Product Abbreviation Taxonomy Identification Using LLM and RAG
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for identifying product taxonomy from product abbreviations require significant manual effort and time, as experts manually search and read product specifications to classify products, which is inefficient and labor-intensive.
Innovation Solution
A method and system utilizing a Large Language Model (LLM) to predict brand names from product abbreviations, generate expansions using an acronym expansion dictionary, apply web scraping and web search techniques, and employ Retrieval Augmented Generation (RAG) for taxonomy classification to automate the taxonomy identification process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual methods are used to identify product taxonomy from abbreviations, then accuracy can be maintained through expert knowledge, but time consumption and labor effort increase significantly
Solution Approach 1:
The patent introduces an intermediary system comprising acronym expansion dictionaries, brand name prediction models, and web search components that mediate between the product abbreviation input and taxonomy identification output. This intermediary automated system replaces direct manual expert analysis, significantly reducing time while maintaining accuracy through multiple verification steps including web scraping and candidate ranking.
2Measurement precision
If expert manual analysis is used to identify product taxonomy, then business understanding can be applied accurately, but labor intensity and operational complexity increase
Solution Approach 1:
The system enables self-service automation where the taxonomy identification process performs itself without requiring expert intervention. The automated pipeline includes acronym expansion, brand name prediction, web search, and candidate ranking that collectively replace manual expert operations, making the process operationally simple while maintaining classification accuracy through embedded business logic.
3Adaptability or versatility
If conventional web search methods are used without preprocessing, then all possible product pages can be found, but the search space becomes too large and time-consuming to process manually
Solution Approach 1:
The patent segments the product search process into distinct stages: acronym expansion to generate candidate names, brand name prediction to filter relevant brands, web search to find product pages, and candidate ranking to prioritize results. This segmentation divides the large search space into manageable segments processed sequentially, maintaining comprehensive product search coverage while dramatically improving identification efficiency through focused processing at each stage.
Data Source
AI summary
The present invention generally relates to the field of taxonomy identification. Identifying product taxonomy from a product abbreviation is currently performed manually and consumes lot of time and effort. Hence, embodiments of present disclosure provide an automated method for identification of product taxonomy from product abbreviation. First, a brand name of product abbreviation is predicted using a Large Language Model (LLM) and a brand list. Then, possible expansions of the abbreviation are generated based on the brand name using acronym expansion dictionary. Among the generated possible expansions, a relevant one is identified using the LLM. Later, web scraping and web search techniques are applied on the relevant expansion to obtain associated top k matches of a supergroup, a product group and module of the predicted relevant expansion. Finally, product taxonomy is predicted based on the top k matches using a LLM augmented taxonomy classification by Retrieval Augmented Generation.