CppSATD: A Reusable Self-Admitted Technical Debt Dataset in C++

CppSATD is the first dedicated C++ SATD comments dataset comprising multi-class SATD comments along with their code context. It is the largest available multi-type SATD dataset with surrounding code context, comprising over 13k manually annotated SATD comments grouped under different types of SATD.

silver_standard_cpp.csv: The silver standard dataset comprises a total of 531,367 source code comments from all five repositories identified. We perform an automated annotation for SATD on the entire dataset using the established SATD text patterns from the PENTACET corpus. The automated annotation resulted in 18,973 SATD comments. The remaining 512,394 are treated as NON-SATD comments. This dataset does not involve any human verification but is solely based on the SATD patterns from the PENTACET corpus.

gold_standard_cpp.csv: We utilize the SATD comments from the silver standard data. We manually verified the 18,973 comments to validate if they are SATD. To ascertain that the 512,394 are NON-SATD comments at 99% confidence level and 1% margin of error, we take a random sample of 16,125 source code comments from 512,394 comments. Overall, the manual verification resulted in a total of 13,069 SATD comments and 22,029 NON-SATD comments. The 13,069 is further classified into one of the five types of SATD as identified in the seminar work that introduced the concept of SATD.

repos.zip: This includes the source code of 5 C++ repositories, from which we collected the data.