A BERT and Topic Model Based Approach to reviews Requirements Analysis

With the rise of mobile applications, user reviews are an important avenue of user feedback in which users may mention different issues in using the software. For example, unresponsiveness, low-level privacy, etc. In order to extract effective requirement information and problematic feedback from these huge user reviews, this paper proposes a review non-functional requirement analysis method based on BERT model and topic model (NRABL). Firstly, we use BERT model to classify the reviews with multi-labels classification and then use LDA (latent Dirichlet allocation) to extract topic and review analysis, this method can help developers to quickly understand the user's requirements and the specific usage problems.