Why similarity measures find related text in NLP - Why Metrics Matter

Metrics & Evaluation - Why similarity measures find related text

Which metric matters and WHY

When we want to find related text, we measure how close or similar two pieces of text are. The key metrics are Cosine Similarity and Jaccard Similarity. Cosine similarity measures the angle between two text vectors, showing how similar their meaning is regardless of length. Jaccard similarity compares shared words or features. These metrics help us find texts that talk about the same ideas or topics.

Confusion matrix or equivalent visualization

Related Text Pairs (Positive) vs Not Related (Negative):

               Predicted Related   Predicted Not Related
Actual Related       TP = 80             FN = 20
Actual Not Related   FP = 15             TN = 85

Total samples = 200

From this:
Precision = TP / (TP + FP) = 80 / (80 + 15) = 0.842
Recall = TP / (TP + FN) = 80 / (80 + 20) = 0.8
F1 Score = 2 * (0.842 * 0.8) / (0.842 + 0.8) ≈ 0.82

This shows how well similarity measures help find truly related text pairs.

Precision vs Recall tradeoff with examples

If we want to find related text, sometimes we want to be very sure the pairs we find are truly related (high precision). For example, in a legal document search, wrong matches waste time.

Other times, we want to find as many related texts as possible (high recall). For example, in research, missing related papers is bad.

Improving precision may lower recall and vice versa. Choosing the right balance depends on the task.

What good vs bad metric values look like

Good: Precision and recall both above 0.8 means most found pairs are truly related and most related pairs are found.

Bad: Precision below 0.5 means many unrelated pairs are marked related. Recall below 0.5 means many related pairs are missed.

For similarity measures, a good threshold to decide relatedness is key to get good precision and recall.

Common pitfalls in metrics

• Accuracy paradox: If most text pairs are unrelated, a model that always says "not related" can have high accuracy but is useless.
• Data leakage: Using the same text in training and testing can inflate similarity scores.
• Overfitting: Tuning similarity thresholds too closely on one dataset may not work on new texts.

Self-check question

Your similarity model finds related text pairs with 98% accuracy but only 12% recall. Is it good for finding related texts? Why or why not?

Answer: No, because it misses most related pairs (low recall). It finds very few related texts even if it is usually correct when it does. For related text search, missing many related pairs is a big problem.

Key Result

Cosine and Jaccard similarity metrics help find related text by measuring closeness; balancing precision and recall is key for good results.

Practice

1. Why do similarity measures help find related text in NLP?

easy

A. Because they compare numeric representations of texts to find closeness

B. Because they translate text into images for comparison

C. Because they count the number of words in each text

D. Because they randomly select texts to compare

Solution

1. Step 1: Understand text representation in NLP

   Texts are converted into numbers (vectors) so computers can compare them easily.

2. Step 2: Role of similarity measures

   Similarity measures calculate how close these numeric vectors are, showing relatedness.

3. Final Answer:

   Because they compare numeric representations of texts to find closeness -> Option A

4. Quick Check:

   Similarity = Numeric comparison [OK]

Hint: Similarity means comparing numbers, not words directly [OK]

Common Mistakes:

• Thinking similarity compares raw words directly
• Confusing similarity with random selection
• Believing similarity translates text into images

2. Which of the following is the correct way to calculate cosine similarity between two vectors A and B in Python?

easy

A. cos_sim = np.linalg.norm(A - B)

B. cos_sim = np.sum(A + B)

C. cos_sim = np.dot(A, B) / (np.linalg.norm(A) * np.linalg.norm(B))

D. cos_sim = np.dot(A, B) * (np.linalg.norm(A) + np.linalg.norm(B))

Solution

1. Step 1: Recall cosine similarity formula

   Cosine similarity = dot product of vectors divided by product of their lengths.

2. Step 2: Match formula to code

   cos_sim = np.dot(A, B) / (np.linalg.norm(A) * np.linalg.norm(B)) matches this formula exactly using numpy functions.

3. Final Answer:

   cos_sim = np.dot(A, B) / (np.linalg.norm(A) * np.linalg.norm(B)) -> Option C

4. Quick Check:

   Cosine similarity formula = cos_sim = np.dot(A, B) / (np.linalg.norm(A) * np.linalg.norm(B)) [OK]

Hint: Cosine similarity = dot product ÷ product of norms [OK]

Common Mistakes:

• Adding vectors instead of dot product
• Multiplying dot product by sum of norms
• Using norm of difference instead of cosine similarity

3. Given two texts converted to sets of words: text1 = {'apple', 'banana', 'cherry'} and text2 = {'banana', 'cherry', 'date'}, what is the Jaccard similarity between them?

medium

A. 0.25

B. 0.6

C. 0.75

D. 0.5

Solution

1. Step 1: Calculate intersection and union of sets

   Intersection = {'banana', 'cherry'} (2 items), Union = {'apple', 'banana', 'cherry', 'date'} (4 items).

2. Step 2: Compute Jaccard similarity

   Jaccard similarity = size of intersection ÷ size of union = 2 ÷ 4 = 0.5.

3. Final Answer:

   0.5 -> Option D

4. Quick Check:

   Jaccard = intersection/union = 0.5 [OK]

Hint: Jaccard = common words ÷ total unique words [OK]

Common Mistakes:

• Counting union incorrectly
• Using sum instead of division
• Confusing intersection with union size

4. The following Python code tries to compute cosine similarity but gives an error. What is the main issue?

import numpy as np
A = np.array([1, 2, 3])
B = np.array([4, 5])
cos_sim = np.dot(A, B) / (np.linalg.norm(A) * np.linalg.norm(B))
print(cos_sim)

medium

A. np.linalg.norm is used incorrectly

B. Vectors A and B have different lengths causing dot product error

C. Division by zero error

D. Missing import statement for numpy

Solution

1. Step 1: Check vector sizes

   Vector A has length 3, vector B has length 2, so dot product is invalid.

2. Step 2: Understand dot product requirements

   Dot product requires vectors of same length; mismatch causes error.

3. Final Answer:

   Vectors A and B have different lengths causing dot product error -> Option B

4. Quick Check:

   Dot product needs equal length vectors [OK]

Hint: Dot product needs vectors of same length [OK]

Common Mistakes:

• Assuming norm causes error
• Thinking division by zero happened
• Ignoring vector length mismatch

5. You want to find related news articles using similarity measures. Which approach best improves accuracy when articles have different lengths and some common words?

hard

A. Use cosine similarity on TF-IDF vectors to reduce common word impact

B. Use raw word counts and Jaccard similarity without preprocessing

C. Compare articles by counting total words only

D. Use random similarity scores to guess relatedness

Solution

1. Step 1: Understand TF-IDF role

   TF-IDF reduces weight of common words, highlighting unique terms in articles.

2. Step 2: Why cosine similarity on TF-IDF helps

   Cosine similarity measures angle between vectors, handling different lengths well.

3. Final Answer:

   Use cosine similarity on TF-IDF vectors to reduce common word impact -> Option A

4. Quick Check:

   TF-IDF + cosine similarity = better relatedness [OK]

Hint: TF-IDF + cosine similarity handles length and common words best [OK]

Common Mistakes:

• Ignoring word importance by using raw counts
• Using Jaccard without preprocessing
• Relying on random scores