Automatic Component Prediction for Issue Reports

Every day, there's a constant influx of software problems emerging during the testing and maintenance phases. With software becoming larger and more intricate, this issue count is on the rise, necessitating swift management. However, handling these issues manually proves challenging due to their complexity and sheer volume, often leading to inefficient and costly out-comes. Previous research endeavors have explored automating this triage process through machine learning and word-based language models, aiming to predict the component related to an issue. This component information is crucial for software engineers to pinpoint the problem's location. Yet, existing methods have fallen short of expectations due to their structural limitations and failure to grasp the context of words. To address this, we propose a novel approach leveraging pretrained language models, particularly fine-tuning BERT on a diverse dataset of issue reports. By doing so, we surpass the limitations of LSTM-based methods and enhance performance in predicting is-sue components