Co-engagement Graph for Item Attribute Assignment

The patent introduces a co-engagement graph as an intermediary data structure between the raw item data and the attribute value determination process. This graph mediates the relationship by capturing co-engagement patterns between items, allowing the system to infer attribute values through graph-based propagation rather than complex machine learning models. The graph serves as a simplified intermediary that transforms the problem from direct classification to relational inference.

The system creates a simplified copy of the item relationship structure through the co-engagement graph, which replicates the essential co-engagement patterns without requiring the full complexity of the original data. This copied representation enables efficient attribute value propagation while avoiding the computational burden of processing the complete raw dataset through complex models.

The patent segments the attribute value determination process into two distinct phases: (1) constructing the co-engagement graph from item engagement data, and (2) propagating attribute values through the graph structure. This segmentation allows each phase to be optimized independently, with the graph construction capturing essential patterns once, and the propagation phase efficiently reusing this structure to determine attribute values for multiple items without reprocessing the entire dataset.

The system performs preliminary action by pre-computing the co-engagement graph and storing it for future use. This graph captures the co-engagement relationships between items in advance, allowing the system to quickly infer attribute values for new items by simply querying the pre-built graph structure, rather than re-running complex machine learning models each time an attribute value is needed.

The co-engagement graph serves as an intermediary that bridges the gap between unformatted free text item data and structured attribute value requirements. Instead of attempting to directly parse and extract attributes from free text, the system uses the graph to capture co-engagement patterns that implicitly encode attribute information, transforming the unstructured data into a form that enables efficient attribute inference.

The system enables self-service by allowing attribute values to be automatically inferred through graph-based propagation without requiring manual data formatting or complex text processing. The co-engagement graph structure itself provides the mechanism for automatic attribute value determination, with the system serving itself by leveraging the inherent relational patterns in the data rather than requiring external formatting interventions.