Building a Developer’s Reputation Model for Programming Contest Platforms based on Source Code Analysis

Abstract: Last years developers became interested in competitive programming, producing the burst of online platforms which design competitions, where a set of programming problems should be solved. People realize this format is also useful for recruiting and training purposes. As developers participate in contests they gain points and are classified in rankings, according to the performance they exhibit in each competition. Today, these ranks are analogous to reputation systems since they indicate the programmers’ expertise. In this thesis dissertation, it is explored the relationship between this reputation and source code features, extracted via source code analysis, of the solutions sent by participants. The aim of this research is to find a set of features, classified into categories lexical, syntactic, quality and readability, which are shared by developers in the same level of knowledge. These features allow us to build a model to predict the programmers’ expertise, considering two classes to divide the developers, who can be assigned to beginners or advanced. In that sense, an experiment was designed to create a classifier model to accomplish that task. Overall, the results look promising, achieving an approximate accuracy of 80 %, and having the lexical features as the most relevant for the model.

• Autores: Garavito Mendez, Oscar Javier
• Fecha de publicación: 2019
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/76553
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/76553; http://bdigital.unal.edu.co/73057/
• Palabra clave: Reputation; Source Code Analysis; Classifier Models; Random Forest; Reputación; Análisis de Código Fuente; Modelos de Clasificación
• Rights: openAccess
• License: Atribución-NoComercial 4.0 Internacional