Why machines need numerical text representation in NLP - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why can't machines understand raw text directly?

Machines process numbers, not words. Why is it necessary to convert text into numbers before feeding it to a machine learning model?

A Because machines only understand numerical data and cannot process raw text directly.

B Because numerical text representation removes all grammar and meaning from the text.

C Because converting text to numbers makes the text shorter and easier to read for humans.

D Because raw text contains too many spelling mistakes that machines cannot fix.

💡 Hint

Think about how computers store and process information internally.

Computers operate using numbers (binary digits). Text must be converted into numbers so machines can perform calculations and learn patterns.

❓ Predict Output

intermediate

Output of simple text to number mapping

What is the output of this Python code that converts words to their length?

NLP

words = ['cat', 'dog', 'elephant']
lengths = [len(word) for word in words]
print(lengths)

A [3, 3, 8]

B ['cat', 'dog', 'elephant']

C [5, 5, 5]

D [0, 0, 0]

💡 Hint

len(word) returns the number of characters in each word.

The code creates a list of lengths of each word: 'cat' and 'dog' have 3 letters, 'elephant' has 8.

❓ Model Choice

advanced

Choosing the right text representation for sentiment analysis

You want to build a model to detect positive or negative feelings in movie reviews. Which numerical text representation is best to capture word meanings and context?

A One-hot encoding of each word ignoring similarity

B Pretrained word embeddings like Word2Vec or GloVe capturing semantic meaning

C Bag-of-Words vector counting word frequencies without order

D Random numbers assigned to each word

💡 Hint

Consider which method captures word meanings and relationships best.

Pretrained embeddings represent words as vectors capturing semantic similarity, improving sentiment detection over simple counts or random numbers.

❓ Metrics

advanced

Evaluating text classification model performance

After converting text to numbers and training a classifier, which metric best tells you how well the model correctly identifies positive reviews?

A Recall - proportion of actual positive reviews correctly identified

B Accuracy - overall correct predictions divided by total predictions

C Precision - proportion of predicted positive reviews that are actually positive

D Loss - the error value during training

💡 Hint

Think about the metric that measures correctness of positive predictions.

Precision measures how many predicted positives are truly positive, important when false positives are costly.

🔧 Debug

expert

Debugging numerical text representation error

Given this code snippet converting text to numbers, what error will it raise?

NLP

text = 'hello world'
word_to_index = {'hello': 1, 'world': 2}
numbers = [word_to_index[word] for word in text.split() + ['!']]
print(numbers)

A No error, output will be [1, 2, 0]

B SyntaxError due to invalid list concatenation

C TypeError because split() returns a string not a list

D KeyError because '!' is not in word_to_index dictionary

💡 Hint

Check if all words exist in the dictionary before accessing.

The list includes '!' which is not a key in word_to_index, causing a KeyError when accessed.