Fine-grained sentiment (5-class) in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Model Choice

intermediate

Choosing the best model for 5-class sentiment classification

You want to classify movie reviews into 5 sentiment classes: very negative, negative, neutral, positive, very positive. Which model is best suited for this task?

A A regression model predicting a continuous sentiment score from 1 to 5.

B A binary logistic regression model trained to predict positive vs negative sentiment.

C A clustering algorithm like K-means to group reviews into 5 clusters.

D A multi-class logistic regression model trained to predict one of the 5 sentiment classes.

💡 Hint

Think about the type of output needed: discrete classes vs continuous values.

Since the task is to classify into 5 distinct sentiment categories, a multi-class classification model like logistic regression with 5 output classes is appropriate. Binary classification only predicts two classes, clustering is unsupervised and may not align with sentiment, and regression predicts continuous values, not discrete classes.

❓ Metrics

intermediate

Evaluating a 5-class sentiment classifier

You trained a 5-class sentiment classifier. Which metric best measures overall accuracy across all classes?

A Accuracy (percentage of correct predictions)

B Precision for the positive class only

C Recall for the negative class only

D Mean Squared Error (MSE)

💡 Hint

Consider a metric that accounts for all classes equally.

Accuracy measures the proportion of correct predictions across all classes, making it suitable for multi-class classification. MSE is for regression, and precision or recall for a single class ignore other classes.

❓ Predict Output

advanced

Output of softmax layer for 5-class sentiment

Given the logits output from a neural network for 5 sentiment classes as [2.0, 1.0, 0.1, 0.5, 3.0], what is the predicted class index after applying softmax and choosing the highest probability?

NLP

import numpy as np
logits = np.array([2.0, 1.0, 0.1, 0.5, 3.0])
exp_logits = np.exp(logits - np.max(logits))
probs = exp_logits / exp_logits.sum()
predicted_class = np.argmax(probs)
print(predicted_class)

A 0

B 4

C 1

D 2

💡 Hint

Softmax picks the class with the highest logit after normalization.

The highest logit is 3.0 at index 4, so softmax probability is highest for class 4, making it the predicted class.

❓ Hyperparameter

advanced

Choosing batch size for training a 5-class sentiment model

You are training a neural network for 5-class sentiment classification on a dataset of 10,000 reviews. Which batch size is likely to give a good balance of training speed and stable gradient updates?

A Batch size of 64

B Batch size of 1 (stochastic gradient descent)

C Batch size of 10,000 (full batch gradient descent)

D Batch size of 1,000,000

💡 Hint

Consider typical batch sizes used in practice for stability and speed.

Batch size 64 is a common choice balancing speed and stable gradients. Batch size 1 is slow and noisy, full batch (10,000) is slow and memory heavy, and 1,000,000 exceeds dataset size and is invalid.

🔧 Debug

expert

Debugging incorrect predictions in 5-class sentiment model

Your 5-class sentiment model always predicts the 'neutral' class regardless of input. Which is the most likely cause?

A The optimizer learning rate is set to a moderate value like 0.001.

B The model uses ReLU activation in hidden layers.

C The model's output layer uses softmax activation but the loss function is mean squared error.

D The training data is perfectly balanced across all 5 classes.

💡 Hint

Think about compatibility between output activation and loss function.

Using softmax output with mean squared error loss can cause poor gradient signals, leading the model to predict the most frequent or default class. Balanced data, ReLU activations, and moderate learning rate are not likely causes of this issue.