Top

COLING-16

This is the page which provides the source code and dataset we used for COLING-16 System Demonstrations.

• Paper: Kotonush: Understanding Concepts Based on Values behind Social Media
• Authors: Tatsuya Iwanari, Kohei Ohara, Naoki Yoshinaga, Nobuhiro Kaji, Masashi Toyoda, and Masaru Kitsuregawa.

Errata

There was a bug in our program which prevented us from counting antonyms. We fixed it and the fix affected some evaluation results and case studies. Please see this.

Source

We made a front-end system (Ruby on Rails) and a back-end system (C++, Ruby, etc.). They are not refactored and will not be maintained.

• See a snapshot of the front-end system at GitHub.
• See a snapshot of the back-end system at GitHub.
  • We have a plan to maintain the back-end programs in an open repository (Under construction…).

Dataset

We make 2 types of dataset available on this page.

humans (for gold-standard)

This data contains the human annotations to make the gold standard ordering. They are written in Japanese.

Download:

• data

The data has 3 domains;

• twitter
  • Twitter users regardless of their genders
• twitter_men
  • male Twitter users
• twitter_women
  • female Twitter users

Please see human annotation format section for more detail.

orderings (system-generated orderings)

This data contains the gold and system-generated orderings. They are written in Japanese and English.

Download:

• data (Japanese, Original)
• data (English, Translated from Japanese)

The data has results of 4 types methods for 2 domains;

types;

• gold
• baseline (in men/women)
• svm_domain_unaware (in men/women)
• svm_domain_aware (in men/women)

domains;

• all
  • Twitter users regardless of their genders
• men
  • male Twitter users
• women
  • female Twitter users

Please see orderings format section for more detail.

Contact

If you have trouble to use them, please contact me (Tatsuya Iwanari nari@tkl.iis.u-tokyo.ac.jp).