Extracting the Discussion Structure in Comments on NewsArticles

1
Extracting the Discussion Structure in Comments
onNews-Articles

• Speaker ???
• Anne Schuth, Maarten Marx, Maarten de Rijke,
• Universiteit van Amsterdam
• Proceedings of the 9th annual ACM international
  workshop on Web information and data management
• Year of Publication 2007

2
Outline

• Introduction
• Data description and acquisition
• Data exploration
• Thread structure
• Conclusion

3
Introduction

• In the past few years, the World Wide Web saw
  some fundamental changes among which the
  phenomenon known as user generated content
  users, as opposed to owners of websites, are now
  able to add their content.

4
Data description and acquisition

• Terminology- An article is a complete
  news-article published on a news-site.- The
  comments on articles consist of at least the
  name of the author and the comment itself.- A
  comment-thread, or just thread is a flat list
  consisting of comments, in reverse
  chronological order.- A commenter is person who
  comments on an article.

5
Data description and acquisition
6
Data exploration

• Articles
• Comment threads

7
Data exploration
8
Data exploration

• Comments

9
Data exploration
10
Data exploration
11
Data exploration

• Thread structure- Commenters not only comment on
  news articles, they also react on other
  comments, using the news article commenting
  facility.- We can define this reacts-on
  relation declaratively as follows

12
Thread structure

• Now we describe our baseline algorithm for
  identifying the reacts-on relation between
  comments.
• It implements the clue commenters name appears in
  comment by a case-insensitive string match
  between each preceding commenter and the full
  text of the current comment.

13
Thread structure
14
Thread structure

• We will describe three methods for mining the
  reacts-on relation.- method B Word boundaries
  In this method we tokenize both the commenters
  name and the comment, and check whether the
  first list of tokens occurs as a sublist of the
  second list of tokens.- method C POS-tagging
  plus loose match Do a part-of-speech tagging of
  all comments.- method D _at_-trigger plus loose
  match A name and an n-gram match if the n-gram
  is preceded by the _at_-symbol and the name and
  n- gram are similar.

15
Thread structure
16
Conclusion