Towards Author Profiling - A Methodology for Gender Expression Classification
summary
Methodology collects tweets via the Twitter API and Python and extracts about 160 linguistic and meta features — characters, word counts, morphology, punctuation, emoticons and psycholinguistic dictionaries — to classify gender expression. Uses Naive Bayes, decision trees and SVM; on ~200k tweets from 60 profiles SVM hits ~68% per-tweet accuracy and ~80% when concatenating tweets per user. Model variants are already deployed for investigative and marketing uses but raw data and some models remain restricted.
AI-written summary · about
chapters (12)
- 0:00Introduction and motivation
- 3:08Problem of anonymity and fake profiles
- 4:36Gender expression classification goal
- 6:47Methodology and preprocessing
- 12:14Meta-attributes and model architecture
- 13:39Dataset and machine learning algorithms
- 14:28Performance evaluation results
- 17:04Psycholinguistic cues analysis
- 18:15Applications beyond forensics
- 19:20Dedication and demo introduction
- 20:06Demo on meta-attributes extraction
- 22:13Conclusion and resources
related talks
22mMaking Natural Language Processing Robust to Sociolinguistic Variation
23mText mining online data with scikit-learn
46mPractical Machine Learning and Rails
39mThe art and science of making computers understand sentiment and emotion
52mKeynote address
01h 23mBeyond Sentiment Emotion Mining with Python and machine learning
11mClassifying Spoken Location Names
38mMeasuring Psychological Traits using Social Media