What if a Robot Could Laugh at Your Jokes? (Laughter Prediction with Natural Language Processing)
summary
Transcripts total about 160,000 words and 5,500 laughs from Carlin and Chappelle across 19 hours, preprocessed into rolling groups of three sentences and converted to embeddings (Word2Vec). A convolutional neural net (embedding layer, kernel size 3 with 256 filters) outperforms RNN/LSTM on this small set, yielding roughly 76% weighted F1 but with false positives and overfitting after ~4 epochs. Next steps: add more comedians, try transfer learning, and fuse audio/video for audience and facial cues.
AI-written summary · about
chapters (18)
- 0:00Intro
- 2:50What is NLP?
- 4:36Joke Structure
- 8:27Data Science Process
- 11:07About the Data
- 14:35Word Embeddings
- 16:14Data Summary
- 16:46Transcript Visualization
- 17:34Word Cloud
- 18:05Topic Visualization with LDA
- 20:07CNN Hyper Parameters
- 22:05Model Results
- 23:45Code Demo
- 26:32Recommendations
- 28:06Next Steps
- 28:26Questions
- 31:31Credits
- 31:46Contact Me
related talks
33mWhen Stand-up Comedy Meets Machine Learning
20mCapturing, Representing, and Interacting with Laughter
01h 51mNatural Language Processing in Python
13mTransfer Learning: Leveraging Pretrained Models with Limited Data
20mNo Need to Laugh Out Loud: Predicting Humor Appraisal of Comic Strips Based on Physiological ...
31mRobot on Rails
45mArtificial Irreverence: Improvised Theatre with Robots
59mTechnical Workshop: Sarcastically Speaking: Unlocking Multi-modal Sentiment Analysis With NLP and Facial Expressions