Sentiment analysis pipeline in NLP - Model Pipeline Trace

Model Pipeline - Sentiment analysis pipeline

This pipeline reads text reviews and learns to tell if the feeling is positive or negative. It cleans the text, turns words into numbers, trains a model, and then predicts feelings on new reviews.

Data Flow - 5 Stages

1Raw Text Input

1000 reviews x 1 column→Collect raw text reviews→1000 reviews x 1 column

"I love this product!"

↓

2Text Cleaning

1000 reviews x 1 column→Lowercase, remove punctuation and stopwords→1000 reviews x 1 column

"love product"

↓

3Tokenization & Vectorization

1000 reviews x 1 column→Convert words to sequences of numbers→1000 reviews x 10 tokens

[12, 45, 78, 0, 0, 0, 0, 0, 0, 0]

↓

4Train/Test Split

1000 reviews x 10 tokens→Split data into training (80%) and testing (20%) sets→Training: 800 reviews x 10 tokens, Testing: 200 reviews x 10 tokens

Training sample: [12, 45, 78, ...], Testing sample: [34, 56, 23, ...]

↓

5Model Training

Training: 800 reviews x 10 tokens→Train neural network to classify sentiment→Trained model

Model learns patterns to predict positive or negative

Training Trace - Epoch by Epoch

Loss
0.7 |****
0.6 |*** 
0.5 |**  
0.4 |*   
0.3 |*   
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.65	0.60	Model starts learning, accuracy is low
2	0.50	0.75	Loss decreases, accuracy improves
3	0.40	0.82	Model learns better features
4	0.35	0.86	Training converges, accuracy rises
5	0.30	0.89	Good performance, loss low

Prediction Trace - 5 Layers

Layer 1: Input Text

Layer 2: Tokenization

Layer 3: Embedding Layer

Layer 4: Neural Network Layers

Layer 5: Prediction

Model Quiz - 3 Questions

What happens to the text during the 'Text Cleaning' stage?

A The model is trained

B Punctuation and stopwords are removed

C Numbers are assigned to words

D Data is split into train and test sets

Key Insight

This visualization shows how raw text is cleaned and turned into numbers so a model can learn patterns. As training progresses, the model gets better at guessing feelings, shown by loss going down and accuracy going up. The final prediction step turns model output into a simple positive or negative label.