Hybrid AI Product Catalogue Mapping for Multilingual Attributes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for creating product catalogs struggle with accurately mapping product attributes across diverse data formats, languages, and modalities, leading to inefficiencies and errors due to reliance on manual rules, keyword matching, and limited scalability, especially in e-commerce environments.
Innovation Solution
A hybrid model integrating rule-based, Named Entity Recognition (NER), and Generative-AI models for extracting and mapping multilingual product attributes and values, utilizing Term Frequency-Inverse Document Frequency (TF-IDF) preprocessing and web crawling/scraping techniques to handle structured, unstructured, and semi-structured data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual rules or keyword matching methods are used for attribute-value mapping, then transparency and control are maintained, but labor intensity increases and scalability decreases
Solution Approach 1:
The patent combines multiple extraction methods (rule-based extraction, NER-based extraction, and Generative AI-based extraction) into a unified hybrid model. This merging allows the system to leverage the transparency of rule-based methods, the precision of NER, and the flexibility of Generative AI, thereby maintaining mapping accuracy while significantly improving processing efficiency and scalability.
Solution Approach 2:
The hybrid model is designed to handle diverse data formats (structured, unstructured, and semi-structured) and multiple languages through a single unified framework. This multi-functionality enables the system to maintain high mapping accuracy across different data types and languages while improving overall productivity through automated processing.
2Measurement precision
If supervised machine learning methods are used for attribute extraction, then accuracy can be improved with labeled data, but the requirement for large amounts of labeled data increases time and cost
Solution Approach 1:
The patent merges supervised NER methods with rule-based and Generative AI approaches. This combination allows the system to achieve high attribute extraction accuracy through NER while reducing dependency on large labeled datasets by leveraging rules and generative capabilities, thereby significantly reducing data preparation time and costs.
Solution Approach 2:
The extraction process is segmented into multiple independent modules (rule-based extraction, NER-based extraction, and Generative AI-based extraction), each handling specific aspects of attribute extraction. This segmentation allows the system to achieve high overall accuracy without requiring all components to rely heavily on labeled data, thus reducing time and resource requirements.
3Adaptability or versatility
If deep learning models are used for attribute mapping, then handling complex data patterns improves, but model complexity and difficulty in interpretation increase
Solution Approach 1:
The patent combines deep learning-based NER models with interpretable rule-based methods and Generative AI. This merging allows the system to handle complex data patterns effectively through the NER model while maintaining interpretability through rule-based approaches, thereby reducing overall model complexity and improving understandability.
Solution Approach 2:
The system segments the attribute extraction task into multiple specialized modules, each handling specific aspects. This segmentation reduces the complexity of individual models while maintaining high adaptability to complex data patterns through the collective capability of the hybrid system.
4Ease of manufacture
If rule-based approaches are used for attribute extraction, then domain-specific customization is improved, but flexibility and scalability decrease when handling diverse data formats
Solution Approach 1:
The patent merges rule-based extraction (providing domain-specific customization) with NER-based and Generative AI-based extraction (providing flexibility). This combination allows the system to maintain easy customization through rules while achieving high adaptability to diverse data formats through the complementary strengths of NER and Generative AI.
Data Source
AI summary
Disclosed is a method and system for creating a product catalogue using a hybrid model. A data reception module receives multimodal data related to a product from one or more data sources that may include structured, unstructured, and semi-structured data. An extraction module extracts text from the product data and is then preprocessed using a preprocessing module, which is further converted into numerical vectors. The hybrid model, an integration of a rule-based model, a Named Entity Recognition (NER) model, and a Generative-AI model, is contextually employed to extract multilingual attributes and values from the text. An attribute-value module generates one or more attribute-value pairs and maps them to the product in a structured format. Finally, a catalogue creation module creates a product catalogue using the one or more attribute-value pairs.