Abstract

In recent years, crowd-based content in the form of screencast videos has gained in popularity among software engineers. Screencasts are viewed and created for different purposes, such as a learning aid, being part of a software project’s documentation, or as a general knowledge sharing resource. For organizations to remain competitive in attracting and retaining their workforce, they must adapt to these technological and social changes in software engineering practices.
In this thesis, we propose a novel methodology for mining and integrating crowd-based multi- media content in existing workflows to help provide software engineers of different levels of experience and roles access to a documentation they are familiar with or prefer. As a result, we first aim to gain insights on how a user’s background and the task to be performed influence the use of certain documentation media. We focus on tutorial screencasts to identify their important information sources and provide insights on their usage, advantages, and disadvantages from a practitioner’s perspective. To that end, we conduct a survey of software engineers. We discuss how software engineers benefit from screencasts as well as challenges they face in using screencasts as project documentation.
Our survey results revealed that screencasts and question and answers sites are among the most popular crowd-based information sources used by software engineers. Also, the level of experience and the role or reason for resorting to a documentation source affects the types of documentation used by software engineers. The results of our survey support our motivation in this thesis and show that for screencasts, high quality content and a narrator are very important components for users.
Unfortunately, the binary format of videos makes analyzing video content difficult. As a result, dissecting and filtering multimedia information based on its relevance to a given project is an inherently difficult task. Therefore, it is necessary to provide automated approaches for mining and linking this crowd-based multimedia documentation to their relevant software artifacts. In this thesis, we apply LDA-based (Latent Dirichlet Allocation) mining approaches that take as input a set of screencast artifacts, such as GUI (Graphical User Interface) text (labels) and spoken words, to perform information extraction and, therefore, increase the availability of both textual and multimedia documentation for various stakeholders of a software product. For example, this allows screencasts to be linked to other software artifacts such as source code to help software developers/maintainers have access to the implementation details of an application feature.
We also present applications of our proposed methodology that include: 1) an LDA-based mining approach that extracts use case scenarios in text format from screencasts, 2) an LDA-based approach that links screencasts to their relevant artifacts (e.g., source code), and 3) a Semantic Web-based approach to establish direct links between vulnerability exploitation screencasts and their relevant vulnerability descriptions in the National Vulnerability Database (NVD) and indirectly link screencasts to their relevant Maven dependencies. To evaluate the applicability of the proposed approach, we report on empirical case studies conducted on existing screencasts that describe different use case scenarios of the WordPress and Firefox open source applications or vulnerability exploitation scenarios.