing information. Further, the lack of brand-oriented machine learning models can limit the ability to identify fraudulent listings and reduce the impact of fraudulent listings on item listing systems.

[0040] Accordingly, conventional item listing systems can be improved by overcoming these limitations based on leveraging advanced machine learning models and techniques and item listing system tools that can enhance machine learning functionality for protecting brands in item listing systems. Thus, a more comprehensive item listing system (with the alternative infrastructure to perform brand-oriented artificial intelligence management) can improve computing operations and interfaces to provide brand protection in item listing systems.

[0041] Embodiments of the present invention relate to systems, methods, and computer storage media for providing brand-oriented artificial intelligence management, among other things, using a brand-oriented artificial intelligence system associated with an item listing system. The brand-oriented artificial intelligence system provides brand-oriented artificial intelligence functionality to support different applications and services in the item listing system. The brand-oriented artificial intelligence system includes a brand-oriented machine learning model training engine, a brand-oriented security administrator engine, and a brand-oriented listing management engine.

[0042] Description of Technical Solution

[0043] At a high level, item listing system providers enforce brand security policies within the platform. To simplify the item listing process, item listing systems can allow sellers to leave certain fields, including the brand field, blank. This omission can occur intentionally or unintentionally. When the brand field is left blank, the system must employ various techniques, such as natural language processing, image recognition, or database cross-referencing, to infer the brand of the item.

[0044] The lack of a specified brand can severely hinder a provider’s ability to enforce brand security agreements. For example, without explicit brand identification, it becomes challenging to implement brand-specific security measures. Additionally, malicious sellers can intentionally omit brand information to circumvent security policies designed to protect specific brands. This vulnerability requires a robust brand inference and verification mechanism to maintain the integrity and security of the item listing system.

[0045] The brand-oriented artificial intelligence system provides a machine learning engine for training a machine learning model that can predict the brand of an item listing that does not include brand information. The machine learning training engine is a software system or framework designed to facilitate the process of training a machine learning model. It typically includes functionality for data preprocessing, model selection, hyperparameter tuning, and optimization algorithms to iteratively improve the performance of the model based on a given dataset.

[0046] The machine learning model is trained based on different data sources (e.g., project listing system data; verified rights owner (VeRO) data and post sale authentication (PSA) data) and further trained to take project listing information (i.e., indicators) for a project listing (e.g., title, category, listing site, and images) to predict what brand the project listing should be. The machine learning model is trained using algorithms (e.g., word embedding algorithms and image recognition algorithms) to generate a text and / or image based machine learning model that supports the predicted brand. Importantly, the machine learning is trained based on a multi-dimensional authenticity analysis dataset that has specific training data (i.e., verification and authentication data and historical project listing system data) for training brand-oriented machine learning models. In this way, based on input of a project listing with missing brand information, a predicted brand can be identified.

[0047] Predicting missing brand information can significantly improve the ability to measure and fight the rate of counterfeits, as well as improve the accuracy of detecting counterfeits (capture) and identifying false negatives (misses). Enhancing counterfeit detection can be based on improved algorithm accuracy. Predicting and filling in missing brand information improves the quality of data used to train machine learning algorithms, resulting in more accurate detection of counterfeits. Having brand information ensures that data patterns remain consistent, making it easier for algorithms to identify anomalies that can indicate counterfeit goods.

[0048] Since the predicted brand information allows listings to be cross-referenced with existing databases of authentic and known counterfeit goods, better cross-referencing can be achieved, enhancing the identification process. With brand data, the system can more effectively filter and flag suspicious listings, increasing the chances of capturing counterfeit goods. Measuring the rate of counterfeits can be based on accurate metrics for a brand (e.g., Brand A or Brand B), rather than just a category (e.g., handbags or shoes). Predicting missing brand information creates a more complete dataset, which provides a more accurate measure of the prevalence of counterfeits. Improved data allows for more reliable trend analysis, helping to identify patterns in counterfeit activity over time. Comparative analysis can also be performed by comparing counterfeit detection rates before and after brand prediction is implemented, and the platform can measure improvements in identifying counterfeit goods. With accurate brand data, the platform can generate brand-specific rates of counterfeits, providing valuable insights to both platform administrators and brand owners.

[0049] Additionally, identification capture and miss can be improved as the prediction of missing brand information reduces the number of reports from third parties or external entities (miss) by enabling the algorithm to capture more instances of fakes. The improved data quality gives the detection results higher credibility, ensuring that fewer fake items go undetected. Better performance tracking is possible as the item listing platform can better track the performance of its detection system by comparing the rate of capturing fakes (capture) to the rate of missing fakes and refining its metrics using the predicted brand information. Detailed analysis of capture and miss allows for continuous improvement of the detection algorithm, making it more effective over time.

[0050] Additionally, the prediction of missing brand information significantly improves the quality of item listings in multiple ways. First, it improves searchability and discoverability by allowing products to appear in brand-specific searches and filters, enhancing the overall search experience and potentially increasing sales. Second, complete listings with brand information establish user trust and confidence, leading to higher customer satisfaction and fewer returns or disputes. Third, including brand information ensures consistency across all listings, making the platform appear more professional and user-friendly. Fourth, accurate brand information improves recommendation algorithms, enhancing the personalization of the shopping experience and increasing user engagement.

[0051] Additionally, complete brand information enables the generation of more accurate analytics and insights, providing valuable data for both the platform and sellers to refine their strategies. Moreover, predicted brand information aids in detecting fake items, reducing the risk of fraudulent listings and improving the platform's reputation for hosting authentic products. Finally, accurate brand information ensures compliance with legal and regulatory requirements, helping the platform avoid legal issues and maintain its integrity. Overall, by predicting and filling in missing brand information, the item listing system can significantly improve the quality of item listings, resulting in a better user experience, increased buyer trust, and improved performance and reputation for the platform.

[0052] For example, a brand-oriented machine learning model can include one or more machine learning models. A brand-oriented machine learning model can be a word embedding model or a text classification model. A particular word embedding machine learning algorithm can be used as a particular implementation within the broader category of word embedding and text classification models (also including other models such as FastText, Word2Vec, and GloVe). Machine learning algorithms can be tools used for natural language processing tasks, particularly when dealing with large vocabularies, morphologically rich languages, or when focusing on computational efficiency. Word embedding and text classification models are designed to convert words and text into numerical vectors that capture semantic meaning, thereby facilitating various natural language processing tasks. For example, a machine learning algorithm can be a Word2Vec model, which further includes the ability to handle sub-word information, which is particularly useful for morphologically rich languages and for dealing with words outside of a vocabulary.

[0053] A brand-oriented machine learning model can be or include an image recognition model. An image recognition model can be based on a deep convolutional neural network (CNN), which is designed to facilitate training of very deep networks by addressing the vanishing gradient problem. For example, ResNet is a deep convolutional neural network (CNN) that performs well on image classification and object detection tasks. It introduces the concept of residual learning through skip connections, which allow gradients to flow more easily through network layers, enabling networks with hundreds or even thousands of layers to be effectively trained. The architecture of ResNet is typically composed of multiple residual blocks, where each block contains several convolutional layers and a skip connection that bypasses one or more of these layers, thereby adding the input directly to the output. This approach helps to maintain performance and accuracy even as the depth of the network is significantly increased.

[0054] By using image recognition models and word embedding models, the brand-oriented machine learning model (as a multi-modal model) can process and analyze both textual data and visual data, thereby improving its ability to automatically identify brands. The brand-oriented machine learning model can combine text classification functionality with image recognition functionality, thereby forming a multi-modal or multi-input model. This approach enables the model to simultaneously process and extract features from both textual descriptions and associated images. The brand-oriented machine learning model can analyze product descriptions and images to provide a comprehensive understanding of the listed items. For example, the text classification portion of the model can use natural language processing (NLP) techniques to classify products or analyze customer sentiment, while the image recognition component can employ computer vision techniques to perform tasks such as object detection or brand logo recognition. By integrating both text processing capabilities and image processing capabilities, the model can leverage the unique information provided by each modality, thereby improving its ability to make accurate predictions or classifications. This fusion of modalities enables more nuanced insights, such as identifying brands associated with a product listing by leveraging information from both textual descriptions and accompanying images.

[0055] The brand-oriented machine learning model is trained using a multi-dimensional authenticity analysis dataset (i.e., item listing system data, VeRO data, and PSA data “brand protection and verification data”). The multi-dimensional authenticity analysis dataset as training data refers to a comprehensive collection of diverse and interrelated information sources used to train machine learning. This dataset integrates various dimensions related to brand authenticity, legitimacy, and completeness from multiple data sources such as item listing data, brand protection and verification data. It includes rich sources of data covering product descriptions, images, seller information, customer feedback, trademark registrations, copyright applications, and legal documents. Each dimension within this dataset is represented by a set of features (including textual, visual, numerical, and contextual features) extracted from relevant data sources.

[0056] VeRO (Verified Rights Owner) data refers to information provided by rights owners to online marketplaces to help protect their intellectual property. Rights owners, such as trademark or copyright owners, submit VeRO reports to inform the marketplace of listings that infringe upon their intellectual property. When rights owners identify listings that infringe upon their intellectual property, they can submit a VeRO report to the online marketplace. The report includes details of the infringement, such as the specific listing URL, the type of infringement (e.g., trademark infringement, copyright infringement), and evidence supporting the claim. Once the marketplace receives the VeRO report, they review the information provided and can take action, such as removing the infringing listing, contacting the seller to resolve the issue, or even suspending the seller's account in cases of repeated violations. VeRO data helps online marketplaces maintain the integrity of their platforms by addressing intellectual property infringement and ensuring a fair and safe environment for both buyers and legitimate sellers.

[0057] Post-sale authentication data refers to information and verification records collected after a transaction is completed to confirm the authenticity of the purchased item, which is especially important in e-commerce platforms that sell high-value, branded, or luxury items. This data includes transaction details, expert physical inspection reports, certification certificates from recognized authorities, buyer feedback and claims, tracking and shipping records, return and refund data, and historical sales data of similar items. It ensures that the item delivered to the buyer is genuine and matches the seller's description, maintaining trust and credibility in the marketplace. By verifying the sold item, the platform protects buyers from counterfeit goods, improves buyer satisfaction, and upholds market integrity.

[0058] VeRO (Verified Rights Owner) data, post-sale authentication data, and item listing system data are valuable for training machine learning models to predict brands that are missing brand information in item listings. Machine learning models can predict brands and further provide metadata associated with the predicted brand and any related brand safety and item listing information (e.g., brand features) from VeRO data, PSA data, and item listing system data. VeRO data includes verified brand information, trademark and rights ownership details, which serve as the ground truth for brand identification. Post-sale authentication data includes transaction details, physical inspection reports, certification certificates, buyer feedback, tracking and shipping records, return and refund data, and historical sales data, providing a comprehensive understanding of item authenticity and brand characteristics. Item listing system data also includes transaction details, seller data, buyer feedback, tracking and shipping records, return and refund data, and historical sales data.

[0059] Features such as brand names, logos, associated trademarks, product descriptions, categories, and legal documents can be extracted from VeRO data. PSA data provides features from transaction records, inspection reports, customer feedback, and historical sales data, enabling a comprehensive understanding of brand characteristics. Project listing system data can include project titles, project descriptions, images, listing sites, prices, categories, and product categories. By combining textual features (e.g., project descriptions, customer reviews) with visual features (e.g., project images, logos) and incorporating transaction sales data and historical sales data, a robust feature set is created for training the model.

[0060] Convolutional Neural Networks (CNNs) are particularly effective at extracting high-level features from images, such as logo patterns, product design elements, and color schemes. Transaction data from PSA records includes details about seller identification, sales volume, pricing trends, and buyer feedback. This data is converted into numerical features representing patterns of transactions, historical sales trends, and levels of customer satisfaction. VeRO data provides verified brand names, trademarks, and legal ownership information. Features extracted from this data include brand ownership history, trademark registration details, and previous infringement reports, which are crucial for understanding brand authenticity and legal compliance. Contextual information such as geographic data, temporal trends, and user behavior patterns can be integrated to provide additional insights into the authenticity and market presence of brands. For example, geographic data might reveal regional popularity of certain brands, while temporal data might indicate seasonal trends in brand sales.

[0061] VeRO data, PSA data, and project listing system data are used to train machine learning models to predict brands that are missing brand information in project listings. VeRO data serves as a verified source of brand information, ensuring that the model is trained based on accurate and legally identified brand details. This data serves as a foundation of truth for model training, enhancing the reliability and credibility of brand predictions. PSA data provides detailed insights into the post-transaction verification process, including physical inspections and customer feedback. This information ensures the identification of counterfeit items and verifies the authenticity of brand listings, making it an important component of the feature set. Project listing system data includes structured information about products available for sale, maintained by e-commerce platforms or any similar online marketplaces. This data includes various attributes and details about each item listed on the platform, facilitating search, discovery, and purchase by customers.

[0062] A variety of features are introduced that span textual, visual, and transactional dimensions in conjunction with VeRO, PSA data, and project listing system data. This diversity allows the model to capture different aspects of brand characteristics, from legal trademarks to customer perception and sales patterns. Integrating visual features from product images (enabled by PSA data) with textual and legal features from VeRO data allows the model to more effectively detect counterfeit listings. For example, discrepancies between visual logos and textual brand descriptions can indicate potential fraud. PSA data adds a layer of context such as transaction history and buyer behavior that helps the model understand the market context of a brand. This contextual information is crucial for making accurate brand predictions, especially in cases where brand names are missing or unclear. In this way, the model can predict brands and further provide metadata (e.g., brand features) associated with the predicted brand and any related brand safety and project listing information from VeRO data, PSA data, and project listing system data.

[0063] Machine learning algorithms such as natural language processing techniques for text analysis, convolutional neural networks for image recognition, and supervised or semi-supervised learning for integrating these features can be employed to predict missing brand information. The dataset incorporates ground truth labels and annotations derived from brand protection and verification data and expert judgment to provide a reference point for authenticity assessment and model training. It evolves over time to adapt to changes in market dynamics, legal regulations, and brand-related activities, capturing temporal trends and longitudinal patterns in brand authenticity in conjunction with real-time updates and historical data. Scalable and flexible, the dataset includes new data sources, features, and dimensions as needed, supporting a wide variety of machine learning algorithms and analysis techniques for authenticity analysis and brand safety. Overall, the multi-dimensional authenticity analysis dataset serves as a comprehensive and diverse resource for training machine learning models, facilitating effective brand prediction and detection and preventing counterfeit, infringing, or false representation brand listings.

[0064] The multi-dimensional authenticity analysis dataset can significantly enhance the training of machine learning models to predict brands with missing brand information in project listings by leveraging its rich and diverse data sources and features. The multi-dimensional authenticity analysis dataset integrates various data sources, including project listing data (e.g., product descriptions, images, seller information) and brand protection and verification data (e.g., trademark registrations, legal documents). This diverse dataset provides a comprehensive understanding of brand characteristics and associations.

[0065] Relevant features are extracted from the integrated dataset to represent different aspects of the brand. For example: textual features, extracting keywords, phrases, and semantic patterns from product descriptions, reviews, and legal documents; visual features, analyzing product images and logos to identify visual patterns associated with a particular brand; contextual features, using seller information, geographic data, and temporal trends to understand the context in which the brand appears.

[0066] Through feature fusion techniques, these multiple features can be fused into a unified representation using feature fusion. This can involve concatenation, weighted sum, or advanced methods such as attention mechanisms to ensure that the model can effectively utilize complementary information. Feature fusion in machine learning involves combining features from multiple data sources to create a more comprehensive and informative representation for model training. In the context of brands that lack brand information in the project list, feature fusion enables the integration of multiple data types, such as text, visual, and transactional features, to improve the accuracy and robustness of the model. Natural language processing (NLP) techniques can be used to process textual data from the project list, including product titles, descriptions, and customer reviews. These methods, such as TF-IDF (Term Frequency-Inverse Document Frequency), word embeddings (e.g., Word2Vec, GloVe), or advanced language models (e.g., BERT), convert textual information into numerical vectors that capture semantic meaning. Image recognition techniques are used to process visual data, such as product images and brand logos.

[0067] In implementing feature fusion, textual data is cleaned and labeled, images are preprocessed for input into CNNs, and transactional data is normalized. Legal documents are parsed to extract relevant brand information. NLP techniques are used to extract textual embeddings, CNNs are used to extract visual embeddings, and statistical methods are used to extract transactional features. Legal features are encoded based on trademark and ownership details. Then, the features from different sources are combined into a single feature vector. Techniques such as concatenation, weighted sum, or advanced methods like attention mechanisms can be used to integrate these features. The integrated feature vector is used to train machine learning models, such as neural network models or gradient boosting models, generative AI models, and LLMs.

[0068] In the context of machine learning (unsupervised learning, transfer learning, generative AI, and large language models), machine learning model training can be used. For example, these techniques improve the ability to fuse multiple features from multiple data modalities, resulting in richer, more robust, and versatile machine learning models across various tasks and domains. By leveraging each other's strengths, such as discovering underlying patterns, transferring learned knowledge, generating synthetic data, and understanding complex linguistic contexts, these techniques synergistically facilitate more effective feature fusion and ultimately improve overall model performance.

[0069] Unsupervised machine learning models aim to find patterns and structure in data without explicit supervision or labeled outcomes. They are used for tasks such as clustering (grouping similar data points together) and dimensionality reduction (reducing the number of variables under consideration). In the context of feature fusion, unsupervised learning can be employed to combine features from different sources or modalities into cohesive representations that capture underlying patterns or relationships in the data.

[0070] Transfer learning involves leveraging knowledge gained in solving one problem to help solve a different but related problem. In machine learning, this often means transferring knowledge from a pre-trained model (typically on a large dataset) to a new model for a different task or domain. Transfer learning can facilitate feature fusion by allowing a model to incorporate features learned from one domain to another, thereby improving the model's ability to generalize and perform well on new tasks.

[0071] Generative AI refers to models and techniques that create new instances of data similar to the training data. This includes models such as generative adversarial networks (GANs) and variational autoencoders (VAEs). In terms of feature fusion, generative AI can synthesize new features or representations that capture complex relationships between original features or across different data modalities, thereby enriching the feature space available for training models.

[0072] Large language models are deep learning models trained on vast amounts of text data to understand and generate human language. Examples include models such as the GPT (Generative Pretrained Transformer) family. By extracting semantic features from text or integrating text information with other types of data (such as images or structured data), LLMs can be used for feature fusion. This integration can improve the richness and context of features used by machine learning models.

[0073] In this way, techniques and feature fusion are able to combine information from multiple sources or modalities to create unified representations that improve model performance. Unsupervised learning techniques can help merge features by discovering patterns and relationships between different data sources. Transfer learning facilitates the integration of features learned from one domain to another, thereby improving the robustness and effectiveness of fused features. Generative AI and LLMs contribute by generating new features or embeddings that capture complex dependencies and semantics, thereby enriching the feature space used to train more powerful machine learning models.

[0074] By using the fused features, a machine learning model (such as a neural network, random forest, or support vector machine, generative AI model, and LLM) is trained. Even if there is no explicit brand information in the item listing, the model learns to recognize patterns and associations indicative of a particular brand. The ground truth labels from the brand protection and verification data and expert annotated dataset provide a solid foundation for supervised learning. The model can compare its predictions against these labels during training to adjust and improve its accuracy. The performance of the model is evaluated using metrics such as accuracy, precision, recall, and F1 score. Based on these evaluations, the model is iteratively refined and retrained to improve its predictive capabilities. Hyperparameter tuning and iterative feature selection are performed to optimize the model performance. By leveraging feature fusion, the integration of VeRO data, PSA data, and item listing system data creates a robust multi-dimensional dataset that significantly enhances the ability of the machine learning model to accurately predict brands in item listings. This comprehensive approach ensures that the model is able to make informed predictions based on a diverse array of complementary information sources, thereby improving brand safety and authenticity verification on e-commerce platforms. Once trained, the model can be used to predict brands in new item listings that lack explicit brand information. This helps maintain the integrity and authenticity of brand listings on the platform.

[0075] When deployed within the item listing system, the model can automatically predict and fill in missing brand information in item listings, providing sellers with suggestions or automatic updates to ensure brand accuracy. Additionally, the model can flag suspicious listings where the predicted brand does not match the information provided by the seller or known brand patterns, thereby enhancing fraud detection and prevention. By leveraging the multi-dimensional authenticity analysis dataset, the platform can significantly improve its ability to accurately predict and verify brand information, thereby improving the overall quality and credibility of the marketplace.

[0076] Consider a scenario in the item listing system where a new item listing is received without explicit brand information. The item listing system utilizes a machine learning model trained on a multi-dimensional authenticity analysis dataset to address this situation.

[0077] Upon receiving the item listing, the item listing system processes the listing through the trained machine learning model. The model utilizes text features from the product description, visual features from the product images, and contextual features from the seller information to make predictions about the missing brand. For example, if the product description mentions specific attributes or characteristics that are typically associated with a particular brand, or if the product images contain recognizable brand logos or trademarks, the model can use this information to infer the most likely brand associated with the item.

[0078] Once the missing brand is identified, the platform can take proactive measures to strengthen brand safety. If the predicted brand matches the registered trademarks or intellectual property data (e.g., VeRO data) stored in the platform’s database, the listing will be approved and flagged with the correct brand information. However, if the predicted brand does not match any known trademarks, or if there are discrepancies that indicate potential counterfeit or infringement activity, the listing is flagged for further review.

[0079] In addition to brand identification and safety enforcement, the predictions of the machine learning model contribute to improving the overall listing quality. By automatically filling in the missing brand information for incomplete listings, the model improves the completeness and accuracy of product listings on the platform. This can improve the user experience for shoppers by providing them with more informative and trustworthy product information. Furthermore, by ensuring consistency and compliance with brand information standards, the listing system fosters trust between sellers and buyers, thereby increasing user satisfaction and engagement. This ensures that only authentic brand listings are allowed on the listing system, thereby maintaining brand safety and protecting the rights of brand owners.

[0080] Advantageously, embodiments of the present technical solution use a brand-oriented artificial intelligence system associated with the listing system to support providing brand-oriented artificial intelligence management. The brand-oriented artificial intelligence system provides resources for implementing a brand-oriented machine learning model training engine, a brand-oriented safety administrator engine, and a brand-oriented listing management engine. The brand-oriented artificial intelligence system operates to provide a solution to problems (e.g., limited integration of artificial intelligence in brand protection) in the listing system. The ordered combination of brand-oriented artificial intelligence system components, infrastructure, and steps is an improvement over conventional listing systems that lack support for brand-oriented artificial intelligence system brand protection functionality.

[0081] For a brand-oriented machine learning model training engine, conventional item listing systems have limited understanding of items (e.g., items or products listed in the item listing system) and, in particular, brand information associated with the items. In particular, in the context of products, items, or services, a brand can refer to a logo or name that is associated with the overall perception and reputation of the product, service, or company in the minds of consumers. Accordingly, a brand-oriented machine learning model is trained based on a multi-dimensional authenticity analysis dataset (with specific training data for training the brand-oriented machine learning model). For example, machine learning model training can be based on metadata and unique identifiers from VeRO data and PSA data, item listing system data for sellers and buyers, text, and data. Integrating these data categories allows for an overall approach for counterfeit detection and brand authenticity assessment. This comprehensive approach significantly improves the accuracy and reliability of predictions related to counterfeit detection and brand authenticity.

[0082] For a brand-oriented security administrator engine, conventional item listing systems have limited brand-oriented security measures, security planning and monitoring, and security visualization tools. Accordingly, a brand-oriented machine learning model trained based on a multi-dimensional authenticity analysis dataset is deployed in the item listing system. The brand-oriented model machine learning model is integrated into a brand-oriented security administrator engine that supports monitoring, analysis, and enforcement of brand security measures. The brand-oriented machine learning model focuses on protecting the integrity and authenticity of brands represented on the item listing system while reducing the risks associated with counterfeits and unauthorized brand usage. The brand-oriented security administrator engine generates comprehensive reports and analysis of brand-related security metrics, including brand compliance rates, infringement events, enforcement actions taken, and overall item listing system integrity. The brand-oriented security administrator engine also supports visualization of brand-oriented security analysis result data in a way that enhances usability and decision making by presenting complex brand-oriented security analysis result data in a clear, actionable format. In this way, the brand-oriented security administrator tool leverages the novel brand-oriented machine learning model to provide novel functionality associated with brand monitoring and detection, dashboards, real-time alerts and notifications, and trend analysis graphs.

[0083] For a brand-oriented item listing management engine, a regular item listing system has limited item listing functionality that is supported by machine learning functionality, in particular, brand-oriented machine learning models. Accordingly, a brand-oriented machine learning model trained on a multi-dimensional authenticity analysis dataset is deployed in the item listing system. The brand-oriented model machine learning model is integrated into a brand-oriented listing management engine that supports listing management functionality, including seller listing processes and listing quality signal analysis. The brand-oriented item listing management engine supports listing accuracy compliance with item listing marketplace standards and optimization of search visibility and customer experience. The brand-oriented listing management engine also supports identification of intentional circumvention or violation of product listing guidelines of the item listing system and takes corrective action to improve listing quality. In this way, the brand-oriented listing management tool leverages novel brand-oriented machine learning models to provide novel functionality associated with listing management, including seller listing processes and visibility optimization based on listing quality signals.

[0084] Example systems and resources

[0085] Aspects of the technical solution can be described by way of example and with reference to FIGS. 1A-1B FIG. 1A An item listing system 100 is shown, including: an item listing system with integrated artificial intelligence 100A; a brand-oriented artificial intelligence system 110, including item listing system services 100B with integrated API 100C; and brand-oriented applications 100D; brand-oriented artificial intelligence system resources 112; multi-dimensional authenticity analysis datasets 114; brand-oriented machine learning model training engine 120A; brand-oriented security administrator engine 120B; brand-oriented listing management engine 120C; and brand-oriented machine learning model 140; item listing system client 130A; item listing system client 130B; and item listing system client 130C.

[0086] ​The item listing system 100 provides a system (e.g., item listing system with integrated artificial intelligence 100A) that includes systems (e.g., brand-oriented artificial intelligence system 110) and engines (e.g., brand-oriented machine learning model training engine 120A; brand-oriented security administrator engine 120B; brand-oriented listing management engine 120C) for providing brand-oriented artificial intelligence system functionality using resources (e.g., brand-oriented artificial intelligence system resources). For example, brand-oriented artificial intelligence system resources 112 include operational, interface, and data resources. Operations include brand monitoring, sentiment analysis, product classification, and fraud detection, thereby ensuring continuous vigilance against brand-related violations, customer sentiment, and fraudulent activity. Interfaces include user-friendly dashboards, APIs for seamless integration, and reporting tools for administrators and brand owners to effectively monitor and manage their brand image. Data sources include brand protection and verification databases, product listings, customer feedback, and market data, thereby providing comprehensive insights into brand protection strategies, law enforcement actions, and market trend analysis. These resources help maintain brand integrity, thereby fostering customer trust and maintaining regulatory compliance within the item listing ecosystem.

[0087] The brand-oriented artificial intelligence system 110 can also include item listing system services 110B corresponding to different services of the item listing system. The item listing system services can include brand prediction services, brand security services, and listing management services, for example, employing brand-oriented machine learning models 140 to provide item listing system functionality. The item listing system services 100B can include integrated application programming interfaces (APIs) 110C for integrating the item listing system services with the brand-oriented machine learning models 140. Brand-oriented applications 110D can also operate to employ the brand-oriented machine learning models 140 to provide functionality associated with the generative AI applications 110D. Other variations and combinations of item listing system services are contemplated by the embodiments described herein.

[0088] The item listing system 100 can also include item listing system clients (e.g., item listing system client 130A (an administrator client), item listing system client 130B (a seller client), and item listing system client 130C (a buyer client)). Administrators, who are responsible for overseeing the functionality of the item listing system 100, employ the item listing system client 130A to configure and maintain the item listing system 100, establish categories, attributes, and monitor user activity to ensure compliance with platform policies. Sellers, as part of the product offering of the platform, utilize the item listing system client 130B to create and manage listings, update product details, pricing, and inventory levels, while handling orders and addressing customer inquiries. Interacting through a dedicated dashboard, sellers proactively manage their storefronts, adjust listings, and seamlessly communicate with buyers. Buyers, the end users of the platform, rely on the item listing system 130C via the item listing system client 130C to explore, evaluate, and purchase products. They navigate through the intuitive interface of the system to browse product listings, filter results, and view detailed information before making a purchase decision. Through interaction with the platform, buyers seamlessly complete transactions, track orders, and receive updates throughout the process. The item listing system clients facilitate the client functionality associated with the brand-oriented artificial intelligence system 110.

[0089] Referring to FIG. 1B , FIG. 1B The item listing system 100 is shown with components in FIG. 1A and further includes brand protection and verification data 114A, item listing data 114B, security data 150, and item listings 160.

[0090] Brand-oriented machine learning model training engine

[0091] The brand-oriented machine learning model training engine 120A accesses a multi-dimensional authenticity analysis dataset 114 associated with a plurality of brands and uses the multi-dimensional authenticity analysis dataset 114 to train a brand-oriented machine learning model 140. The multi-dimensional authenticity analysis dataset 114 includes intellectual property enforcement data, authenticated memorabilia data, and item listing system data (e.g., brand protection and verification data 114A and item listing system data 114B) that provide ground truth labeling and annotations associated with model training.

[0092] The brand-oriented machine learning model 140 is trained based on brand multidimensional authenticity features. The brand-oriented machine learning model 140 is trained based on machine learning techniques that incorporate brand multidimensional authenticity features corresponding to intellectual property enforcement data, authenticated merchandise data, and project listing system data. The brand-oriented machine learning model 140 is trained to receive project listing information for a project listing (e.g., project listing 160) that does not have a brand as input for generating a predicted brand, which is used as an indicator with features related to generating a predicted brand.

[0093] The brand-oriented machine learning model training engine 120A deploys the brand-oriented machine learning model 140 in the project listing system to support one or more applications (e.g., brand-oriented application 100D). The one or more applications include a brand-oriented security administrator tool and a brand-oriented listing management tool. The brand-oriented machine learning model 140 is configured to generate a predicted brand for a project listing that does not include a brand, where the predicted brand is generated based on project listing information associated with the project listing. The brand-oriented machine learning model 140 is a multi-modal model that includes a text-based model and an image recognition model. A first predicted brand score is associated with the text-based model and a second predicted brand score is associated with the image recognition model, such that the generated predicted brand is based on a combination of the first predicted brand score and the second predicted brand score.

[0094] Brand-oriented security administrator engine

[0095] The brand-oriented security administrator engine 120B accesses brand-oriented security data 150. The brand-oriented can be a project listing that does not include brand information, or a combination of project listings with and without brand information. Accessing the brand-oriented security data is based on a request by a security administrator for brand-oriented security analysis result data that is associated with a brand identifier in the request. The brand identifier is used to indicate that brand-oriented security analysis result data should be generated for the brand identifier.

[0096] The brand-oriented security administrator engine 120B uses the brand-oriented machine learning model 140 trained based on the multidimensional authenticity analysis dataset 114 and brand multidimensional authenticity features to analyze brand-oriented security data 1250. The brand-oriented machine learning model 140 is integrated into a brand-oriented security administrator tool that employs the brand-oriented security administrator engine 120B of the project listing system 100 to support generating brand-oriented security analysis result data.

[0097] The brand-oriented security administrator engine 120B generates brand-oriented security analysis result data. Generating brand-oriented security analysis result data is based on: identifying a predicted brand for a project list, where the project list does not include a brand input; extracting project list information associated with the project list; and updating existing brand-oriented security analysis result data for the predicted brand based on the extracted project list information.

[0098] The brand-oriented security administrator engine 120B can also generate brand security strategies based on the brand-oriented security analysis result data. Generating brand strategies based on brand-oriented security analysis involves analyzing data on security incidents, reputation, and customer sentiment. Specifically, generating brand strategies can include identifying threats, assessing brand awareness, and defining strategy goals. The process includes developing a communication plan, implementing security measures, and monitoring performance.

[0099] The brand-oriented security administrator engine 120B communicates the brand-oriented security analysis result data to enable display of one or more visualizations based on the brand-oriented security analysis result data. The brand-oriented security administrator engine 120B can create a security incident trend chart or a geospatial incident heat map. For example, a visualization displays a trend chart showing the frequency and distribution of security incidents over time. Each incident type (e.g., data breach, phishing attack, malware infection) is represented by a different color or symbol on the chart. The x-axis represents time (e.g., days, weeks, months), while the y-axis shows the number of incidents. This chart helps visualize the overall trends of security incidents affecting the brand and identify periods of increased activity or vulnerability.

[0100] Another visualization presents a geospatial heat map showing the geographic distribution of security incidents. Each incident is plotted on the map, with color intensity representing the concentration or severity of incidents in different regions. Stakeholders can zoom into specific regions to identify hotspots of security activity or vulnerability. This visualization helps understand the geographic distribution of security threats and prioritize mitigation measures for high-risk areas.

[0101] Brand-oriented list management engine

[0102] The brand-oriented list management engine 120C accesses a project list associated with a project list system and analyzes the project list using a brand-oriented machine learning model 140 trained based on the multi-dimensional authenticity analysis dataset 114 and brand multi-dimensional authenticity features. The brand-oriented machine learning model 140 is integrated into a brand-oriented list management tool that employs the brand-oriented list management engine 120C of the project list system 100 to support brand-oriented list management.

[0103] Based on the analysis of the item listing, the brand-oriented listing management engine 120C generates a brand-oriented security notification associated with the item listing. The generation of the brand-oriented security notification is based on identifying a predicted brand for the item listing, where the item listing does not include a brand input. The brand-oriented security notification is a communication or alert specifically designed to inform stakeholders, such as customers, partners, and employees, of a security event or threat that directly impacts the reputation, credibility, or integrity of the brand.

[0104] The brand-oriented listing management engine 120C communicates the brand-oriented security notification. The brand-oriented security notification is communicated to the seller associated with the item listing and a security administrator, or administrator, of the item listing system, or another service of the brand-oriented listing management engine 120C. Based on the brand-oriented security notification, the brand-oriented listing management engine 120C performs a listing quality analysis based in part on brand-oriented security data associated with the brand-oriented security notification. The listing quality analysis involves a comprehensive evaluation of the item listing, covering various attributes such as title, description, images, price, and reviews.

[0105] The listing quality analysis can focus on fraud, such that the listing quality analysis involves a meticulous examination of the item listing to identify potential fraudulent activities or listings. This comprehensive assessment includes a careful review of various attributes such as product title, description, images, price, and seller information. Data collection involves gathering information from the listing and analyzing it for consistency, irregularities, or red flags that may indicate fraudulent behavior. Attributes such as excessively low prices, misleading descriptions, or suspicious seller profiles are flagged for further investigation. Content analysis delves into the accuracy and authenticity of product information, ensuring that the listing is transparent and compliant with platform policies. Image quality assessment verifies the legitimacy of product images and their correspondence to the listed item. Pricing analysis includes comparing the price to market averages and identifying anomalies that suggest fraudulent pricing practices.

[0106] Additionally, the seller's reputation and past transaction history are also examined to assess credibility and detect patterns of fraudulent behavior. Recommendations for fraud prevention and mitigation strategies are then generated based on the analysis results, focusing on strengthening the listing verification process, implementing stricter seller review processes, and enhancing fraud detection algorithms. Continuous monitoring and iteration are performed to adapt to evolving fraud strategies and ensure the integrity and security of the item listing system. Based on the listing quality analysis, the item listing can be flagged for remedial actions.

[0107] In one embodiment, listing management can include restricting the visibility of listings until brand information is confirmed or corrected. Listings with uncertain or unverified brand details can be temporarily hidden from general view, reminding sellers to update and verify this information in a timely manner. Initially, if a listing is missing verified brand details, these listings are flagged and hidden from general view, reminding sellers to update and confirm the information in a timely manner through a user interface. Machine learning models predict the brand based on listing attributes such as description and images, verify or suggest corrections. Sellers are notified to review and modify brand information, once verified, triggering a reevaluation and updating the visibility status.

[0108] Turning to FIG. 2A , FIG. 2A An example item listing interface 200A associated with missing brand information is shown. The item listing interface 200A for the wallet item 202A has an image showing the wallet design details as well as a descriptive heading 206A specifying the style and any unique features. Price and availability information is presented in the item listing interface section 208A, allowing a user to evaluate the value and inventory status of the item. The item listing interface section 210A includes item details and item description from the seller. However, the brand information is missing and is listed as “No Brand.” The brand information can have been omitted inadvertently or intentionally. Thus, a brand-oriented artificial intelligence system can be used to analyze this type of item listing to predict the brand, enforce brand safety policies, or listing management.

[0109] Turning to FIG. 2B , FIG. 2BAn example brand-oriented artificial intelligence system associated with providing brand-oriented artificial intelligence management is shown in accordance with embodiments described herein. At a high level, inputs 210 are processed and analyzed via a machine learning (ML) model 230B (e.g., a brand-oriented machine learning model). The inputs 210 serve as the basis for the ML model 230B to make predictions to generate outputs 240 (e.g., predicted brands). The brand-oriented artificial intelligence system integrates different types of applications (or services) that employ the output 240 information. The inputs 210 can include a listing title, a listing site 214B, a category 216B, an item description 218B, a listing image 220, feedback 222B, a return messaging 224B, machine-to-machine (M2M) messages, and item details 228B. Each input serves as an indicator to the ML model 230 to predict a brand for a listing of an item associated with the inputs 210 that does not have a brand. For example, the listing site 214B can employ a custom approach to evaluate fraud patterns, market dynamics, payment methods, language and cultural factors, and geographic location data. Fraud activities often vary based on geographic location. Certain types of scams or fraudulent behavior can be more prevalent in specific regions due to cultural factors, economic conditions, or technical infrastructure. Analyzing regional fraud patterns helps develop fraud detection strategies to address specific risks associated with a brand.

[0110] The ML model 230B can include a word embedding model 232B and an image recognition model 234B. It is contemplated that the ML model 230B can operate independently or in combination, or as a multi-modal model to provide brand prediction functionality. For example, the two models operate independently, each producing its own predicted brand score for a given listing of an item. However, to improve accuracy and reliability, an ensemble method such as a voting ensemble, where the predicted brand scores from each model are treated as “votes” for the predicted brand, can be used to combine the predicted brand scores from the two models. The final prediction is determined by aggregating these votes, such as taking the mean or weighted mean of the scores. This approach leverages the collective insights of multiple models to make a more informed decision. A stacking ensemble uses the predicted brand scores from the two models as features to train a meta-model that learns how to best combine these scores to make a final prediction. This meta-model effectively learns the strengths and weaknesses of each base model and best integrates their predictions to improve overall performance.

[0111] In addition to generating predicted brand scores to predict missing brands within a listing, various types of information can be generated and integrated into the output 240B to enhance the accuracy and depth of the analysis. In addition to the predicted missing brands themselves, key contextual details such as product descriptions, categories, and price ranges also provide valuable insights into potential brand associations. Analyses of similar products, user reviews or ratings, and seller information can further enrich the understanding by shedding light on common brand mentions, sentiment, and seller-brand relationships. Utilizing image recognition technology can provide visual cues from product images that help identify logos or trademarks associated with specific brands. Additionally, accessing external databases or repositories of brand information enables comprehensive comparison and verification of predicted brands against established brand catalogs. By synthesizing these different sources of information, machine learning models can generate more accurate predictions, ultimately facilitating improved decision-making and data-driven insights in the field of brand identification and analysis within listings.

[0112] The output 240B can be communicated to application-associated proactive counterfeit detection 252B, listing quality signals 254B, report insights and analysis 256B, and seller listing processes 258B. Proactive counterfeit detection 252B can include practices that actively identify and prevent the sale of counterfeit or fraudulent items before they are listed or sold on the platform. It involves processes such as detecting suspicious listings, verifying product authenticity, and taking preemptive actions to remove or block fraudulent sellers or listings. Listing quality signals 254B are indicators or measures used to assess the quality, accuracy, and reliability of listings on the platform. Listing quality signals 254B can include factors such as detailed product descriptions, high-quality images, accurate categorization, competitive pricing, positive user reviews, and seller reputation. Monitoring and analyzing these signals help the platform ensure a positive user experience, maintain trust, and reduce risks such as counterfeit items or misleading listings.

[0113] Report insights and analysis 256B involve collecting, analyzing, and deriving actionable insights from reports submitted by users, moderators, or automated systems regarding various aspects of platform activity, such as suspicious behavior, policy violations, or user feedback. Report insights and analysis enable the platform to identify trends, patterns, and areas of concern, make informed decisions, and implement targeted intervention measures to address issues and improve platform integrity and user satisfaction.

[0114] Seller listing flow 258B refers to the step-by-step process that sellers follow to create and publish listings on the platform. Seller listing flow 258B typically includes tasks such as entering product details, uploading images, setting price and inventory information, selecting categories or tags, and submitting the listing for review or publication. Optimizing the seller listing flow involves simplifying the process, providing guidance or tools to improve listing quality, and implementing checks or verifications (e.g., missing brand information) to prevent errors or policy violations, ultimately improving the efficiency and effectiveness of sellers' activities on the item listing platform.

[0115] Turning to FIG. 2C , FIG. 2C An example brand-oriented machine learning model flow associated with generating predicted brands is shown in accordance with embodiments described herein. For example, the brand-oriented machine learning model can be a word embedding model that acts as input 202C, and applies hashing 204C, averaging 206C, linear transformation 208C, and softmax 210C operations on the pre-processed reviews 212C to generate word vectors 214C, hidden review representations 215C, transformed review vectors 218C, and predicted brand outputs 220C.

[0116] Input preprocessing: Preprocess the annotations (e.g., L1 preprocessed annotations) to remove punctuation, convert text to lowercase, and tokenize the words.

[0117] Hashing 204C: Hashing is used to map each word to a fixed-size vector representation. Instead of using a traditional lookup table for word embeddings, a hash function is applied to convert each word into a fixed-size hash value. This ensures efficient memory usage and scalability, especially for large vocabularies.

[0118] Averaging 206C: Average the hashed word vectors (e.g., L2 N word vectors) to obtain a single vector representation for the entire review. This average vector captures the overall semantics of the review by aggregating information from individual word embeddings.

[0119] Hidden review representation 210C: The hidden review vector (e.g., L3 hidden review representation) serves as a hidden representation of the review. This vector encapsulates learned features and semantic information extracted from the review, which is crucial for sentiment analysis.

[0120] Linear transformation 208C: Apply a linear transformation to the average vector to project it onto a higher-dimensional space (e.g., L4 transformed review vector). This transformation helps capture complex relationships and patterns in the review data that may not be captured by the original word embeddings.

[0121] Predicted output: Finally, the hidden review representation is fed into a classifier (such as a softmax layer) to predict the brand of the review (e.g., GUCCI, COACH, or LOUIS VUITTON). The softmax operation computes the probability of each possible sentiment label based on the learned features. Each brand label corresponds to a class, and the classifier outputs the predicted brand label with the highest probability as the predicted brand of the item listing. The output of the classifier provides the predicted brand label based on the learned features and representation extracted from the review data.

[0122] Turning to FIG. 3 , FIG. 4 , FIG. 5 and FIG. 6 , these figures illustrate corresponding interfaces and diagrams associated with brand safety. For example, these interfaces can be associated with a safety administrator tool that supports reviewing details of individual item listings and their corresponding details. Referring to FIG. 3 , FIG. 3 , a chart 302 is shown that presents an overall view of missed events that occur throughout the year, for example, focusing on December, April, and August. The x-axis 304 of the chart represents the progression of time divided by the months of the year (i.e., December, April, and August). Meanwhile, the y-axis 306 quantifies the total number of missed events recorded during each respective month. For each month, a data point is plotted on the chart to represent the total number of missed events observed. These data points are interconnected to form lines (e.g., 310 and 320), thereby creating different visual representations for December, April, and August.

[0123] In addition to showing missed event data on a monthly basis, the chart includes dotted lines to divide two components: annual target misses (e.g., 310) and historical baseline (e.g., 310). The dotted line indicating annual target misses 310 acts as a benchmark or predetermined goal to measure performance against. These lines extend horizontally across the chart, intersecting with each monthly data point. They provide a clear visual reference, enabling stakeholders to assess whether the total missed events for a given month aligns with the target established for the year. Furthermore, the chart features a historical baseline 320 represented by another dotted line. This baseline reflects the average or historical trend of missed events observed throughout the year. It acts as a comparison point, allowing viewers to assess current performance relative to past patterns.

[0124] In this context, "bps" refers to basis points, a measurement unit used to quantify small percentage changes. In the context of this chart, bps can indicate deviation from a historical baseline or target miss expressed in percentage points. For example, if actual miss events exceed the baseline or fail to meet the target, the deviation can be quantified in bps to provide a precise understanding of the size of the difference. Overall, this chart provides a comprehensive visualization of miss events, facilitating informed decision making and strategic planning by providing insights into performance relative to targets and historical trends.

[0125] Referring to FIG. 4 , FIG. 5 and FIG. 6 , FIG. 4 shows an example dashboard interface 402 in which a chart (e.g., chart 410 and chart 420) is associated with a total list of suspicious items identified using an active approach (e.g., y-axis 412) and a total list of suspicious items identified using a passive approach (e.g., y-axis 422). The chart can be plotted on a weekly basis (e.g., x-axis 412 and x-axis 422). For example, chart 410 provides a visual representation of the total number of lists of suspicious items identified on a weekly basis by an active approach. The x-axis of this chart represents the progression of time, broken into weeks, while the y-axis quantifies the total count of suspicious lists. Active approaches for detecting actual or potential ("suspicious") fraud lists involve employing a proactive strategy and advanced algorithms to identify potential fraud instances before they escalate. These approaches often involve data analysis, machine learning algorithms, and pattern recognition techniques to detect anomalies or suspicious patterns in list behavior. Each data point on chart 410 represents the cumulative count of lists of suspicious items identified by the active approach during a particular period. These data points are interconnected to form a line graph that illustrates the trend of active detection efforts over time. This chart can demonstrate fluctuations in the number of suspicious lists identified week-over-week, reflecting changes in the effectiveness of active detection measures or changes in patterns of fraudulent activity.

[0126] Turning to the second chart 420, which illustrates lists of suspicious items identified using a passive approach, chart 430 depicts the total number of lists of suspicious items identified on a weekly basis by a passive approach. Similar to chart 410, the x-axis represents time in weeks, while the y-axis indicates the total count of suspicious lists.

[0127] The passive approach for detecting actual or potential ("suspect") listings involves instances of reporting suspicious activity or anomalies identified in response to customer complaints, manual review, or post-transaction monitoring. These approaches rely on manual intervention and investigative techniques to resolve fraudulent activity after it has already occurred. Each data point on chart 420 represents the cumulative count of suspect listing identified by the passive approach during a particular period. Similar to the first chart 410, these data points are connected to form a trend line, which shows the trajectory of passive detection efforts over time. This chart can reveal fluctuations in the number of suspect listings identified by the passive approach, reflecting changes in the number of reporting incidents, the efficiency of the investigation process, or changes in consumer behavior patterns. Collectively, these two charts provide valuable insights into the effectiveness of both the active and passive approaches for detecting suspect listings, enabling stakeholders to assess the effectiveness of their fraud detection strategies and make informed decisions to reduce risk.

[0128] Referring to FIG. 5 and FIG. 6 These figures show a dashboard interface 502 and gauges and dashboard 602. The dashboard interface 502 includes a first interface portion 510 that provides selectable filters (filter 512 and filter 514) that can be used to filter data associated with the dashboard interface. The dashboard interface 602 includes a second interface portion 610 that provides sources (source 3 612 and source 9 614) that indicate corresponding counts or percentages of active and / or passive listings associated with the corresponding sources. The dashboard interface can be associated with different types of brand safety functionality associated with brand monitoring, listing analysis, geographic analysis, user feedback and reporting, interface features, and graphs and charts.

[0129] For example, the dashboard can show dynamic trend graphs that further elucidate the pattern of fluctuations in fraudulent activity over time, enabling administrators to identify emerging trends and respond accordingly. The dashboard facilitates vigilant brand monitoring, which is a critical component of fraud detection and prevention. Administrators are presented with an intuitive interface through which they can manage the listings of monitored brands, making it easy to add or remove supervised entities. Real-time alerts are seamlessly integrated, immediately notifying administrators of any suspect listings associated with a monitored brand, ensuring prompt intervention to protect brand integrity.

[0130] In addition, the dashboard can also support presenting data associated with the analysis of each list flagged as fraudulent. Each flagged list is meticulously categorized with thumbnails for quick visual recognition. Relevant details such as seller information, listing date, price, and location are meticulously categorized, providing administrators with a comprehensive understanding of current fraudulent activities. Administrators can delve deeper into each listing, accessing a wealth of detailed information and actionable insights, enabling them to take decisive measures such as removing listings or banning sellers.

[0131] The dashboard can provide geographic visualization, presenting administrators with an interactive map interface depicting the spatial distribution of fraudulent listings. A dynamic heat map overlays the map, vividly illustrating areas with the highest density of fraudulent activities. Administrators can seamlessly drill down to specific regions or cities, facilitating targeted interventions to curb fraudulent activities at their source.

[0132] The dashboard includes robust user feedback and reporting mechanisms. Users are empowered to report suspicious listings or activities, contributing to the collective effort to combat fraud. A dedicated section supports flagging potential fraudulent activities for administrative review. Flagged listings are highlighted, ensuring they immediately grab the attention of administrators. Clicking on a flagged listing reveals a wealth of details, enabling administrators to delve deeper into the intricate details of each case. From product descriptions to seller histories, every aspect of a listing is meticulously documented, enabling administrators to make informed decisions. Furthermore, the interface possesses seamless communication tools, enabling administrators to directly interact with sellers, facilitating dialogue and driving solutions. For escalated cases, an effective escalation mechanism is established, ensuring serious incidents are promptly reported to the appropriate authorities for further action. Administrators can leverage the power of data analysis to analyze fraudulent item listings, identifying recurring patterns or emerging trends to strengthen fraud prevention measures.

[0133] Additionally, a wealth of graphs and charts can be accessed via the dashboard, transforming raw data into actionable insights. Time series graphs track the trajectory of fraudulent activities over time, enabling administrators to discern temporal patterns and predict future trends. Pie charts depict the distribution of fraudulent listings across various product categories, revealing areas that require intensified scrutiny. Bar graphs illustrate top brands involved in fraudulent activities, facilitating targeted interventions to maintain brand integrity. Finally, interactive heat maps provide a spatial perspective, highlighting regions grappling with rampant fraudulent activities, guiding administrators to strategically deploy resources for effective fraud combat. In essence, the dashboard is capable of implementing brand safety measures associated with brand-oriented machine learning models to identify fraudulent activities, provide administrators with the tools and insights needed to maintain the integrity of the item listing system, and protect the interests of users and stakeholders alike.

[0134] Turning to FIG. 7A , FIG. 7B , FIG. 7C , FIG. 8 , FIG. 9A and FIG. 9B , these figures illustrate corresponding interfaces associated with item listings. For example, these interfaces can be associated with a security administrator tool that supports reviewing details of individual item listings and their corresponding details. FIG. 7A , FIG. 7B and FIG. 7C illustrate cases of true positives, where the brand-oriented machine learning model correctly predicts the correct brand; FIG. 8 illustrate instances of false positives, and FIG. 9A and FIG. 9B illustrate cases of anomalies.

[0135] A true positive occurs when the model correctly identifies that a listing belongs to a particular brand and after verification the prediction is indeed accurate. In essence, it indicates that the model correctly detected the presence of the brand it was designed to identify. A false positive occurs when the model incorrectly predicts the presence of a particular brand in a listing, but after verification the listing does not actually belong to that brand. An anomaly refers to an instance or case that deviates significantly from the norm or expected patterns within a given context (e.g., item listing information). These cases often exhibit characteristics, behaviors, or outcomes that are uncommon, unexpected, or irregular compared to typical occurrences in the domain under consideration. False positives and anomalies can be alternative processing mechanisms for brand safety. For example, these types of cases can be escalated, such as instances flagged as outliers or anomalies being elevated to higher-level review or decision-making within an organization. This can entail sending the anomaly prediction to a domain expert, a senior data scientist, or management for further analysis, verification, or intervention. The goal is to ensure that these cases receive appropriate attention and action, including manual review, model refinement, or process improvement, to enhance the accuracy and reliability of the brand prediction system. The embodiments described herein envision other variants and combinations of handling false positives and anomalies.

[0136] Finally, the confidence score (i.e., the predicted brand score) represents the level of certainty or confidence that the model has in its prediction. It indicates the model's belief in the correctness of its classification output. The confidence score typically ranges between 0 and 1, where 0 indicates low confidence and 1 represents high confidence. For example, if the model assigns a confidence score of 0.8 to a prediction, it means that the model is 80% confident in its classification decision.

[0137] Referring to FIG. 7A , FIG. 7B and FIG. 7CAs discussed, the input data associated with the item can be used to generate a predicted brand. To illustrate, the example item listing information interface 702 is provided, and we will discuss the seller tags shown in the seller tags interface portion 710, including the brands provided by the seller and the images in the image portion 720 of the interface. In FIG. 7A Brand 702A is “No Brand” and processing the item listing via a brand-oriented machine learning model can predict the brand to be “Brand A” with a predicted brand score of: 0.99996. This prediction can be based on the seller tags 704A (e.g., faux leather, letter combination, tweed) and / or the corresponding image 708A. In FIG. 7B Brand 702B is “No Brand” and processing the item listing can predict the brand to be “Brand B” with a predicted brand score of: 1.0. This prediction can be based on the seller tags 704B (e.g., leather, crossbody bag) and / or the corresponding image 706B. In FIG. 7C Brand 702C is “No Brand” and processing the item listing can predict the brand to be “Brand C” with a predicted brand score of: 0.99494. This prediction can be based on the seller tags 704C (e.g., HD03, 1600W, professional hair dryer) and / or the corresponding image 706C.

[0138] Referring to FIG. 8 which shows a false positive. In FIG. 8 Brand 802C is “Many” and processing the item listing can predict the brand to be “Brand X” with a predicted brand score of: 0.800098. This prediction can be based on the seller tags 804C (e.g., paper cloth, all occasions, gift tote bag) and / or the corresponding image 806C.

[0139] Referring to FIG. 9A and FIG. 9B these address exceptional cases. FIG. 9A includes the seller-provided brand 902A “Brand X” but the product line in the seller tags 904A is “Brand Y” and an image of a Brand X wallet. The generated predicted brand can be “Brand X” with a predicted brand score of: 0.9992. However, this information conflicts with the product line. Thus, this can be a candidate for further review (e.g., item listing quality analysis and notification to update the seller tags).

[0140] FIG. 9B includes the seller-provided brands 902B “Brand A, Brand X, Brand Y, Brand Z” and several seller tags (e.g., mixed, leather mixed, mixed) and the predicted brand is “Brand Z” with a predicted brand score of: 0.3422. Thus, this can be a candidate for further review.

[0141] You can use examples and refer to FIG. 1A and FIG. 1B This is how we describe various aspects of a technical solution. FIG. 1A Based on reference FIGS. 19-21 A block diagram of an exemplary technical solution environment for implementing embodiments of the technical solution is described. Typically, this technical solution environment includes a technical solution system suitable for providing an example item list system 100 that can employ the methods of this disclosure. Specifically, FIG. 1A A high-level architecture of an item list system 100 according to an embodiment of this disclosure is shown. In addition to other engines, managers, generators, selectors, or components not shown (collectively referred to herein as "components"), FIG. 1A Project list platform system 100 corresponds to FIG. 1B .

[0142] Example Method

[0143] Reference FIGS. 10-18 Flowcharts illustrate methods for providing brand-oriented AI management capabilities within a project listing system. These methods can be performed using the project listing system described herein. In embodiments, one or more computer storage media have computer-executable or computer-usable instructions that, when executed by one or more processors, cause the one or more processors to perform methods (e.g., computer-implemented methods) within a project listing platform system (e.g., a computerized system or computer system).

[0144] Turning FIG. 10 A flowchart is provided illustrating method 1000 for providing brand-oriented AI management capabilities in a project listing system. At box 1002, a multidimensional authenticity analysis dataset associated with multiple brands is accessed; at box 1004, the multidimensional authenticity dataset is used to train a brand-oriented machine learning model; and at box 1006, the brand-oriented machine learning model is deployed in the project listing system to support one or more applications.

[0145] Turning FIG. 11, a flowchart is provided that illustrates a method 1100 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1102, a multidimensional authenticity analysis dataset associated with a plurality of brands is accessed; at block 1104, a word embedding model of a brand-oriented machine learning model is trained using the multidimensional authenticity dataset; at block 1106, an image recognition model of the brand-oriented machine learning model is trained using the multidimensional authenticity dataset; and the brand-oriented machine learning model is deployed in the project listing system that selectively employs the word embedding model and the image recognition model to predict a brand of a project listing.

[0146] Turning to FIG. 12 , a flowchart is provided that illustrates a method 1200 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1202, a multidimensional authenticity analysis dataset is identified; one or more machine learning algorithms are selected to train a brand-oriented machine learning model; and the brand-oriented machine learning model is deployed for operation with one or more applications of the project listing system.

[0147] Turning to FIG. 13 , a flowchart is provided that illustrates a method 1300 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1302, brand-oriented security data is accessed; at block 1304, the brand-oriented security data is analyzed using a brand-oriented machine learning model trained based on a multidimensional authenticity analysis dataset; at block 1306, brand-oriented security analysis result data is generated; and at block 1308, the brand-oriented security analysis result data is communicated to cause display of one or more visualizations based on the brand-oriented security analysis result data.

[0148] Turning to FIG. 14 , a flowchart is provided that illustrates a method 1400 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1402, a predicted brand of a project listing is identified, the project listing not including a brand input; at block 1404, project listing information associated with a project listing update is extracted; at block 1406, existing brand-oriented security analysis result data of the predicted brand is updated based on the extracted project listing information.

[0149] Turning to FIG. 15, a flowchart is provided showing a method 1500 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1502, brand-oriented security data is communicated; at block 1504, brand-oriented security analysis data is received, the brand-oriented analysis result data being generated using a brand-oriented machine learning model that analyzes the brand-oriented security data. The brand-oriented machine learning model is trained based on multi-dimensional authenticity analysis data; at block 1506, the brand-oriented security analysis result data is caused to be displayed.

[0150] Turning to FIG. 16 , a flowchart is provided showing a method 1600 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1602, a project listing associated with the project listing system is accessed; at block 1604, the project listing is analyzed using a brand-oriented machine learning model trained based on a multi-dimensional authenticity analysis data set; at block 1606, based on analyzing the project listing, a brand-oriented security notification associated with the project listing is generated; and at block 1608, the brand-oriented security notification is communicated.

[0151] Turning to FIG. 17 , a flowchart is provided showing a method 1700 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1702, a brand-oriented security notification related to a project listing is accessed; at block 1704, a listing quality analysis is performed based in part on brand-oriented security data associated with the brand-oriented security notification; and at block 1706, based on the project listing quality analysis, the project listing is flagged for remedial action to be taken.

[0152] Turning to FIG. 18 , a flowchart is provided showing a method 1800 of providing brand-oriented artificial intelligence management functionality in a project listing system. At block 1802, a request to generate a project listing is communicated; at block 1804, a brand-oriented security notification generated using a brand-oriented machine learning model that analyzes the project listing is received. The brand-oriented machine learning model is trained based on a multi-dimensional authenticity analysis data set; at block 1806, the brand-oriented security notification is caused to be displayed.

[0153] Additional support for the detailed description of the invention

[0154] Example project listing system environment

[0155] Reference is now made to FIG. 19 , FIG. 19 An example project listing system 1900 computing environment in which embodiments of the present disclosure can be employed is shown. In particular, FIG. 19A high-level architecture of an example project listing platform 1910 that can host a technical solution environment or a portion thereof is shown. It should be understood that this and other arrangements described herein are set forth as examples. For example, as described above, many of the elements described herein can be implemented as discrete or distributed components or in conjunction with other components, and in any suitable combination and location. Other arrangements and elements (e.g., machines, interfaces, functions, commands, and function groupings) can be used in addition to or instead of the arrangements and elements shown.

[0156] The project listing system 1900 can be a cloud computing environment that provides computing resources for functions associated with the project listing platform 1910. For example, the project listing system 1900 supports the delivery of computing components and services, including servers, memory, databases, networks, applications, and machine learning associated with the project listing platform 1910 and client devices 1920. Multiple client devices (e.g., client devices 1920) include hardware or software that accesses resources on the project listing system 1900. The client devices 1920 can include applications (e.g., client applications 1922) and interface data (e.g., client application interface data 1924) that support client functions associated with the project listing system. Multiple client devices can access computing components of the project listing system 1900 via a network (e.g., network 1930) to perform computing operations.

[0157] The project listing platform 1910 is responsible for providing a computing environment or architecture that includes infrastructure that supports providing project listing platform functions (e.g., e-commerce functions). The project listing platform supports storing projects in a project database and provides a search system for receiving queries and identifying search results based on the queries. The project listing platform can also provide a computing environment with features for managing, selling, buying, and recommending different types of projects. The project listing platform 1910 can be dedicated to a content platform, such as the EBAY content platform or e-commerce platform developed by EBAY Inc. of San Jose, California.

[0158] The project listing platform 1910 can provide project listing operations 1930 and project listing interfaces 1940. The project listing operations 1930 can include service operations, communication operations, resource management operations, security operations, and fault tolerance operations that support specific tasks or functions in the project listing platform 1910. The project listing interfaces 1940 can include service interfaces, communication interfaces, resource interfaces, security interfaces, and management and monitoring interfaces that support functions between project listing platform components. The project listing operations 1930 and project listing interfaces 1940 can enable communication, coordination, and seamless operation of the project listing system 1900.

[0159] By way of example, the functions associated with the item listing platform 1910 can include: shopping operations (e.g., product search and browse, product selection and cart, checkout and payment, and order tracking); user account operations (e.g., user registration and authentication, and user profile); seller and product management operations (e.g., seller registration, product listing and inventory management); payment and financial operations (e.g., payment processing, refunds and returns); order fulfillment operations (e.g., order processing and fulfillment, and inventory management); customer support and communication interfaces (e.g., customer support chat / email and notifications); security and privacy interfaces (e.g., authentication and authorization, payment security); recommendation and personalization interfaces (e.g., product recommendations and customer reviews and ratings); analytics and reporting interfaces (e.g., sales and inventory reporting, and user behavior analytics); and API and integration interfaces (e.g., APIs for third-party integrations).

[0160] The item listing platform 1910 can provide item listing platform databases (e.g., item listing platform databases 1950) to efficiently manage and store different types of data. The item listing platform databases 1950 can include relational databases, NoSQL databases, search databases, cache databases, content management systems, analytics databases, payment gateway databases, customer relationship management databases, log and error databases, inventory and supply chain databases, and multichannel databases, which are used in combination to efficiently manage data and provide an e-commerce experience for users.

[0161] The item listing platform 1910 supports applications (e.g., applications 1960) that are computer programs or software components or services that serve a particular function or set of functions to meet specific item listing platform requirements or user requirements. Applications can be client-side (user-facing) and server-side (backend). Applications can also include applications without any AI support (e.g., applications 1962), applications supported by traditional AI models (e.g., applications 1964), and applications supported by generative AI models (e.g., applications 1966). By way of example, the applications can include online storefront applications, mobile shopping applications, administration and management consoles, payment gateway integrations, user account and authentication applications, search and recommendation engines, inventory and stock management applications, order processing and fulfillment applications, customer support and communication tools, content management systems, analytics and reporting applications, marketing and promotion applications, multichannel integration applications, log and error tracking applications, customer relationship management (CRM) applications, security applications, and API and web services, which are used in combination to efficiently provide an e-commerce experience for users.

[0162] The project listing platform 1910 can include a machine learning engine (e.g., machine learning engine 1970). The machine learning engine 1970 refers to a machine learning framework or machine learning platform that provides infrastructure and tools for designing, training, evaluating, and deploying machine learning models. The machine learning engine 1970 can serve as the backbone for developing and deploying machine learning applications and solutions. The machine learning engine 1970 can also provide tools for visualizing data and model results and interpreting model decisions to understand how the model makes predictions.

[0163] The machine learning engine 1970 can provide necessary libraries, algorithms, and utilities to perform various tasks within a machine learning workflow. The machine learning workflow can include data processing, model selection, model training, model evaluation, hyperparameter tuning, scalability, model deployment, inference, integration, customization, data visualization. The machine learning engine 1970 can include pre-trained models for various tasks, thereby simplifying the development process. In this way, the machine learning engine 1970 can simplify the entire machine learning process, from data preparation and model training to deployment and inference, making it accessible and efficiently usable by different types of users (e.g., customers, data scientists, machine learning engineers, and developers) engaged in a wide range of machine learning applications.

[0164] The machine learning engine 1970 can be implemented in the project listing system 1900 as a component that leverages machine learning algorithms and techniques (e.g., machine learning algorithms 1972) to enhance various aspects of the functionality of the project listing system. The machine learning engine 1970 can provide a range of machine learning algorithms and techniques for teaching computers to learn from data and make predictions or decisions without being explicitly programmed. These techniques are widely applied in a variety of applications across different industries and can include the following examples: supervised learning (e.g., linear regression: classification, support vector machines (SVM); unsupervised learning (e.g., clustering, principal component analysis (PCA), association rules (e.g., apriori); reinforcement learning (e.g., Q-learning, deep Q-networks (DQN); deep learning (e.g., neural networks, convolutional neural networks (CNN), and recurrent neural networks (RNN); and ensemble learning random forests.

[0165] Machine learning training data 120 supports the process of building, training, and fine-tuning machine learning models. Machine learning training data 120 consists of labeled datasets that are used to teach machine learning models to recognize patterns, make predictions, or perform specific tasks. Training data generally includes two main components: input features (X) and labels or target values (Y). Input features can include variables, attributes, or characteristics that are used as inputs to a machine learning model. Input features (X) can be numerical, categorical, or even textual, depending on the nature of the problem. For example, in a model for predicting house prices, input features might include the number of bedrooms, square footage, neighborhood, etc. Labels or target values (Y) include the values that the model aims to predict or classify. Labels represent the desired output or ground truth for each corresponding set of input features. For example, in a spam classifier, the labels would indicate whether each email is spam (i.e., a binary classification). The training process involves presenting the training data to the model, and the model learns to make predictions or decisions by identifying patterns and relationships between the input features (X) and the target values (Y). Machine learning algorithms adjust their internal parameters during training in order to minimize the difference between their predictions and the actual labels in the training data. Machine learning engine 1970 can use historical and real-time data to train models and make predictions, thereby continually improving performance and user experience.

[0166] Machine learning engine 1970 can include machine learning models generated using machine learning engine workflows (e.g., machine learning models 1976). Machine learning models 1976 can include generative AI models and traditional AI models, both of which can be used in project list system 1900. Generative AI models are designed to generate new data (typically in the form of text, images, or other media) based on patterns and knowledge learned from existing data. Generative AI models can be used in a variety of ways, including: content generation, product image generation, personalized product recommendations, natural language chatbots, and content summarization. Traditional AI models encompass a wide range of algorithms and techniques, and can be used in a variety of ways, including: recommendation systems, predictive analytics, search algorithms, fraud detection, customer segmentation, image classification, natural language processing (NLP), and A / B testing and optimization. In many cases, a combination of both generative AI models and traditional AI models can be used to provide a comprehensive and efficient e-commerce experience, combining data-driven insights and creativity.

[0167] Machine learning engines 1970 can be used to analyze data, make predictions, and automate processes to provide a more personalized and efficient shopping experience to users. By way of example, product recommendation search and filtering: pricing optimization, inventory and stock management: customer segmentation, customer churn prediction and retention, fraud detection, sentiment analysis, customer support and chatbots, image and video analysis, and ad targeting and marketing. The specific applications of machine learning within the project listing platform 1910 can vary depending on the specific goals, available data, and resources.

[0168] Example distributed computing system environment

[0169] Referring now to FIG. 20 , FIG. 20 An example distributed computing environment 2000 in which embodiments of the present disclosure can be employed is shown. In particular, FIG. 20 A high-level architecture of an example cloud computing platform 2010 that can host a technology solution environment or a portion thereof (e.g., a data fiduciary environment) is shown. It should be understood that such arrangements and other arrangements described herein are set forth by way of example only. For example, many of the elements described herein can be implemented as discrete or distributed components or in conjunction with other components, and in any suitable combination and location. Other arrangements and elements (e.g., machines, interfaces, functions, orders, and groupings of functions) can be used in addition to or instead of those shown.

[0170] A data center can support a distributed computing environment 2000 that includes a cloud computing platform 2010, racks 2020, and nodes 2030 (e.g., computing devices, processing units, or blades) in the racks 2020. The technology solution environment can be implemented using a cloud computing platform 2010 that runs cloud services across different data centers and geographic regions. The cloud computing platform 2010 can implement a fabric controller 2040 component for providing and managing resource allocation, deployment, upgrade, and management of cloud services. Generally, the cloud computing platform 2010 is used to store data or run service applications in a distributed manner. The cloud computing infrastructure 2010 in a data center can be configured to host and support the operation of endpoints of a particular service application. The cloud computing infrastructure 2010 can be a public cloud, a private cloud, or a dedicated cloud.

[0171] Node 2030 can be configured with a host 2050 (e.g., an operating system or runtime environment) that runs a defined software stack on node 2030. Node 2030 can also be configured to perform specialized functions within cloud computing platform 2010 (e.g., compute nodes or storage nodes). Node 2030 is allocated to run one or more portions of a service application of a tenant. A tenant can refer to a customer that utilizes resources of cloud computing platform 2010. Service application components of cloud computing platform 2010 that support a particular tenant can be referred to as a multi-tenant infrastructure or a tenancy. In this document, the terms service application, application, or service are used interchangeably and broadly refer to any software or software portion that runs on top of a data center or accesses storage and compute device locations within a data center.

[0172] When node 2030 is supporting more than one individual service application, node 2030 can be divided into virtual machines (e.g., virtual machine 2052 and virtual machine 2054). Physical machines can also run individual service applications simultaneously. Virtual machines or physical machines can be configured as individualized computing environments supported by resources 2060 (e.g., hardware resources and software resources) in cloud computing platform 2010. It is contemplated that resources can be configured for a particular service application. Further, each service application can be divided into functional portions such that each functional portion is capable of running on a separate virtual machine. In cloud computing platform 2010, multiple servers can be used to run service applications and perform data storage operations in a cluster. Specifically, servers can independently perform data operations but are exposed as a single device referred to as a cluster. Each server in a cluster can be implemented as a node.

[0173] Client device 2080 can link to service applications in cloud computing platform 2010. Client device 2080 can be any type of computing device described with reference to computing device 2000, e.g., client device 2080 can be configured to issue commands to cloud computing platform 2010. In embodiments, client device 2080 can communicate with service applications by directing communication requests to a virtual internet protocol (IP) and load balancer or other means of a designated endpoint in cloud computing platform 2010. Components of cloud computing platform 2010 can communicate with each other over a network (not shown), which can include, without limitation, one or more local area networks (LANs) and / or wide area networks (WANs). FIG. 20

[0174] Example Computing Environment

[0175] Having briefly described an overview of embodiments of the application, an example operating environment in which embodiments of the application can be implemented is described below to provide a general context for various aspects of the application. Specifically, reference is made first to FIG. 21 ​An example operating environment for implementing embodiments of the application is shown and generally designated as computing device 2100. Computing device 2100 is but one example of a suitable computing environment and is not intended to imply any limitation as to the scope of use or functionality of the application. Neither should computing device 2100 be interpreted as having any dependency relating to or requiring any of the components, or combinations thereof, illustrated.

[0176] The application can be described in the general context of computer code or machine-useable instructions, including computer-executable instructions such as program modules, being executed by a computer or other machine, such as a personal data assistant or other handheld device. Generally, program modules including routines, programs, objects, components, data structures, and the like, refer to code that performs particular tasks or implements particular abstract data types. The application can be practiced in a variety of system configurations, including hand-held devices, consumer electronics, general- purpose computers, more specialty computing devices, and the like. The application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network.

[0177] With reference to FIG. 21 Computing device 2100 includes bus 2110 which directly or indirectly couples the following devices: memory 2112, one or more processors 2114, one or more presentation components 2116, input / output port(s) 2118, input / output components 2120, and power supply 2122. Bus 2110 represents what can be one or more busses (such as an address bus, data bus, or combination thereof). Alternatively, virtual components of the described components and / or functionality can be provided by a virtualized computing device or other cloud computing resource. FIG. 21 Various blocks were shown with dashed, double dashed, and dotted lines to indicate that the depicted components and / or FIG. 21 The diagram of FIG. 2 merely illustrates an example computing device that can be used with one or more embodiments of the application. There is no distinction between such categories as "workstation", "server", "laptop", "handheld device", etc. as all of these categories are within the scope of FIG. 21 and are considered with reference to "computing device".

[0178] Computing device 2100 typically includes a variety of computer-readable media. Computer-readable media can be any available media that can be accessed by computing device 2100 and includes both volatile and nonvolatile media, removable and non-removable media. By way of example, and not limitation, computer-readable media can comprise computer storage media and communication media.

[0179] Computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by the computing device 2100. Computer storage media does not include signals per se.

[0180] Communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. The term "modulated data signal" means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Combinations of the any of the above should also be included within the scope of computer readable media.

[0181] Memory 2112 includes computer storage media in the form of volatile and / or non-volatile memory. The memory can be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid state memory, hard drives, optical drives, etc. Computing device 2100 includes one or more processors that read data from various entities such as memory 2112 or I / O components 2120. Presentation component(s) 2116 present data indications to a user or other device. Exemplary presentation components include a display device, speaker, printing component, vibrating component, etc.

[0182] I / O ports 2118 allow computing device 2100 to be logically coupled to other devices including the I / O components 2120, some of which can be built in. Illustrative components include a microphone, joystick, game pad, satellite dish, scanner, printer, wireless device, etc.

[0183] Additional structural and functional features of embodiments of the technical solution

[0184] Having identified various components used herein, it will be understood that any number of components and arrangements can be employed to implement the desired functionality within the scope of the present disclosure. For example, components in the embodiments depicted in the figures are shown with lines for conceptual clarity. Other arrangements of these and other components can also be implemented. For example, although some components are depicted as single components, many of the elements described herein can be implemented as discrete or distributed components or combined with other components and implemented in any suitable combination and location. Some elements can be omitted entirely. Further, as described below, various functions of one or more entities described herein can be performed by hardware, firmware, and / or software. For example, various functions can be performed by a processor executing instructions stored in memory. Accordingly, other arrangements and elements (e.g., machines, interfaces, functions, orders, and groupings of functions) can be used in addition to or instead of the ones shown.

[0185] Embodiments described in the following paragraphs can be combined with one or more of the specifically described alternatives. In particular, claimed embodiments can contain references to more than one other embodiment in the alternatives. Claimed embodiments can specify further limitations of the claimed subject matter.

[0186] The subject matter of embodiments of the present application is described with specificity herein to meet statutory requirements. However, the description itself is not intended to limit the scope of this patent. Rather, the inventors have contemplated that the claimed subject matter might also be embodied in other specific forms without departing from the spirit or essential characteristics thereof. In particular, the present embodiments are described with reference to the attached figures, which are provided as illustrative examples so as to enable those with ordinary skill in the art to

[0187] For purposes of this disclosure, the word "comprising" is intended to mean the same as the word "including" and the word "accessing" includes "receiving," "referencing," or "retrieving." Further, the word "communication" has the same broad meaning as the words "receiving" or "sending," facilitated by a software or hardware-based bus, receiver, or transmitter using the communication media described herein. Additionally, words such as "a" and "one" encompass the plural as well as the singular. Thus, for example, when there is a reference to "a" feature, this satisfies the constraint that there is "one or more" of the feature. Further, the term "or" includes conjunctive, disjunctive, and both (thus, a or b includes a or b as well as a and b).

[0188] For the foregoing detailed description of purposes, embodiments of the application are described with reference to a distributed computing environment; however, the distributed computing environment described herein is merely exemplary. The components can be configured for performing novel aspects of the embodiments, where the term "configured for" can refer to "programmed to" use code to perform a particular task or implement a particular abstract data type. Furthermore, while embodiments of the application can generally be described with reference to the technology solution environment and schematic diagrams described herein, it should be understood that the described technology can extend to other implementation contexts.

[0189] Embodiments of the application have been described with respect to specific embodiments for purposes of illustration and not limitation. Alternative embodiments will become apparent to those having ordinary skill in the art without departing from the scope of the application.

[0190] From the foregoing, it will be apparent that the application is well adapted to attain all the ends and objectives hereinabove set forth, together with other advantages which are obvious or inherent to the structure. It will be understood that certain features and subcombinations are of utility and can be employed without reference to other features or subcombinations. This is contemplated by and is within the scope of the claims.

[0191] It should be understood that certain features and subcombinations are of utility and can be employed without reference to other features or subcombinations. This is contemplated by and is within the scope of the claims.

Claims

1. A computerized system, comprising: One or more computer processors; as well as A computer memory that stores computer-usable instructions, which, when used by the one or more computer processors, cause the one or more computer processors to perform operations, the operations including: Access brand-oriented security data; The brand-oriented security data is analyzed using a brand-oriented machine learning model trained on a multidimensional authenticity analysis dataset and brand multidimensional authenticity features. Generate brand-oriented security analysis results data; and The brand-oriented security analysis results data are transmitted so that one or more visualizations are displayed based on the brand-oriented security analysis results data.

2. The system according to claim 1, wherein, The brand-oriented machine learning model is integrated into the brand-oriented security administrator tool of the project list system to support the generation of the brand-oriented security analysis results data.

3. The system according to claim 1, wherein, The multidimensional authenticity analysis dataset includes intellectual property enforcement data, certified souvenir data, and item listing system data, which provide basic fact labels and annotations associated with model training.

4. The system according to claim 1, wherein, The brand-oriented machine learning model is trained based on machine learning technology, which integrates multi-dimensional brand authenticity features corresponding to the intellectual property enforcement data, the certified souvenir data, and the project listing system data.

5. The system according to claim 1, wherein, The brand-oriented security analysis results data are generated based on: Identify predicted brands for a list of items, wherein the list of items does not include brand input; Extract the project list information associated with the project list; and The existing brand-oriented security analysis results data of the predicted brand are updated based on the extracted project list information.

6. The system according to claim 1, wherein, Access to brand-oriented security data is based on a security administrator's request for brand-oriented security analytics results data, which is associated with the brand identifier in the request.

7. The system according to claim 1, the method further includes generating a brand security strategy based on the brand-oriented security analysis result data.

8. One or more computer storage media having computer-executable instructions thereon, said computer-executable instructions, when executed by a computing system having a processor and a memory, causing the processor to perform operations, said operations including: Access brand-oriented security data; The brand-oriented security data is analyzed using a brand-oriented machine learning model trained on a multidimensional authenticity analysis dataset and brand multidimensional authenticity features. Generate brand-oriented safety analysis results data; as well as The brand-oriented security analysis results data are transmitted so that one or more visualizations are displayed based on the brand-oriented security analysis results data.

9. The medium according to claim 8, wherein, The brand-oriented machine learning model is integrated into the brand-oriented security administrator tool of the project list system to support the generation of the brand-oriented security analysis results data.

10. The medium according to claim 8, wherein, The multidimensional authenticity analysis dataset includes intellectual property enforcement data, certified souvenir data, and item listing system data, which provide basic fact labels and annotations associated with model training.

11. The medium according to claim 8, wherein, The brand-oriented machine learning model is trained based on machine learning technology, which integrates multi-dimensional brand authenticity features corresponding to the intellectual property enforcement data, the certified souvenir data, and the project listing system data.

12. The medium according to claim 8, wherein, The brand-oriented security analysis results data are generated based on: Identify predicted brands for a list of items, wherein the list of items does not include brand input; Extract the project list information associated with the project list; and The existing brand-oriented security analysis results data of the predicted brand are updated based on the extracted project list information.

13. The medium according to claim 8, wherein, Access to brand-oriented security data is based on a security administrator's request for brand-oriented security analytics results data, which is associated with the brand identifier in the request.

14. The medium according to claim 8, the method further comprising generating a brand security strategy based on the brand-oriented security analysis result data.

15. A computer-implemented method, the method comprising: Access brand-oriented security data; The brand-oriented security data is analyzed using a brand-oriented machine learning model trained on a multidimensional authenticity analysis dataset and brand multidimensional authenticity features. Generate brand-oriented safety analysis results data; as well as The brand-oriented security analysis results data are transmitted so that one or more visualizations are displayed based on the brand-oriented security analysis results data.

16. The method according to claim 15, wherein, The brand-oriented machine learning model is integrated into the brand-oriented security administrator tool of the project list system to support the generation of the brand-oriented security analysis results data.

17. The method according to claim 15, wherein, The multidimensional authenticity analysis dataset includes intellectual property enforcement data, certified souvenir data, and item listing system data, which provide basic fact labels and annotations associated with model training.

18. The method according to claim 15, wherein, The brand-oriented machine learning model is trained based on machine learning technology, which integrates multi-dimensional brand authenticity features corresponding to the intellectual property enforcement data, the certified souvenir data, and the project listing system data.

19. The method according to claim 15, wherein, The brand-oriented security analysis results data are generated based on: Identify predicted brands for a list of items, wherein the list of items does not include brand input; Extract the project list information associated with the project list; and The existing brand-oriented security analysis results data of the predicted brand are updated based on the extracted project list information.

20. The method of claim 15, wherein, Access to brand-oriented security data is based on a security administrator's request for brand-oriented security analytics results data, which is associated with the brand identifier in the request.

Images