Detecting TV Program Highlight Scenes Using Twitter Data Classified by Twitter User Behavior

Detecting TV Program Highlight Scenes Using Twitter Data Classified by Twitter User Behavior and Evaluating It to Soccer Game TV Programs

“…The categorized user group 3 and 4 have very similar features but difference on whether there is significant difference between the use rates of plain-tweet and retweet/replytweet, so that the properties within the categorized user groups are different to the ones within the categorized user groups in my previous work [3]. However, both of the user group 3 and 4 were adopted to investigate detecting the highlight-scenes in the experiment since the balance of using plain-tweet and retweet/reply-tweet affected how many keywords which represent the event contents were included in the tweets on the detected time periods according to the previous work.…”

“…• The number of hashtags, which represents the degree of contribution of information sharing. In my previous study [3], to detect highlight events and generate metadata to the events from twitter data of a soccer game, the twitter data was categorized into four user groups of which each had different behaviors on Twitter as follows; "heavy use of hashtags," "heavy use of retweets", "parallel use of retweets/plain-tweets," and "heavy use of plain-tweets." Therefore, the system adopted classifying the Twitter data into the four user groups by using the Ward's method [17] because of using the same features and the clustering method as the previous research.…”

“…To create the frequency time series of the tweets for each user type, tweet frequencies for certain a time span are calculated and arranged into the timeline of the TV program. The time span of the current system is set to 5 s, which was determined by investigating the time gaps between highlight scenes of the TV program and the tweets regarding these scenes [3].…”

“…Given the above problems, this paper proposes a highlight scene detection method for TV programs based on Twitter data categorized by Twitter user behavior with high accuracy [3], [4]. Using the categorized Twitter data, the proposed method detects highlight scenes of TV programs.…”