Week 2 - Coding Exercise is not advanced

Week 2 - Coding Exercise

I tested sentiment analysis by adding a variety of real-world reviews typically found on platforms like Google Maps and Yelp. After running at least 40 prompts, I found around 20 of them to be inaccurate.

Here are a couple of examples:

  • Example 1: “Loved the burger. But fries were not great” – This should be a neutral or mixed review, but it’s labeled as positive.
  • Example 2: “Just alright! Ha Ha Ha!” – This is a neutral or negative comment, but it’s tagged as positive. If I remove the “Ha Ha Ha,” it correctly gets labeled as neutral.

In Exercise 2, I noticed the following:

  • Changing the 5th review to “3.5/5” correctly labels it as neutral.
  • However, when I change the 4th review (while keeping the 5th review as “3.5/5”) to “Love it!”, the 5th review is incorrectly tagged as positive.

These inconsistencies make me think the sentiment analysis still has room for improvement.

If anyone has any game-changing resources or tips for mastering sentiment analysis, I’d love to hear them! I’m a beginner

1 Like

in your exercise 2, you mentioned changing the 4th review to love it but didn’t mentioned what review marks was given out of 5 compare to the 5th review?

Sentiment analysis is considered one of the evaluation metric related to an llm, but when it comes to going more precise to getting more appropriate llm, one needs to add/use more evaluation metrics such as cosine similarity, perplexity, ROUGE and BLEU SCORE, Precision, Recall, F1 score.