CleanCodeReview

To investigate the problem of classifying source code reviews, we have created a dataset suitable for evaluating and testing various methods for solving this problem. We combined four open datasets, and manually marked up 3200 code review comments. We have created our own classification based on the available datasets. The final dataset contains 10045 comments and is divided into 16 classes and 5 groups for hierarchical classification.

The dataset contains comment classes such as:

Style - readability, code layout, indentation issues, and other common programming conventions

Naming - uniform style and convenience of naming variables, methods, classes

Questioning - questions to the author of the code, requests clarification of the code or examples of use

Response - appointment of other reviewers, writes acknowledgements, agreements with others, additions to the developer's opinion

Convention - discussion of the software development process

Testing - requests tests to verify the functionality of the code

Design - architecture and code design, program structure control

Refactoring - logical structure, object creation, logical errors

Functionality - identification of code defects

Roadmap - further development of the program

Optimization - code optimization, parallelism, synchronization

Error - identifies problems with exception and error handling

Documentation - problems with documentation or comments in the source code

Support - compatibility with other systems, support systems

Input/Output - input/output in the graphical user interface, problems with pop-up windows.

Other - comments that do not carry a semantic load without context

Union of classes:

Code style (Style, Naming)

Discussion (Questioning, Response, Convention, Testing)

Development (Design, Refactoring, Functionality, Roadmap, Optimization, Error)

User (Documentation, Support, Input/Output)

Other (Other)