Getting Sentimental: exploration of iOS sentiment analysis
summary
Built a bare‑bones iOS app (SwiftUI, Swift Data) that scores sentiment with Apple's Natural Language framework (polarity range -1.0→1.0), discovers NL mislabels with slang and multilingual input, tests OpenAI (GPT) which returns more accurate labels but flags a violent invite as -2.5 (illegal) and raises privacy/cost/offline concerns, and trains a local Create ML classifier on roughly 60 labeled examples that runs on‑device and matches ChatGPT‑like accuracy for this narrow domain.
AI-written summary · about
chapters (20)
- 0:00Introduction and welcome
- 3:12Building a feedback app
- 6:42What is NLP?
- 11:58Natural Language framework basics
- 16:30Sentiment scoring with NL
- 18:58Limitations of basic sentiment analysis
- 20:05Sentiment analysis in the real world
- 21:30Breaking down opinions
- 23:17Problematic feedback examples
- 26:27Modern sentiment analysis methods
- 27:31How ChatGPT uses NLP
- 28:11Transformer models explained
- 31:46BERT and contextual embeddings
- 33:10Testing ChatGPT for sentiment
- 35:58Challenges with API-based solutions
- 36:19Training models with Create ML
- 37:45Evaluating custom sentiment models
- 39:12Key takeaways from experiments
- 40:34Embracing experimentation and risk
- 41:55Closing thoughts and thanks
related talks
17mExploring Natural Language Processing
39mSentiment Analysis, its techniques and applications
45mBuilding Emotionally Intelligent Apps using Azure AI
24mParsing natural languages in Swift
32mUnpaired Sentiment-to-Sentiment Translation: Using Reinforcement Learning
01h 05mSocial Text, Sentiment, and Tone Analysis
25mGetting Started with Machine Learning in your iOS apps
43mSentiment Analysis in Action: Building Your Real-time Pipeline