Demand Group Generation Using NLP Clustering and Decision Trees
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current demand modeling systems face computational challenges when dealing with large product inventories, often requiring simplifications that compromise accuracy, such as estimating or eliminating cross elasticities, or aggregating products into groups, which can lead to loss of granularity and increased costs due to the need for manual review and time-consuming processes.
Innovation Solution
A system and method for semi-automatically generating demand groups using product attributes, such as size, flavor, and brand, through natural language processing, clustering algorithms like k-means, and decision trees, to efficiently create demand groups that can be fed into pricing optimization systems, reducing manual effort and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If products are aggregated into groups for demand modeling, then computational requirements are reduced, but modeling accuracy and granularity are lost
Solution Approach 1:
The patent segments the product inventory into hierarchical levels: individual products, demand groups (clusters of substitutable products), and broader categories. This segmentation allows demand modeling to operate at the demand group level for computational efficiency while preserving the ability to drill down to individual product level for granular analysis and decision-making.
Solution Approach 2:
The patent introduces a new dimension of analysis by creating demand groups based on substitutability relationships rather than traditional product categorization. This transforms the modeling approach from product-centric to relationship-centric, enabling efficient computation across large inventories while maintaining accuracy through the substitutability metric.
2Reliability
If manual review and grouping of products is performed, then demand groups can be created with domain knowledge, but the process is time-consuming and costly
Solution Approach 1:
The system performs self-service by automatically generating demand groups using algorithms that analyze product attributes, sales data, and substitutability relationships. This eliminates the need for manual domain expert review while maintaining high accuracy, as the automated system continuously learns and refines groupings based on actual market behavior and product characteristics.
3Productivity
If cross elasticities are estimated or eliminated to reduce computations, then computational load decreases, but demand model accuracy deteriorates
Solution Approach 1:
The patent extracts and focuses computational resources on calculating cross elasticities only within demand groups where substitutability relationships exist. By eliminating the need to compute cross elasticities across all product pairs in the entire inventory, the system achieves significant computational savings while maintaining accuracy for the relevant product relationships.
Data Source
AI summary
The present invention relates to a system and method for generating demand groups. The system receives demand group modeling data including a product listing, point of sales data, available econometric data and product information. Attributes may then be assigned to the products based upon product identifiers, size, flavor, brand, and product descriptions utilizing natural language processing. The products may then be clustered according to the attributes and point of sales data. One or more decision trees may be generated for the product listings using the point of sales data. Demand rules may be received, which may be applied to the product clusters and the decision trees to generate demand groups. A confidence score may be generated for each product indicating how well that product fits within the demand group. These confidence scores may be compared against a threshold. Products with scores below the threshold may be flagged for user review.