Hybrid approaches in NLP - Practice Problems & Coding Challenges

Challenge - 5 Problems

🧠 Conceptual

intermediate

Understanding Hybrid Models in NLP

Which of the following best describes a hybrid approach in Natural Language Processing (NLP)?

A Combining rule-based methods with machine learning models to improve text understanding.

B Using only deep learning models without any handcrafted rules.

C Applying unsupervised learning exclusively for text classification.

D Relying solely on dictionary lookups for language translation.

💡 Hint

Think about mixing different techniques to get better results.

Hybrid approaches combine rule-based and machine learning methods to leverage the strengths of both, improving NLP tasks.

❓ Model Choice

intermediate

Choosing Models for a Hybrid Sentiment Analysis System

You want to build a sentiment analysis system that uses both a lexicon-based method and a machine learning classifier. Which combination below fits a hybrid approach?

A Use only a pre-trained transformer model without any lexicon.

B Use a sentiment dictionary to score words and a logistic regression model trained on labeled reviews.

C Apply k-means clustering on unlabeled text data.

D Use a rule-based system that assigns sentiment based on fixed patterns only.

💡 Hint

Look for a mix of dictionary and machine learning.

Combining a lexicon-based scoring with a machine learning classifier is a classic hybrid approach for sentiment analysis.

❓ Predict Output

advanced

Output of Hybrid Text Classification Pipeline

What is the output of the following Python code that combines TF-IDF features with a rule-based keyword count for classification?

NLP

from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
import numpy as np

texts = ["I love sunny days", "I hate rain", "Sunny weather is great", "Rainy days are gloomy"]
labels = [1, 0, 1, 0]

# Rule-based feature: count of positive words
positive_words = {'love', 'sunny', 'great'}
rule_features = np.array([[sum(word in positive_words for word in text.lower().split())] for text in texts])

# TF-IDF features
vectorizer = TfidfVectorizer()
tfidf_features = vectorizer.fit_transform(texts).toarray()

# Combine features
X = np.hstack((tfidf_features, rule_features))

model = LogisticRegression().fit(X, labels)
predictions = model.predict(X)
print(predictions.tolist())

A [0, 0, 0, 0]

B [0, 1, 0, 1]

C [1, 0, 1, 0]

D [1, 1, 1, 1]

💡 Hint

Check how the rule-based feature and TF-IDF features help the logistic regression model.

The model learns from both TF-IDF and the positive word count, correctly predicting the original labels.

❓ Hyperparameter

advanced

Tuning Hybrid Model Parameters

In a hybrid NLP model combining a rule-based sentiment score and a neural network, which hyperparameter adjustment is most likely to improve the balance between the two components?

A Increase the number of epochs for training the neural network only.

B Use a smaller batch size without changing the model architecture.

C Remove the rule-based component to simplify the model.

D Adjust the weight given to the rule-based score in the final prediction layer.

💡 Hint

Think about how to control the influence of each part in the combined output.

Adjusting the weight of the rule-based score helps balance its contribution with the neural network's output.

🔧 Debug

expert

Debugging a Hybrid NLP Pipeline Error

Consider this hybrid NLP pipeline code snippet that combines a rule-based feature with a machine learning model. It raises a ValueError: shapes (4,5) and (4,1) not aligned. What is the cause?

NLP

import numpy as np
from sklearn.linear_model import LogisticRegression

texts = ["happy day", "sad night", "joyful morning", "gloomy evening"]
labels = [1, 0, 1, 0]

# Rule-based feature: count of positive words
positive_words = {'happy', 'joyful'}
rule_features = np.array([[sum(word in positive_words for word in text.split())] for text in texts])

# Dummy TF-IDF features with wrong shape
tfidf_features = np.random.rand(4, 5)

# Incorrect feature combination
X = np.dot(tfidf_features, rule_features)

model = LogisticRegression().fit(X, labels)

A Using np.dot to combine features with incompatible shapes causes the error.

B The rule-based feature calculation is incorrect and returns empty arrays.

C LogisticRegression cannot be trained on combined features.

D Labels array length does not match feature rows.

💡 Hint

Check how features are combined and their shapes.

np.dot requires matching inner dimensions; here (4,5) and (4,1) do not align, causing the ValueError.