Improving the Classification of Q&A Content for Android Fragmentation Using Named Entity Recognition

Abstract

Despite the huge amount of high quality information available in socio-technical sites, it is still challenging to filter out relevant piece of information to a specific task in hand. Textual content classification has been used to retrieve only relevant information to solve specific problems. However, those classifiers tend to present poor performance when the target classes have similar content. We aim at developing a Named Entity Recognizer (NER) model to recognize entities related to technical elements, and to improve textual classifiers for Android fragmentation posts from Stack Overflow using the obtained NER model. The proposed NER model was trained for the entities API version, device, hardware, API element, technology and feature. The proposed classifiers were trained using the recognized entities as attributes. To evaluate the performances of these classifiers, we compared them with other three textual classifiers. The obtained results show that the constructed NER model can recognize entities efficiently, as well as discover new entities that were not present in the training data. The classifiers constructed using the NER model produced better results than the other baseline classifiers. We suggest that NER-based classifiers should be considered as a better alternative to classify technical textual context compared to generic textual classifiers.