Fuzzy Multi-intent Classifier For User Generated Software Documents

ABSTRACTUser generated software documents (e.g. app reviews, bug reports, developer emails, code comment, etc) often contain a sense of purpose, or intent. They would help the analyst in categorizing, in assigning suitable resources, or in further extracting important information (e.g. user opinions in reviews). However, recent studies were using mostly human effort to extract and encode linguistic patterns to detect such intents from very specific domains (i.e. app reviews and developer emails). In this work, we formalize the problem statement of multi-intent classification for user generated software documents, and propose a Fuzzy Multi-Intent Classifier that can train a classification model on annotated text examples to predict intents using fuzzy logic on matched intent patterns. Our classifier utilizes the novel concept of Lexeme Sequence Pattern which can approximately match to similar text without having a human to encode its Parse Tree. We evaluated our approach against 6 common classification algorithms and out-performed them.