This invention discloses an indicator attribution
analysis method based on expression tree
decomposition and adaptive
drill-down. The method includes:
parsing the calculation expression of the target indicator to generate an operator tree; obtaining base period and
current period dimension-grouped indicator data, and calculating the total change of the target indicator at the root node of the operator tree; calling the corresponding
decomposition rules according to the operator type of each internal node, completing the operator
decomposition layer by layer from the root node, satisfying the change conservation constraint throughout the process, and obtaining the attribution contribution of each dimension value in the current layer; calculating the attribution
gain of a single dimension path, and generating
drill-down or stop decisions for each path after independently comparing it with a preset threshold; continuing to
drill down and repeating the calculation process for paths that meet the
gain threshold, and stopping drill-down for paths that do not meet the threshold, finally generating an asymmetric attribution tree with differentiated
branch depths, and outputting a structured attribution report. This invention can achieve accurate and verifiable indicator attribution, and balances analytical accuracy and computational efficiency through an adaptive drill-down mechanism.