CodeReviewCommentsNER

This dataset contains code review comments from various sources (Github projects, Android Gerrit, Tizen Gerrit). It contains 3000 comments with 15420 manually-labeled named entities for the token classification task. The list of classes is following: Variable, Function, Class, Value, File_Name, File_Type, Keyword, Data_Type, Library_Package, Error_Name, HTML_XML_Tag, Operating_System, Programming_Language, External_Tool, Website. IOB2 format was used. The archive also contains data already divided into train and test. Our best model, based on CodeBERT, showed F1 score for tokens = 0.7347; Type F1 = 0.7703, Strict F1 = 0.7290, according to metrics from the nervaluate library.