Sentiment Analysis with Deep Learning.
Second Session: All models and datasets hosted at: -Google Drive Link
Datasets: -sentiment140 Dataset Samples: 1.6 million polarity: 0 = negative. 2 = neutral. 4 = positive.
Epochs:10 Accuracy: 75% (Low Accuracy causes: Not enough data, Poor Deep learning model. [Recurrent Neural Nets/LSTM are better for Natural Language modeling.])
First Session:
Dataset: 10k lines of positive and negative samples of data. Number of Hidden Layers = 3
Number of Nodes in Hidden Layer 1 = 2000 Number of Nodes in Hidden Layer 2 = 2000 Number of Nodes in Hidden Layer 3 = 2000
Number of Epochs = 20 Batch Size = 100
Output Types = 2 Number of Lexicons = 423 Pickle Size = 139.9mb
O/P:
Epoch 1 completed out of 20 loss: 1985189.31934 Epoch 2 completed out of 20 loss: 723707.9021 Epoch 3 completed out of 20 loss: 647706.482666 Epoch 4 completed out of 20 loss: 443022.09021 Epoch 5 completed out of 20 loss: 307819.837387 Epoch 6 completed out of 20 loss: 161528.994003 Epoch 7 completed out of 20 loss: 186569.646027 Epoch 8 completed out of 20 loss: 352946.313339 Epoch 9 completed out of 20 loss: 429838.99939 Epoch 10 completed out of 20 loss: 52362.7623653 Epoch 11 completed out of 20 loss: 14503.3057673 Epoch 12 completed out of 20 loss: 12115.1989224 Epoch 13 completed out of 20 loss: 12971.1519582 Epoch 14 completed out of 20 loss: 11645.6373348 Epoch 15 completed out of 20 loss: 14256.5142536 Epoch 16 completed out of 20 loss: 13370.8477809 Epoch 17 completed out of 20 loss: 14245.3728524 Epoch 18 completed out of 20 loss: 14793.5029753 Epoch 19 completed out of 20 loss: 16494.0327399 Epoch 20 completed out of 20 loss: 19093.1549654 Accuracy: 0.634146 --- Training Time : 401.4542450904846 seconds ---