Demographic Factors Improve Classification Performance

Introduction
1. Socio Lingusits
  1. Language Use
    1. Demographic Factors
      1. Age
      2. Gender
    2. Results
      1. Yond people and womenare are more creative
      2. Men and older people are less creative
2. Natural language Processing
  1. Initial Assumption
    1. Language is Uniform
  2. Test Data
    1. Social Media
      1. Model
        Bad performance
        Without demographic factors
        Good performance
        With demographic factors
        Multiple studies
      2. NON UNIFORM
    2. News
      1. UNKNOWN
        ?
  3. Text classification tasks
    1. Sentiment analysis
      1. Determining the popularity of a document
    2. Topic identification
      1. Assigning a high level concept to a document that captures its content
    3. Author attribute clasification
      1. Inferring demographic factors from linguistic features
Research Focus
1. Encoding demographic factors?
  1. Embeddings conditioned on respective demographic variable
    1. Distributed representations of words in a vector space, capturing syntactic and semantic regularities among words
2. Effect of demographic factors on performance?
  1. F1 performance of classifiers
    1. Text classification tasks
    2. With demographic info
      1. CONSISTENTLY BETTER
        Small improvements
        On average
        In 8/30 cases statistically significant
        p<0.005 bootstrap sampling tests
        ?
    3. Without demographic info
    4. Five different languages
      1. Danish
      2. Franch
      3. German
      4. English
        US
        British
    5. ?
  2. Using Logistic regression models
    1. ?

Nächster

Demographic Factors Improve Classification Performance

Beschreibung

Zusammenfassung der Ressource

ähnlicher Inhalt

	Erstellt von Ivan Zapreev vor mehr als 9 Jahre