3D Product Similarity Mapping for Assortment Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning techniques face challenges in accurately grouping similar products for assortment analysis, particularly for large retailers with numerous products, as they rely on rigid pre-existing classifications and fail to visualize complex similarities effectively, leading to difficulties in decision-making for assortment managers.
Innovation Solution
A machine learning algorithm trains a model to create a three-dimensional representation of products based on similarity, allowing for geometric calculations and visualizations that can be displayed in both two and three dimensions, enabling better clustering and recommendation services by identifying similar products through a similarity service.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pre-existing product classifications are used for grouping, then the classification process is simple and fast, but the grouping accuracy and ability to capture complex product similarities deteriorates
Solution Approach 1:
The patent transforms product classification from using fixed pre-existing categorical parameters to using dynamic similarity parameters derived from multiple product attributes. The system calculates similarity scores based on weighted combinations of attributes (price, category, brand, specifications) and uses these continuous similarity parameters to group products, thereby improving accuracy while maintaining computational efficiency through parameter optimization.
Solution Approach 2:
The patent creates a composite similarity metric by combining multiple product attributes (category, price, brand, specifications) into a unified similarity score. This composite approach allows the system to capture complex product relationships that single-dimensional classifications miss, achieving more accurate grouping by synthesizing information from multiple sources rather than relying on rigid pre-existing categories.
2Device complexity
If machine learning models use traditional output formats, then the model structure is simple, but the visualization capability and insight generation deteriorates
Solution Approach 1:
The patent enhances traditional machine learning output by adding a visualization dimension. Instead of only providing similarity scores, the system generates graphical representations including scatter plots that map products in 2D/3D space based on similarity, dendrograms showing hierarchical relationships, and heat maps displaying similarity matrices. This dimensional expansion allows assortment managers to visually explore product relationships and gain insights that numerical outputs alone cannot provide.
3Measurement precision
If detailed product attribute analysis is performed, then the assortment decisions are more accurate, but the computational time and resources increase
Solution Approach 1:
The patent implements pre-computation of product embeddings and similarity matrices. During an offline training phase, the system processes product attributes and computes embedding vectors that capture essential product characteristics. These pre-computed embeddings are stored and reused for rapid similarity calculations during assortment analysis, eliminating the need to reprocess raw attributes for each analysis task and significantly reducing computational time while maintaining accuracy.
Solution Approach 2:
The patent employs dynamic attribute weighting that adapts to different analysis contexts and user preferences. The system allows adjustment of weights for different product attributes (price, category, brand, specifications) based on specific assortment management needs. This dynamic approach enables the system to focus computational resources on the most relevant attributes for each analysis scenario, improving efficiency without sacrificing decision accuracy.
Data Source
AI summary
In an example embodiment, a machine learning algorithm is used to train a machine-learned model to create a three-dimensional representation of products, with each product mapped into a coordinate in the three-dimensional space. The model selects the coordinates based on the similarity of the product to other products. Coordinates that are closer geometrically in the three-dimensional space represent products that are similar to each other, whereas as the coordinates get further and further away, this implies that the products are less and less similar. This machine-learned model then not only allows for quick analysis of multiple products, as similarity between products or groups of products can be performed using geometric calculations (e.g., cosine distance), but also can then be tied into a three-dimensional representation that can be displayed either on a two-dimensional display or displayed on a three-dimensional display.