An Evaluation System for News Video Streams and Blogs

1
An Evaluation System for News Video Streams and
Blogs

• Daisuke Kitayama and Kazutoshi Sumiya
• University of Hyogo (???)
• ACM Symposium on Applied Computing 2006

2
Objective

• Evaluate/manipulate the sentiment of a video news
  report by varying the scene order
• Apply similar technique to weblog to evaluate the
  sentiment of a blog post

3
Idea

• The sentiment of a video news report can change
  depending on the order of the scenes
• For example, showing positive scenes after
  negative scenes gives an overall positive
  sentiment

4
Extension to Blogs

• The sentiment of a blog post depends on the
  sentiments of words in the latter part of the
  post
• Under this framework, we can compute a sentiment
  score for each blog post

5
Sentiment of a Word in a News Report

• Use a manually-defined dictionary to determine
  whether a word is positive, negative or neutral
• Compute the sentiment score of a word using the
  formula
• eij is the jth sentiment word in the ith scene
• S is the total number of scenes
• E is the total number of occurrences of sentiment
  words in the entire news stream
• Later-occurring words (high i an j) have higher
  weights

6
Sentiment of a News Report

• Computed as a ratio defined as

7
Sentiment of a Word in a Blog Post

• Treat an entire blog post, including comments and
  trackbacks, as a single continuous document
• The sentiment score of a word is defined as
• ei is the ith sentiment word in the blog post
• B is the total number of occurrences of sentiment
  words in the blog post
• Words occurring later (high i) have higher weights

8
Sentiment of a Blog Post

• The sentiment score of a blog post is defined as

9
Experiment

• Work-In-Progress
• The goal is to compare the sentiment score a news
  stream with the sentiment score of the
  corresponding blog posts