Co-clustering Prediction Model Generation for Customer-Merchandise Accuracy

The patent segments the customer-merchandise relationship space by performing co-clustering on customer IDs and merchandise IDs, dividing them into multiple customer clusters and merchandise clusters respectively. This segmentation allows the system to capture diverse characteristics of different customer groups and merchandise categories, thereby improving prediction accuracy without requiring a single overly complex model.

The patent generates separate prediction models for each combination of customer cluster and merchandise cluster, allowing each local model to be optimized for specific customer-merchandise relationships. This local quality approach enables each model to capture nuanced characteristics of its target segment, improving overall prediction accuracy while keeping individual model complexity manageable.

Instead of creating separate models for each individual customer-merchandise pair, the patent segments customers into clusters and merchandise into clusters, then creates models for each cluster combination. This reduces the number of models from potentially millions of individual pairs to a manageable number of cluster combinations, balancing accuracy with computational feasibility.

The patent creates prediction models that work across multiple customer-merchandise relationships within each cluster combination, making each model universal for its segment. This multi-functionality approach allows a single model to serve multiple prediction purposes within its cluster, reducing the total number of models needed while maintaining high accuracy for each segment.