Item Matching Using Attribute Extraction and ML Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining similar items are computationally inefficient and require significant user interaction data, often leading to biased results, especially for new or low-volume items, and do not efficiently match items across large datasets.
Innovation Solution
A system that uses a processor to determine attributes of items based on description data, computes match scores using machine learning models, and provides a graphical user interface to represent matching items, optimizing the matching process by training models on user feedback and attribute weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If user interaction data (purchases, browsing, shopping cart contents) is used to determine related items, then complementary items can be identified, but the method is computationally inefficient and requires significant user interaction data that may not be available for new or low-volume items
Solution Approach 1:
The patent segments the item matching process into distinct components: extracting attributes from item descriptions, computing attribute values for paired attributes, determining attribute-type indicators, and calculating match scores. This segmentation allows each component to be optimized independently and enables parallel processing, improving computational efficiency while maintaining matching accuracy without requiring user interaction data
Solution Approach 2:
The patent replaces the mechanical system of collecting and processing user interaction data (clickstreams, purchases, browsing behavior) with an automated system that extracts and compares item attributes directly from item descriptions using natural language processing and machine learning models. This substitution eliminates the need for computationally intensive user data processing while providing reliable matching for all items including new and low-volume items
2Reliability
If user interaction data is required for item matching, then matching can be performed, but results are skewed toward incorrect or biased results and sufficient data is not available for new or low-volume items
Solution Approach 1:
The patent enables items to serve themselves by automatically extracting their own attributes from their descriptions and using those attributes for matching. The system processes item descriptions directly to identify and compare relevant attributes without requiring external user interaction data. This self-service approach ensures that all items, including new and low-volume items, have sufficient data available for accurate and unbiased matching based on their inherent characteristics
Solution Approach 2:
The patent extracts relevant attributes directly from item descriptions using natural language processing and machine learning techniques. By taking out and isolating key attributes from unstructured description text, the system creates a structured representation that can be compared across items without relying on user interaction data, ensuring unbiased and reliable matching results for all items regardless of their popularity or age
3Measurement precision
If manual line-by-line searching of databases is performed to evaluate item specifications, then similar items can be identified, but the process is time consuming and impractical for large sets of items
Solution Approach 1:
The patent replaces the manual mechanical process of line-by-line searching and comparing item specifications with an automated computer-implemented system. The system uses natural language processing to extract attributes from item descriptions, computes attribute values programmatically, and calculates match scores automatically. This substitution maintains detailed attribute comparison precision while reducing evaluation time from minutes or hours to seconds, making the process practical for large sets of items
Solution Approach 2:
The patent transforms unstructured item description text into structured attribute parameters that can be efficiently compared. By changing the state of the data from raw text to extracted attribute values with defined data types and formats, the system enables rapid automated comparison while maintaining the precision of detailed attribute evaluation. This parameter transformation allows the system to process large numbers of items quickly without sacrificing measurement precision
Data Source
AI summary
The technology includes an example method for determining matching items. In some implementations, the method may determine a first set and second set of attributes and attribute types associated with a first and second item, respectively, based on description data associated with the items. The method may determine an attribute-type indicator for a pairing of a first attribute of the first set with a second attribute of the second set based on attribute types of the attributes, compute an attribute value for the pairing based on a similarity between the first and second attribute, and compute a match score for a combination of the first and second items based on the attribute value and attribute-type indicator. The method may also, in response to receiving a data query identifying the first item, provide a graphical user interface including a graphical element representing the second item based on the match score.