Sarcasm_Detector
URL: https://github.com/TheMoskowitz/Sarcasm_Detector
Description
A neural network for sarcasm detection trained on the Reddit Sarcasm Database.
Methods and Models
- Dataset: Reddit Sarcasm Database
- Approach: Neural network-based sarcasm detection
- Additional Experimentation:
- Sentiment Analysis: Attempted to use a theory that sarcastic sentences contain both highly positive and highly negative sentiments close together, but it did not improve results.
- Larger Dataset: No significant improvement observed despite using an expanded Reddit dataset.
Results
- Accuracy: ~57%
- The best academic results found were in the 60% range.
- Yahoo claims to have a proprietary model with 80% accuracy.
Future Improvements
- Contextual Analysis: Incorporating preceding comments might enhance sarcasm detection.
- Multimodal Features: Using audio could simplify the problem and significantly improve accuracy beyond 80%.