Lexicon-based approaches (VADER) in NLP - Model Metrics & Evaluation

Metrics & Evaluation - Lexicon-based approaches (VADER)

Which metric matters for Lexicon-based approaches (VADER) and WHY

VADER is used to detect sentiment in text, like positive or negative feelings. The key metrics are Accuracy, Precision, Recall, and F1-score. These show how well VADER guesses the right sentiment.

Accuracy tells us how many texts were labeled correctly overall.

Precision tells us, when VADER says a text is positive, how often it is really positive.

Recall tells us, out of all truly positive texts, how many VADER found.

F1-score balances precision and recall to give a single score.

We use these because VADER works on words and rules, so we want to check if it really matches human feelings well.

Confusion Matrix Example

    Actual \ Predicted | Positive | Neutral | Negative
    -------------------|----------|---------|---------
    Positive           |   80     |   10    |   10
    Neutral            |   15     |   70    |   15
    Negative           |   5      |   10    |   85
    

This shows how many texts VADER labeled correctly or wrongly for each sentiment.

Precision vs Recall Tradeoff with VADER

If VADER has high precision for positive sentiment, it means when it says "positive," it is usually right. This is good if you want to trust positive labels.

If VADER has high recall for positive sentiment, it means it finds most of the positive texts. This is good if missing positive feelings is bad.

For example, if a company wants to find all happy customer reviews, high recall is important.

If a company wants to only show very sure positive reviews, high precision is better.

Good vs Bad Metric Values for VADER Sentiment

Good: Accuracy above 80%, Precision and Recall above 75%, F1-score balanced and high.

Bad: Accuracy below 60%, Precision or Recall very low (below 50%), F1-score low or very unbalanced.

Good metrics mean VADER matches human sentiment well. Bad metrics mean it often guesses wrong or misses feelings.

Common Pitfalls in VADER Metrics

• Accuracy paradox: If most texts are neutral, accuracy can be high by always guessing neutral, but precision and recall for positive/negative will be poor.
• Data leakage: Using test data in tuning VADER rules can give false high scores.
• Overfitting: Tweaking VADER too much on one dataset may not work well on new texts.
• Ignoring class imbalance: If one sentiment is rare, metrics like accuracy can be misleading.

Self Check

Your VADER model has 98% accuracy but only 12% recall on negative sentiment. Is it good for production?

Answer: No. The model misses most negative texts (low recall). It may label almost everything as positive or neutral to get high accuracy. This is bad if finding negative sentiment is important.

Key Result

For VADER sentiment analysis, balanced precision and recall above 75% indicate good performance beyond just accuracy.