Sentiment_Analysis

Implement a Deep Neural Network based on Bidirectional LSTM Structure for Social media Comments Sentiment Classification

The problem

Social media comments sentiment analysis helps you discover hidden gems of business intelligence. It helps you set competitor benchmarks, track your performance based on customer satisfaction, get product insights, and manage your brand reputation. This project solves this problem by implementing a deep neural network model to classify Social media comments sentiments.

Data

The dataset used is the set of reviews on "Foody.vn" containing 27,000 samples labeled. It can be downloaded at:

https://streetcodevn.com/blog/dataset

Pre-Processing Data

• Firstly, we do Word Embedding, using Pretrained-Embedding Set:

  https://github.com/sonvx/word2vecVN

• Randomly split Dataset into Data train, validation and test with the ratio 8:1:1

Model

model = Sequential()

model.add(Bidirectional(LSTM(128, return_sequences=True),
                        input_shape=(200, 400)))
model.add(Dropout(0.3))
model.add(Bidirectional(LSTM(128, return_sequences=False)))                     
model.add(Dropout(0.3))

model.add(Dense(units=128, activation='relu'))
model.add(Dense(units=1, activation='sigmoid'))

model.compile(loss='binary_crossentropy',optimizer='adam',metrics=['accuracy'])
model.summary()

• Bidirectional LSTM

Result

Epoch1/10: 21600/21600 [==============================] - 255s 12ms/step - loss: 0.5280 - acc: 0.7462 - val_loss: 0.4350 - val_acc: 0.8259 
Epoch 2/10 21600/21600 [==============================] - 250s 12ms/step - loss: 0.3597 - acc: 0.8501 - val_loss: 0.3235 - val_acc: 0.8596 
Epoch 3/10 21600/21600 [==============================] - 249s 12ms/step - loss: 0.3000 - acc: 0.8771 - val_loss: 0.3156 - val_acc: 0.8652 
Epoch 4/10 21600/21600 [==============================] - 248s 12ms/step - loss: 0.2627 - acc: 0.8945 - val_loss: 0.2718 - val_acc: 0.8859 
Epoch 5/10 21600/21600 [==============================] - 243s 11ms/step - loss: 0.2340 - acc: 0.9069 - val_loss: 0.2769 - val_acc: 0.8811 
Epoch 6/10 21600/21600 [==============================] - 246s 11ms/step - loss: 0.2122 - acc: 0.9175 - val_loss: 0.2846 - val_acc: 0.8896 
Epoch 7/10 21600/21600 [==============================] - 247s 11ms/step - loss: 0.1802 - acc: 0.9313 - val_loss: 0.3045 - val_acc: 0.8852 
Epoch 8/10 21600/21600 [==============================] - 247s 11ms/step - loss: 0.1476 - acc: 0.9455 - val_loss: 0.3312 - val_acc: 0.8793 
Epoch 9/10 21600/21600 [==============================] - 247s 11ms/step - loss: 0.1293 - acc: 0.9532 - val_loss: 0.2995 - val_acc: 0.8922 
Epoch 10/10 21600/21600 [==============================] - 246s 11ms/step - loss: 0.1045 - acc: 0.9619 - val_loss: 0.3487 - val_acc: 0.8867

scores = model.evaluate(_test_x, test_y, verbose=1)
print('Test loss:', scores[0])
print('Test accuracy:', scores[1])
=>
2700/2700 [==============================] - 46s 17ms/step Test loss: 0.3845756830744169 Test accuracy: 0.8733333333333333