Raw Social Media Comments

Input Layer GloVe Embeddings (300D)

Bidirectional LSTM 2 Layers, 128 Units

Dense Layer 64 Units - Dropout (0.5)

Output Layer 3 Units, Softmax

Tokenizer WordPiece

Transformer Encoder 6 Layers, 768 Hidden Dim

Tokenizer Byte-Pair Encoding

Transformer Encoder 12 Layers, 768 Hidden Dim

Pooling [CLS] Token

Classifier 3 Units, Softmax

Prediction Output

Positive / Negative / Neutral

Accuracy

Precision

Recall

F1-Score

Confusion Matrix

Statistical Analysis

LSTM

DistilBERT

Pooling [CLS] Token

Classifier 3 Units, Softmax

RoBERTa

Comprehensive Analysis