An unsupervised approach for detecting differences in self-presentation on OkCupid.

A final project for Applied Natural Language Processing, Fall 2015.

We will use the following methods to detect differences in self-presentation.

• Pointwise mutual information for generating features
  • unigrams, bigrams, and trigrams used by at least 1% of users
• Dimensionality reduction with Principle Component Analysis
• Clustering with k-means
• Keyphrase extraction for cluster "descriptions"
• Look for differences in demographic distributions by cluster

0: My self summary
1: What I'm doing with my life
2: I'm really good at
3: The first thing people notice about me
4: Favorite books, movies, tv, food
5: The six things I could never do without
6: I spend a lot of time thinking about
7: On a typical Friday night I am
8: The most private thing I am willing to admit
9: You should message me if