• Considerations

    Carl Jamilkowski • 07/03/2022 at 10:22 • 0 comments

    • Things to analyze
      • Word recognition -> specific emoji (e.g. clown = 🤡)
      • Sentiment analysis
        • Basic implementation should be easy with something like TextBlob
        • Next level would be
          • Looking at tone of voices, not just the words
          • NLP
          • Video Facial Expression Recognition System
            • Avoiding this for the initial project
    • Special cases
      • Multiple speakers
        • Who to analyze?
        • Average sentiment between the two?
      • Cringe words/phrases (creepy topics, fringe/conspiracy theories)
        • Build a hard coded library?
        • Is there a list I could auto pull to get the latest? How to keep up with Twitter trends?
    • Try to avoid internet dependencies
      • Offline processing for privacy and bandwidth concerns
      • Use local models for AI/NLP
    • What emoji to show?
      • Don't refresh too often
      • Wait until end of sentence to gauge sentiment?
      • When conversation is neutral sentiment
        • Use direct word/phrase recognition
        • Use "judgement" mode where it responds to things being said