Fig. 11 Top seven association rules based on confidence for Swing obtained from threads containing how-to-do-it question.
In addition, we present, for the covered API elements, the distribution of the numbers of threads (containing how-to-do-it question) to which these were linked.
This may suggest that threads containing debug-corrective question frequently mention more API elements that are frequently mentioned together than threads containing how-to-do-it question.
To what extent does Stack Overflow cover API elements of Swing and Android APIs with threads containing how-to-do-it question?
First, we developed a classifier by using supervised machine learning algorithms to automatically classify Stack Overflow questions, in order to select threads containing how-to-do-it and debug-corrective question types.
In addition, we observed that, for each API, the coverage on threads containing debug-corrective questions is higher than the coverage on threads containing how-to-do-it questions over almost the whole period.
To select threads containing how-to-do-it and debug-corrective question types for coverage analysis, we performed the classification of the collected questions by using the classifiers developed in the "Automatic classification of questions" section.
In this paper, we extend the previous study, providing coverage analysis of the Swing and Android APIs on threads containing how-to-do-it and debug-corrective question types.
The distributions of the numbers of threads that were linked with API elements are long-tailed distributions for both APIs with significant outliers indicating that a small part of the APIs is responsible for the largest numbers of threads containing how-to-do-it question.
Figures 13 and 14 present charts, for Swing and Android, respectively, containing three series: the growth of API coverage on threads containing how-to-do-it (dotted line) and debug-corrective questions (dashed line), and the growth of the number of threads related to the API (solid line).
