Publications by authors named "Mustafa Mhamed"
Article Synopsis
- The paper introduces EMIR, the first dataset for Ethiopian music, featuring 600 recordings of various genres classified by expert judges into four Kiñits.
- The Ethio Kiñits Model (EKM) was developed to classify these clips, demonstrating that Mel-frequency Cepstral coefficients (MFCC) outperform other features for this task.
- In experiments, EKM achieved the highest accuracy (95.00%) compared to existing models, suggesting potential for further research in Ethiopian music classification.
Improving Arabic Sentiment Analysis Using CNN-Based Architectures and Text Preprocessing.
Comput Intell Neurosci
September 2021
Article Synopsis
- Sentiment analysis plays a crucial role in natural language applications, and this paper explores two models specifically for Arabic sentiment analysis using ASTD and ATDFS datasets in both 2-class and multiclass settings.
- The two proposed models, MC1 (a 2-layer CNN with global average pooling) and MC2 (a 2-layer CNN with max pooling followed by a BiGRU), achieved notable accuracy improvements over previous studies, reaching 73.17% on the challenging 4-class task and 90.06% on the simpler 2-class task.
- The research emphasizes the importance of effective Arabic preprocessing, incorporating innovative techniques, and finding that processing emoticons and using a custom stoplist can significantly enhance model performance,