January 26, 2017
NEC Personal Computers, Ltd.
Three I's, Inc.
NEC Personal Computers ("NECPC") has been conducting joint development on artificial intelligence with SRI International ("SRI"). NECPC is pleased to announce today that it provided the resulting technologies to Three I's, Inc. (head office: Koto-ku, Tokyo), and Three I's has commenced sales of the service as the new online advertisement distribution network called CANDY.
CANDY is server software that selects advertisements and articles relevant to Web articles such as news. It understands the meaning of articles written in Japanese by linking with NECPC artificial intelligence, and displays the most relevant articles and advertisements on the screen.
Several media partners have been testing CANDY since October 2016. In these tests, the software has proven effective for the operation of website businesses, primarily because of an approximately three-fold increase*1 in the click-through rate to the content selected by NECPC artificial intelligence, compared with displays of recommended web content based on popularity. Given this promising outcome, Three I's decided to provide CANDY openly in January 2017.
NECPC artificial intelligence was jointly developed by NECPC and SRI, and it provides the functionality to decipher the relevance of a Japanese sentence. It also conducts machine learning of large volumes of content uploaded to the Internet in the Japanese language, accumulating the knowledge in its knowledge dictionary, which is sufficient to understand the topics covered by average people living in areas where Japanese is spoken, and understanding the topics of Web content.
For example, the word "hanazono" means a garden where flowers are planted. The element is equipped with the intelligence to know that this word is also commonly used as the name of a rugby stadium in a rugby context, and it has a strong relevance to such phrases as "high school rugby."
In conventional online advertisements, those advertisements that have a strong relevance are displayed by acquiring browsing histories using cookies, a technology that identifies website visitors (users). This system displays advertisements that have a weak relevance with the articles that interest users and that users are viewing. This results in the challenge of attracting the user's attention to advertisements (example: an advertisement for condominiums or clothes, which a user has previously searched for, pops up when the user is viewing articles about professional sports teams).
CANDY, on the other hand, uses a deciphering process based on relevance with games, teams and leagues, automatically selects and recommends content about additional related information and advertisements for fan goods, considering the articles that users are viewing with interest.
Thanks to these relevant recommendations based on common-sense perspectives, the click-through rate to advertisements and the purchase rate of online shopping products after clicking should increase.
Test operations resulted in an average purchase rate of more than 3% when the number of clicks on product sale advertisements for major Internet shopping is used as a parameter, and the rate of conversions into purchases is calculated*2.
NECPC will incorporate the next-generation version of this NECPC artificial intelligence into My Time Line, its own news application. My Time Line will be equipped with the functionality to estimate which topics have attracted a user from the huge volume of online news, updated daily, by calculation in the CPU of the terminal, and to display news and advertisements that cater to user's preferences. This type of personal information, such as information about interests, is managed inside the user's terminal (if multiple computers are used, data such as information on personal tastes are encrypted for synchronization).
Going forward, NECPC will continue to enhance the functionality of the product in addition to My Time Line, and will strive to provide the functionality to understand the user's preferences and present actions, and will provided recommendations via a personal computer of useful information for the user. Through these initiatives, NECPC will continue to work on further research to ensure that terminals actively support people's lives, like assistants, for purposes such as booking tickets for concerts of their favorite artists or proposing travel schedules.
As a leader in artificial intelligence (AI) R&D, SRI has developed natural language text processing technologies for personalization, prediction and emotion recognition. NECPC has licensed text analytics and machine learning software from SRI. This technology enables personalized context and online advertising that are fundamental to how the CANDY operates.
As a leader in artificial intelligence R&D, SRI has developed natural language text processing technologies for personalization, prediction and emotion recognition. NECPC has licensed text analytics and machine learning software that will enable personalized content and online advertising that is fundamental to how CANDY operates.
From October to December 2016 the click-through rate AB comparison test of related articles was conducted in major online news portal websites. The method used is to display A (CANDY is not used) and B (CANDY is used) articles in same quantity at random to approximately 18 million page views. The CANDY achieved the click-through rate approximately 3 times higher than A.(Note 2)
For three months from October to December 2016, advertisements of products recommended by the CANDY were posted on major online news portal websites, and the number of clicks was measured. The results of purchase history data were received from major Internet outlets, and the purchase rate per click was calculated.