An NLP Approach to Estimating Effort in a Work Environment
Iftinca Dan, Rusu Catalin, Oswald Oliver · 2020
Effort estimation in Software development is becoming increasingly hard to do correctly, this is in part due to the growing complexity of software projects, but also due to the higher amount of projects in total. Specialists in multiple fields are required to work together to obtain a realistic estimate, but even then, there is a good amount of risk involved when planning based on the estimate. This leads to cost increases both for the actual estimation process and the losses taken due to wrong estimations. There are already a few project estimation systems out there which take into account presumed system size, development cycles (design, develop, test) or other project related variables. We want to introduce a more granular estimation system, which uses the text descriptions of various tasks but also takes into account available metadata like seniority of the employee which will be working on said tasks, client and project/project type when estimating. The system is built on a deep neural network. The results we are getting are promising so far and we are working on establishing the human baseline accuracy while the tool is available for company employees to use.