Assessing Software Documents by Comprehension Effort

Recommender systems for software engineering have become increasingly popular in recent years. These systems combine several methodologies to provide suggestions that meet the developer’s needs. Recommender systems collect data from online resources, such as blogs, forums, Q&A websites, and suggest documents or pieces of code that are most likely helpful to the developers. However, these systems are not taking into consideration their comprehension effort, which may vary depending on the document familiarity and readability. In this thesis, we present our approach to calculating the comprehension effort, by creating a language model able to capture a document familiarity, that we combine with the document readability. Usually developers are more interested in documents which they are familiar with. By calculating the comprehension effort, a recommender system can complement the rank and suggest the most comprehensive and appropriate ones to the developer.