Table 4 Accuracy results SemEval-2016 dataset. Results of the external systems from the SemEval-2016: task a competition are compared with the results of the neural network models presented in the current body of work. While a total of 34 systems contributed to the competition, only four systems are illustrated in descending order for comparison [52]
Model | Layer/Filter Size | Accuracy (%) |
|---|---|---|
FFNN | [400, 400] [400, 400, 400] [775, 225, 75, 25] | 59.5 59.0 58.8 |
Single layer CNN | [1, 2] [1, 2, 3] [3, 4, 5] [1, 2, 3, 4, 5] | 61.7 61.1 60.2 60.6 |
Multi-layer CNN | [1, 2] | 62.2 |
GRU | [205] | 60.6 |
LSTM | [205] | 61.2 |
Bi-Dir LSTM | [205] | 62.2 |
TCN | [3] | 59.6 |
External System | Accuracy (%) | |
|---|---|---|
aueb.twitter.sentiment | 62.9 | |
sensei-lif | 61.7 | |
unimelb | 61.6 | |
senti-sys | 60.9 | |
Baseline | 34.2 | |