Figure 1: The proposed methodology of Arabic text-based emotion detection and analysis with imbalanced class problem handling
Emotion |
Training |
Development |
Testing |
Anger |
899 |
215 |
609 |
Anticipation |
206 |
57 |
158 |
Disgust |
433 |
106 |
316 |
fear |
391 |
94 |
295 |
Happiness |
605 |
179 |
393 |
Love |
562 |
175 |
367 |
Optimism |
561 |
165 |
344 |
pessimism |
499 |
125 |
377 |
Sadness |
842 |
217 |
579 |
Supervise |
47 |
13 |
38 |
Trust |
120 |
36 |
77 |
Table 1: Description of SemEval-2018 Dataset
|
MCE without SMOTE |
MCE with SMOTE |
||||
Feature Size |
Precision |
Recall |
F-measure |
Precision |
Recall |
F-measure |
250 |
78.46 |
77.18 |
77.81 |
80.84 |
79.84 |
80.34 |
500 |
82.28 |
81.69 |
81.98 |
84.22 |
83.04 |
83.63 |
750 |
78.42 |
76.25 |
77.32 |
79.97 |
77.51 |
78.72 |
1000 |
82.19 |
83.22 |
82.7 |
83.58 |
84.7 |
84.14 |
Table 2: Performance of meta-classier ensemble (MCE) with and without SMOTE on SemEval-2018 datasets
Figure 1: The proposed methodology of Arabic text-based emotion detection and analysis with imbalanced class problem handling
Figure 2: The Pre-Processing Techniques
Figure 3: The General Architecture of the Proposed Ensemble
Figure 4: Performance of Baseline Classification Models with Proposed Enhanced Methods for Text-Based Arabic Emotion Analysis on SemEval-2018 Corpus
Figure 5: Performance of Meta-Classifier Ensemble (MCE) with and Without the SMOTE on SemEval-2018 Dataset
Tables at a glance
Figures at a glance