Short Text Matching in Performance Management

In “Role” based Performance Appraisal process the evaluation of Individuals is done based on the meeting of target for “Goals” given to that individual in the specified time period. Standardization of goals with the help of a pre defined “template” is important for completeness and correctness of role definition and comparing two individuals. Since a goal is a short textual description of expected activity we pose this as a matching problem and explore two different approaches that use minimal human supervision. First approach is based on co-training framework which uses goal description and available additional information in the form of self comments. The second approach uses semantic similarity using weak supervision framework. We demonstrate superior performance of the two approaches as compared to multiple baselines. First approach gives better recall while second approach scores better in precision. Based on the objective of the end application any of the approaches can be used.