/NLP_DisasterTweets

Kaggle Challenge - Natural Language Processing with Disaster Tweets

Kaggle - Natural Language Processing with Disaster Tweets

Kaggle Challenge

Notes taken from the dataset

  • id (1 to 10.9k)
  • 222 unique values (keyword) out of 7503 unique values (text)
  • keyword = word that shows disaster

Questions and Ideas

  • Q: How does locations affect keywords and how they interact?
    • Maybe subset each location and see which word is more common for each location. Stratify for each location
  • Find ways to detect metaphors of keywords --> Metaphors can lead to false positives
  • Look at words around they keywords
  • Maybe depending on the emojis it can see if the message is bad or not. e.g. happy emoji does not mean danger (DO THIS LATER):
  • Exclamation mark (❗️ ) could indicate danger tweet

Algorithms Result (See which one is the best)

  • Long Short Term Memory Networks (LSTM)