Abstract:
An increasing number of mobile app user reviews is a vital source on user concerns towards existing apps. These reviews
help to optimize and improve the apps. Despite the recent introduction of effective user review analysis methods, analyzing user reviews
still poses significant challenges for researchers. One of them is the overwhelming number of informative reviews make it difficulty
extract and prioritize user concerns. This research proposes a novel framework to prioritize user concerns in mobile app reviews utilizing
Natural Language Processing (NLP) techniques such as sentiment analysis, Latent Dirichlet Allocation (LDA), and word embedding.
This comprehensive framework extracts and ranks user concerns and opinions related to user experience (UX) using a weighted scoring
mechanism; multi-criteria prioritization formula. This formula includes four key metrics: Entropy score, Topic Prevalence score, Thumbsup
count, and Sentiment score for the major topics identified in the reviews. The proposed framework was evaluated using user reviews
from eight mobile apps across four popular categories: education, messaging, business, and shopping. A total of 869,731 reviews were
scraped from the Play Store for this evaluation. To validate the proposed framework, its prioritization results were compared with a
dataset prioritized by expert app developers. Spearman’s rank correlation was used to compare the prioritization trends and the average
correlation was 0.7569. Additionally, the Mean Absolute Error (MAE) was 0.1724. These results show that the proposed prioritization
framework aligns with the expert developers’ priorities with a marginal error. Furthermore, this framework is generalizable, as the
evaluation included apps from diverse categories. This makes the proposed framework an effective and efficient tool for decision-making
in patch, update or version releases in mobile apps, ensuring that critical user concerns are addressed promptly.
