An information retrieval strategy for large multimodal data collections involving source code and natural language

Source code repositories store data from software products. Among this data we can find the evolution of the source code, requirements, bugs and communication between developers. Source code repositories have been growing rapidly in the recent years andwith them the need of extracting information from them. An interesting source code repository that is growing both in usage and information is Stack Overflow (SO), this web site provides one of the biggest Question Answering places used by thousands of developers everyday. In SO the developers can ask any question related to a programming issue and it will be answered by other users. We can find a source code repository with both source code and natural language with thousands of samples and the possibility of combining both sources of information to extract useful and not eye-noticeable information from it. In this thesis, we explore how to represent source code and natural language and how to combine these representations. We try to solve the task of understanding how users in SO talk about the programming language, how similar these programming languages are among them based on how users talk about them, and finally, we provide tools on the building of an information retrieval strategy by identifying duplicated post.

• Autores: Baquero Vargas, Juan Felipe
• Fecha de publicación: 2019
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/76556
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/76556; http://bdigital.unal.edu.co/73062/
• Palabra clave: Stack Overflow; source code analysis; Duplication detection; Predicting programming language; Análisis de código fuente; Detección de duplicados; Predecir el lenguaje de programación
• Rights: openAccess
• License: Atribución-NoComercial 4.0 Internacional