Asia Trend Map: Forecasting “Cool Japan” Content Popularity on Web Data


Published on

Brief introduction about Asia Trend Map: "Cool Japan" content popularity forecasting system

Published in: Education, Technology, Travel
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Asia Trend Map: Forecasting “Cool Japan” Content Popularity on Web Data

  1. 1. Asia Trend Map: Forecasting “Cool Japan” Content Popularity on Web Data Shuhei Iitsuka The University of Tokyo Ohma Inc. 2013/08/20 1
  2. 2. Background •  Anime, Manga and Game has become popular around the world. •  Japanese content industries are willing to promote their products overseas under the brand of “Cool Japan”. •  However, localization processes (translation, promoting etc.) take costs a lot of money and time. 2013/08/20 2 Japan in London: Sushi, Manga, Cosplay and Camden – à Sellers need to estimate the product’s popularity in the target market and allocate their resources strategically.
  3. 3. Purpose •  Forecasting each product's popularity around Asian countries based on web data from Twitter, Wikipedia and a search engine. •  Why Asia? –  Close to Japan geographically and culturally à direct economic effect –  Growing market •  Why web data? –  Unauthorized copies are widely distributed around the country à There’s difficulty in catching the trend from the sales data 2013/08/20 3 ?
  4. 4. Demonstration: Asia Trend Map •  This system can forecast about 4,000 Japanese content’s popularity trends following 6 months for 13 countries in Asia. 2013/08/20 4
  5. 5. Model Overview 2013/08/20 5 Twitter Wikipedia search  engine web   data forecasting   model consumer,   user   tweet edit search crawl crawl crawl attribute   extraction training  data   (Sales  in  Japan) SVR
  6. 6. Wikipedia Data Attributes •  Edit –  Monthly Edit Count, Monthly Unique Editor Count, Average Edit Count Per User ... •  Link –  Number of Forward Links, Number of Backward Links ... •  Content –  Number of International Links, Page Size, Number of Sections ... 2013/08/20 6 NARUTO 나루토 火影忍者 Jump   (Magazine) Ramen Forward  Link Backward  Link International  Link Wikipedia  link  example:
  7. 7. month:  m Twitter and Search Engine •  Twitter: Extract number of tweets which includes the product name (monthly) •  Search Engine: Extract number of times the product name is searched (monthly) •  We get each product’s local name utilizing Wikipedia database. 2013/08/20 7 NARUTO Wikipedia 火影忍者 나루토 Twitter Search   Engine T_(m,  China) T_(m,  Korea) S_(m,  China) S_(m,  Korea)
  8. 8. Pre-processing on Training Data •  Sales of Manga suddenly increases when new volume is out. à We connect the peak with lines and make use of this as training data. 2013/08/20 8
  9. 9. Experimental Results •  Prediction precision is improved by applying attributes of multiple web services. •  Especially, Wikipedia data took an importance role in predicting the trends in more distant future. 2013/08/20 9
  10. 10. Experimental Results •  Among the Wikipedia data attributes, Page Content (Number of international links, Page size, etc.) took the most important role in predicting the trend. 2013/08/20 10
  11. 11. Conclusion •  We built the forecasting system of Japanese cultural products from web data •  We launched a website based on this system: Asia Trend Map •  We'd like to contribute to strategic planning process of "Cool Japan" with this. 2013/08/20 11