Abstract
In recent years, a large number of TV series has been broadcasted. This huge amount of video content does not match the viewer's busy lifestyle. In many cases, people do not have time to watch the entire program, catch up with the episodes of TV programs or re-watch forgotten video content. Therefore, techniques such as video summarization, that attempt to solve these problems, have been studied. TV series, unlike news and sports, do not have an underlying structure that can be useful for summaries. We propose an automated summarization framework for creating video summaries that preserving the story line to the level that user can watch the summary instead of the original content. We comprehensively used features characterizing videos such as semantic information and audiovisual information of the TV metadata. We create a model that assigns importance scores to each segment of the TV series by performing ranking learning using TV metadata. As a result of calculating the average area under the ROC curve for two titles was 0.612 and 0.582 for each title.
| Original language | English |
|---|---|
| Pages | 47-51 |
| Number of pages | 5 |
| Publication status | Published - 2019 |
| Event | 10th International Multi-Conference on Complexity, Informatics and Cybernetics, IMCIC 2019 - Orlando, United States Duration: 12 Mar 2019 → 15 Mar 2019 |
Conference
| Conference | 10th International Multi-Conference on Complexity, Informatics and Cybernetics, IMCIC 2019 |
|---|---|
| Country/Territory | United States |
| City | Orlando |
| Period | 12/03/19 → 15/03/19 |
Keywords
- Machine learning
- Metadata
- Natural language processing
- TV series
- Video summarization