view full project website TopicXP
A visualization for mapping concepts to codeTopicXP was a project undertaken while I was a part of the SEMERU group at the College of William & Mary. Building on past work using information retrieval algorithms to visualize source code, TopicXP used LDA to extract topics (which can be simply described as collections of words that frequently appear together) from the natural language found in comments and identifiers in source code, and then presented this information to the user in a navigable visualization.
With the goal of giving new developers an overview of the concerns present in a system and how they relate to one another, TopicXP starts by showing a visualization of the different topics it extracted and arrows representing how many method calls are made between classes which are assigned to each topic. Users can then navigate into these topics to see an overview the packages and classes which implement the concepts represented by these topics, and finally can navigate deeper again to view the actual source code.
Using this workflow, developers can gain a high-level appreciation of the concerns within a piece of software and then dive deeper to learn more about the specific concerns that they are interested in, leveraging LDA's topics to guide their exploration of the system.
