<![CDATA[Analyzing mobile application user reviews plays a crucial role in software evolution. However, manual processes are often constrained by the large volume of data and language ambiguity. This research develops an automated classification model to categorize reviews into bug reports, feature requests, and non-informative using a Bi-Long Short-Term Memory (LSTM) architecture reinforced with an Attention Mechanism. Experimental results show that the model achieves 94.33% training accuracy and 71.95% testing accuracy, outperforming the standard Bi-LSTM which only reached 87.58% in training accuracy and 70.85% in testing. In terms of efficiency, this model converges faster, reaching peak performance in only 100 epochs, compared to 500 epochs for the standard Bi-LSTM. Furthermore, experiments show that increasing architectural complexity, such as combining Bi-LSTM with Attention, triggers overfitting and weight fluctuations. Thus, the integration of the Attention Mechanism in BiLSTM is proven to provide an optimal balance between computational efficiency and prediction accuracy, effectively supporting decision-making systems for developers.]]>
Copyrights © 2026
