Label/Entity Mapping Accuracy

Explanation

Label/Entity Mapping Accuracy checks whether the AI system correctly assigns each extracted value to the right field or entity, which matters in document-extraction because a value can be read perfectly yet mapped to the wrong label, for example placing a co-applicant's income in the primary applicant's field, producing a correct value in the wrong place and a flawed downstream decision. It is measured by the Mapping Correctness Score, computed as Correct_Mappings/Total_Mappings, the simple proportion of value-to-label assignments that are right. To implement it, maintain a labelled ground-truth set specifying the correct field or entity for each value, compare the model's mappings against it, count correct versus total mappings, and log the score along with examples of mis-mappings as evidence; this check runs after extraction in the pipeline, during pre-deployment validation and on sampled production output. The thresholds are a target of at least 0.90, with investigation triggered below 0.85 and remediation required below 0.80 (proposed): a score in the investigate band prompts a reviewer to examine recurring mapping errors, while a score in the remediate band requires corrective action such as refining the mapping logic or schema and routing affected documents to manual review. Its Immediate priority reflects that mis-mapped data corrupts records even when individual values are accurate.