AI-based system for dynamic product allocation and categorization in real time

Abstract

An AI-based real-time product matching and categorization system (100) comprising: (a) a product data ingestion and normalisation module configured to collect and pre-process product data from multiple structured and unstructured sources; (b) a natural language understanding (NLU) and feature extraction module configured to extract product attributes using machine learning and natural language processing techniques; (c) a dynamic taxonomy assignment and category prediction module configured to assign products to appropriate categories using adaptive classification models; (d) a real-time attribute enrichment and validation module configured to complete, standardise and validate the extracted attributes; (e) an adaptive learning and feedback integration module configured to refine predictive models based on user, merchant and system feedback; f) a real-time API and integration module for connecting to external systems; g) and a monitoring, control and explanation module configured to track performance and provide interpretable decision insights.

Description

[0001] The present invention relates to an AI-based system for dynamic, real-time product assignment and categorization. It enables the automatic extraction, assignment, and updating of product attributes through machine learning and natural language processing. The invention ensures accurate, scalable, and adaptable product classification across various e-commerce and retail platforms.

[0002] In the rapidly evolving landscape of e-commerce and digital retail, accurate product categorization and matching is critical for improving search relevance, personalized recommendations, inventory management, and the customer experience. Traditional methods of manually labeling and classifying products are time-consuming, error-prone, and unscalable, especially when dealing with millions of SKUs and constantly changing product catalogs. This often results in inconsistent product data, poor search accuracy, and missed sales opportunities.

[0003] Existing automated systems rely heavily on static, rule-based engines or keyword matching that can't adapt to new product trends, ambiguous listings, or vendor-specific terminology. These systems are unable to handle unstructured or multilingual product data and struggle with contextual understanding, leading to miscategorization or missing attributes. Furthermore, most current solutions don't operate in real time, resulting in delayed updates and limited responsiveness to changing product information.

[0004] To address these challenges, the present invention presents an AI-based dynamic, real-time product matching and categorization system that leverages machine learning, deep learning, and natural language processing. This system is designed to intelligently learn from massive data sets, continuously adapt to evolving product information, and ensure highly accurate labeling and categorization. The invention enables scalable, automated, real-time product data enrichment, thus significantly improving operational efficiency, customer satisfaction, and the overall performance of digital commerce platforms.

[0005] An objective of the present disclosure is to enable automatic product categorization and attribute extraction in real time.

[0006] Another objective of the present disclosure is to reduce manual effort and errors in labeling product data.

[0007] Another goal of the present disclosure is to dynamically adapt to changing taxonomies and product trends.

[0008] Another objective of the present disclosure is to support multilingual and unstructured product data input.

[0009] Another objective of this disclosure is to improve data consistency and completeness through AI validation.

[0010] Another goal of the present disclosure is continuous improvement through feedback and adaptive learning.

[0011] Another goal of this disclosure is seamless integration with external platforms via real-time APIs.

[0012] Another objective of this disclosure is to provide transparent, verifiable KL decisions with explanatory tools.

[0013] Further objects and advantages of the present disclosure will become apparent from the following description, which is not intended to limit the scope of the present disclosure.

[0014] The present invention relates to an AI-based dynamic, real-time product matching and categorization system designed to automate the extraction, enrichment, and classification of product data across e-commerce and digital retail platforms. It addresses the inefficiencies and inaccuracies of manual tagging and static, rule-based systems.

[0015] Another embodiment of the present invention is that the system includes a product data ingestion and normalization module that collects product information from various formats and sources, cleanses and unifies the data, and prepares it for accurate AI-driven analysis. It processes structured and unstructured inputs in real time.

[0016] Another embodiment of the present invention is the Natural Language Understanding (NLU) and Feature Extraction module, which uses advanced NLP and transformer-based models to analyze product descriptions and identify relevant attributes such as brand, size, material, and usage context. It supports multilingual and domain-specific data.

[0017] Another embodiment of the present invention is the Dynamic Taxonomy Mapping and Category Prediction Module, which uses machine learning algorithms to automatically assign products to the most relevant categories. It adapts to evolving taxonomy structures and supports hierarchical classification schemes.

[0018] Another embodiment of the present invention is the real-time attribute enrichment and validation module, which enhances data by standardizing attribute values, populating missing fields, and checking attribute consistency using a knowledge graph. It flags inconsistencies and improves the overall reliability of the data.

[0019] Another embodiment of the present invention is the adaptive learning and feedback integration module, which continuously improves the system based on feedback from users, dealers, and system audits. Reinforcement learning ensures that the models evolve with new trends and product language patterns.

[0020] Another embodiment of the present invention is the real-time API and integration module, which allows the system to connect to external platforms and tools and supports both batch and streaming operations. It ensures low-latency data delivery, which is critical for fast-paced retail and product updates.

[0021] Another embodiment of the present invention is the monitoring, governance, and explanation module, which provides dashboards, audit trails, and interpretability of AI decisions using explanation methods. This ensures trust, compliance, and performance transparency throughout the system's lifecycle.

[0022] The invention relates to an AI-based dynamic real-time product attribution and categorization system (100) consisting of multiple intelligent and interconnected modules to ensure seamless, automated, and scalable product data classification. These modules work together to extract, understand, and assign relevant attributes and categories for products across various industries, particularly in dynamic retail and e-commerce environments. Module for recording and normalizing product data

[0023] This module is responsible for collecting structured and unstructured product data from various internal and external sources, including vendor feeds, product lists, manufacturer catalogs, and user-generated content. The data can be in various formats (CSV, JSON, XML, plain text) and languages. The module performs initial cleansing, deduplication, format unification, and normalization of the raw input data. It also detects anomalies such as missing fields, inconsistent units, or erroneous text and triggers predefined correction protocols or flags data for human review. This preprocessed, standardized data serves as input for downstream modules. Natural Language Understanding (NLU) and Feature Extraction Module

[0024] At the heart of the system is this module, which uses advanced natural language processing (NLP) models to analyze product titles, descriptions, specifications, and other text elements. Using Named Entity Recognition (NER), dependency parsing, and transformer-based language models (such as BERT or GPT), it extracts relevant attributes such as color, size, brand, material, gender, and usage. This module can handle multilingual input and recognizes domain-specific terminology, abbreviations, and compound attributes. It also uses contextual clues to disambiguate product terms (e.g., distinguishing "apple" as a fruit from "apple" as a brand). Dynamic taxonomy assignment and category prediction module

[0025] This module uses supervised and semi-supervised machine learning models to predict the most appropriate category for a product based on extracted features and learned classification patterns. Unlike static, rule-based systems, it adapts to evolving taxonomy structures and emerging product categories. The module supports hierarchical categorization and can dynamically reclassify products when the taxonomy is updated. Predictions are made in real time and with high accuracy, using algorithms such as decision trees, gradient boosting, or deep neural networks trained on historical data and feedback loops. Module for real-time attribute enrichment and validation

[0026] After extracting features and assigning categories, this module enriches product lists by adding missing attributes, standardizing values ​​(e.g., converting "bluish" to "blue"), and validating the consistency of data points. It compares the attributes against a knowledge graph or product ontology to ensure logical coherence (e.g., a jacket cannot have a shoe size). Furthermore, the system flags conflicting or low-confidence predictions and can automatically forward them to a human-involved interface or trigger self-learning mechanisms to improve future accuracy. Module for adaptive learning and feedback integration

[0027] This module incorporates feedback from user interactions (e.g., clicks, conversions, search refinements), merchant corrections, and quality assurance audits to continuously improve the accuracy and robustness of the AI ​​models. It uses reinforcement learning and continuous retraining pipelines to adapt to evolving trends, seasonal product variations, and language changes. The module ensures that the system becomes smarter and more context-aware over time, without the need for manual reprogramming or model resets. Real-time API and integration module

[0028] Designed for seamless deployment, this module provides RESTful APIs and webhooks that enable real-time interaction with enterprise systems, e-commerce platforms, and product management tools. It supports batch and streaming modes for product updates and offers configurable output formats. The module ensures low-latency responses for live product ingestion and updates, making it suitable for high-velocity retail environments where rapid time to market is essential. Module for monitoring, control and explainability

[0029] To ensure transparency, compliance, and trust, this module provides tools for monitoring system performance, data quality, and model decisions. It includes dashboards for tracking categorization accuracy, attribute coverage, and latency metrics. It also supports explainability frameworks (e.g., SHAP, LIME) to make model predictions interpretable, especially for audits or regulatory compliance. Alerts and logs are generated for unusual behavior, enabling proactive management and continuous optimization.

[0030] The invention is explained again below with reference to the figure. It shows: Fig. : an AI-based dynamic real-time system for product allocation and categorization (100).

[0031] The AI-powered dynamic real-time product mapping and categorization system (100) begins with the product data ingestion and normalization module, which collects and standardizes raw product data from various sources and prepares it for intelligent processing. The cleaned data then flows into the natural language understanding (NLU) and feature extraction module, where advanced NLP techniques extract key product attributes such as color, brand, size, and material from titles, descriptions, and specifications. These extracted features are passed to the Dynamic Taxonomy Mapping and Category Prediction Module, which leverages machine learning models to predict and assign the most appropriate product categories, dynamically adapting to changes in taxonomy structures.The enriched data is passed to the real-time attribute enrichment and validation module, which standardizes attribute values, fills in missing information, ensures logical consistency, and validates against a predefined knowledge base. Meanwhile, the adaptive learning and feedback integration module incorporates feedback from user interactions, merchant inputs, and system performance metrics to continuously refine and retrain the AI ​​models, ensuring the system evolves with changing product trends and language. The enriched and categorized product data is then made accessible via the real-time API and integration module, enabling seamless integration with e-commerce platforms, inventory systems, and product management tools.Throughout the entire process, the Monitoring, Governance, and Explanation module oversees the end-to-end workflow by tracking performance metrics, ensuring data quality, logging system decisions, and providing transparency and interpretability of AI predictions for traceability and compliance. Together, these modules form a coherent, intelligent system capable of delivering highly accurate, real-time product allocation and categorization at scale.

Claims

[1] An AI-based real-time product matching and categorization system (100) comprising: (a) a product data ingestion and normalisation module configured to collect and pre-process product data from multiple structured and unstructured sources; (b) a natural language understanding (NLU) and feature extraction module configured to extract product attributes using machine learning and natural language processing techniques; (c) a dynamic taxonomy assignment and category prediction module configured to assign products to appropriate categories using adaptive classification models; (d) a real-time attribute enrichment and validation module configured to complete, standardise and validate the extracted attributes; (e) an adaptive learning and feedback integration module configured to refine predictive models based on user, merchant and system feedback; f) a real-time API and integration module for connecting to external systems; g) and a monitoring, control and explanation module configured to track performance and provide interpretable decision insights. [2] The system (100) of claim 1, wherein the product data ingestion and normalization module is further configured to process multilingual, inconsistent, and incomplete data entries. [3] The system (100) of claim 1, wherein the NLU and feature extraction module uses transformer-based models for contextual understanding of product descriptions. [4] The system (100) of claim 1, wherein the taxonomy mapping module dynamically adapts to evolving product categories and supports hierarchical classification structures. [5] The system (100) of claim 1, wherein the attribute enrichment and validation module references a domain-specific knowledge graph to detect and correct logical inconsistencies in product data. [6] The system (100) of claim 1, wherein the adaptive learning module uses reinforcement learning to continuously improve categorization accuracy. [7] The system (100) of claim 1, wherein the real-time API module supports both batch and streaming modes for synchronous and asynchronous product updates. [8] The system (100) of claim 1, wherein the monitoring and control module includes explanation tools that generate interpretable visualizations and logs using SHAP or LIME methods.

Images