Improving Stack Overflow Tag Prediction Using Eye Tracking

I) Goals and Purpose Software developers use Stack Overflow to post questions and answers related to programming and computer science problems they need to solve. Questions such as seeking input on some efficient and time-saving methods of coding a particular program, getting help on solving various bottlenecks in coding are commonly seen. When users submit questions on Stack Overflow they need to submit at least one and up to five tags in addition to their question (see Figure 1). These tags attached to each question broadly identify the programming language talked about, the problem type in discussion and maybe some other fine grained categories the question belongs to. The tags associated with each question help with information retrieval or user queries. The goal of this project was to develop a tag prediction system utilizing eye tracking that will improve the accuracy of auto-generated Stack Overflow question tags. These Stack Overflow tags are important because they allow users to be able to further depict a problem within a program, or to precisely answer a programming question when users are on the Stack Overflow network. The main research question for our project is as follows: • To what degree do programmers focus on the keywords that tag extraction techniques generate? Based on the results of the previous question, another follow up study can be done to address the following two questions. In this project however, we only focused on the previous question above. • To what degree do the top n keywords from our approach and the standard approach match our Oracle generated keywords? • What are the best machine learning algorithms that can be successfully used to make such predictions?