Week 4 SOC Homework

[CristyTarantino]

• Prediction | calculate the prediction for 100% pos, and 100% neg review | Required
• Sentiment Analyzer for "Apps for Android" | repeat this same process for "Apps for Android" dataset
• Improve the Sentiment Analyzer | build a better sentiment analyzer and comment both your code and your data exploration
• Change treshold | Change the treshold from 0.5 to 0.2, and rerun the code. Give a commentary in plain English about how that changed precision and recall. What does that mean? What is now included that wasn't before? What part of it is good? What is bad from our Task perspective. Remember: our task was to identify Dissatisfied reviews. https://en.wikipedia.org/wiki/Precision_and_recall
• Add features. | Explain which features you chose, implement them, and write a commentary on your results. Feel welcome to use NLTK's built-in sentiment analyzer or any other research that you can find and understand
• len_features(dataset) | Add two features: (1) length of review (in thousands of character) - truncate at 2,500 (2) percentage of exclamation marks (in %)
• Refactor the code above | "Be lazy. Not just lazy but proactively, agressively lazy." Remove duplication. create a single function that takes in data and spits out all success metrics across all of your algos.