Identifying product groups in ecommerce

Systems and methods are disclosed herein for supplementing product records with product groups that are relevant to the product records. Queries form users may be analyzed to extract keywords. Search results for keywords are evaluated to determine category consistency among product records, including such values as entropy and taxonomy depth. Those keywords with search results having adequate category consistency are selected as product groups and the search results associated with the product groups. Product groups are associated with product records according to a random walk of a graph having as nodes products and product groups and links representing belonging of a product to a product group. Product groups may be selected based on a transition probability based on a random walk and a quality score based on usage of a product group page for the product group.