Investigating the effectiveness of fasttext word representation technique for assignining bug reports

Software bugs are commonly encountered during the software life cycle. A bug can be generated mainly due to human errors, poor program design or early deployment without testing. The assignment of these bugs however can become a potential breach in the process of software development. It is because the process of assigning bug reports is generally manual, which makes the assignment process lengthy, difficult, and prone to errors. To decrease the time and cost consumed in the task of assigning bug reports many classification approaches have been proposed. However, the performance of these classification techniques decreases with due to complexity of the dataset and number of classes. Recently deep learning algorithms have demonstrated great efficiency in this matter. Deep learning algorithms show good results on complex and unstructured datasets. To understand the data and get better classification results conversion of text data to meaningful numerical values is important. Generally, the classical word representation methods show low performance in capturing the semantic and syntactic sense of text data. In our research we propose FastText embedding as word representation technique with Convolutional Neural Network (CNN). FastText captures important information of the text such as word analogies, semantic and syntactic etc. Our proposed model is tested on different dataset and yields Top-K developer accuracies. When compared to past research our model gives improved results.