attribute_change_detection

This supplementary submission provides the dataset splits used in the experiments for the manuscript \u201cDetecting Attribute Changes in Entity Records Using Cross-Attention Mechanisms.\u201d The package includes three Parquet files corresponding to the train, validation, and test sets. Each example contains a paired representation of a current record and an earlier record, along with a binary label indicating whether an attribute change is present. The data is synthetically generated from a public multilingual product-metadata source to emulate realistic change and non-change scenarios, and the evaluation splits emphasize difficult partial-substitution cases to support stress testing.