[Paper] Toward Personality Insights from Language Exploration in Social Media

• Paper
• Demo
• My Slide

The main purpose of this paper is to show how social media can be used to gain psychological insights.

Different from other papers in the past which use a pre-compiled word category list (e.g. LIWC),
it uses an open vocabulary approach that allowing discovery of unanticipated language.

Data

• 75,000 Volunteers
  • Facebook Status Update
  • Age
  • Gender
  • Personality (Through Standard Personality Questionnaire)

Architecture

1. Linguistic Feature Extraction
   • N-Gram
     • Point-Wise Mutual Information
   • Topic
     • Probability a person mentioning a topic (Derived from LDA)
2. Correlation analysis
   • Least Squares Linear Regression
3. Visualization
   • Differential Word Clouds
     • Word size represents correlation strength.
     • Color represents relative frequency
   • Standardized Frequency Plot
     • Plot the word frequency against age

Result

Most results confirm what is already known or obvious.
However, I think this method might still be useful to gain insight in other kinds of datasets.