Abstract: Techniques are described that provide users with performance summarizations over time. In some cases, an enterprise system receives a first performance evaluation for an employee at a first time, and receives a second performance evaluation for the employee at a second time. The enterprise system determines semantic meanings for text strings included in the first and second performance evaluations. The enterprise system inputs the semantic meanings into a machine-learned model trained to determine performance over time based at least in part on semantics of employee feedback. The enterprise system receives a performance score for the employee from the machine-learned model that reflects how the semantic meanings have changed over time. The enterprise system displays the performance score in a user interface, and may also display other performance scores for the employee, a rank of the employee based on the performance score, and/or other performance metrics.