Purchase Embeddings for Accurate Regional Shopping Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems provide minimally useful purchase recommendations based on simple word matching, failing to consider individual and regional shopping habits, vendor preferences, and item characteristics beyond basic descriptions, leading to irrelevant suggestions.
Innovation Solution
Utilizing purchase embeddings, which are multi-dimensional data records representing items and purchases, to calculate correlations and similarities based on past buying histories and regional shopping patterns, enabling targeted recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If simple word matching is used for purchase recommendations, then the system is easy to implement and fast, but the recommendation accuracy and relevance deteriorate
Solution Approach 1:
The patent transforms item descriptions into numerical vector representations (embeddings) that capture semantic meaning. This parameter transformation allows the system to move from simple string matching to sophisticated similarity calculations based on multiple dimensions including item characteristics, purchase patterns, and contextual information, thereby improving recommendation accuracy while maintaining computational efficiency
Solution Approach 2:
The patent introduces multi-dimensional embedding spaces where items are represented not by single keywords but by vectors containing multiple attributes such as item features, purchase history patterns, regional preferences, and temporal information. This dimensional expansion enables the system to capture complex relationships between items that simple word matching cannot detect
2Measurement precision
If multiple item characteristics and regional preferences are considered, then recommendation relevance improves, but system complexity increases
Solution Approach 1:
The patent combines multiple data sources and item characteristics into unified item embeddings and user preference embeddings. By merging item features, purchase history, regional preferences, and temporal patterns into single vector representations, the system manages complexity through consolidation while still capturing nuanced relationships for highly relevant recommendations
Solution Approach 2:
The patent introduces embedding vectors as intermediary representations between raw data and recommendation logic. These embeddings serve as mediators that transform complex multi-dimensional information into compact numerical forms that can be efficiently processed by similarity calculations, bridging the gap between detailed input data and simple recommendation output
3Measurement precision
If purchase embeddings are calculated based on comprehensive purchase history analysis, then correlation accuracy improves, but computational time increases
Solution Approach 1:
The patent pre-calculates item embeddings and stores them in databases before actual recommendation queries. By performing the computationally intensive embedding calculations in advance based on available purchase history and item characteristics, the system avoids repeating these calculations during real-time recommendations, thereby maintaining high correlation accuracy while reducing query response time
Solution Approach 2:
The patent creates and stores copy representations (embeddings) of items and user preferences that capture the essential characteristics without requiring access to the full original data during queries. These embedded copies enable fast similarity comparisons while preserving the nuanced relationships learned from comprehensive purchase history analysis
Data Source
AI summary
Aspects described herein may relate to methods, systems, and apparatuses that provide new capabilities for recommending purchases to a user based on the user's past purchasing history and the purchase history of others. A new descriptor referred to as “purchase embeddings” is disclosed, which are data records in a new multi-dimensional space for describing and tracking purchases of goods and services.