This project comprises of a depression clasifier by finetuning BERT model. The model predicts two labels, where in 1 for Depressed text and 0 for Non Depressed text. This classification model can be extended for classifying chatbot converstion. Currently the model is trained using 20K sentences for both depression and non depression tweet data.
Epoch 1/5
286/286 [==============================] - ETA: 0s - loss: 0.0372 - accuracy: 0.9886
286/286 [==============================] - 526s 2s/step - loss: 0.0372 - accuracy: 0.9886 - val_loss: 0.0176 - val_accuracy: 0.9948
Epoch 2/5
286/286 [==============================] - ETA: 0s - loss: 0.0254 - accuracy: 0.9935
286/286 [==============================] - 526s 2s/step - loss: 0.0254 - accuracy: 0.9935 - val_loss: 0.0245 - val_accuracy: 0.9926
Epoch 3/5
286/286 [==============================] - ETA: 0s - loss: 0.0178 - accuracy: 0.9956
286/286 [==============================] - 526s 2s/step - loss: 0.0178 - accuracy: 0.9956 - val_loss: 0.0195 - val_accuracy: 0.9943
Epoch 4/5
286/286 [==============================] - ETA: 0s - loss: 0.0081 - accuracy: 0.9985
286/286 [==============================] - 536s 2s/step - loss: 0.0081 - accuracy: 0.9985 - val_loss: 0.0174 - val_accuracy: 0.9965
Epoch 5/5
286/286 [==============================] - ETA: 0s - loss: 0.0046 - accuracy: 0.9989
286/286 [==============================] - 525s 2s/step - loss: 0.0046 - accuracy: 0.9989 - val_loss: 0.0183 - val_accuracy: 0.9952