Yahoo! is a technology for evaluating posted comments using the "Natural Language Processing Model [AI] Using Deep Learning" introduced for the purpose of soundening "Yahoo! News Comments" on September 18. The API is now available free of charge. From the same day, we have started accepting applications for posting service providers in Japan.
By utilizing this API, posting service providers can evaluate comments posted on their services with AI, and based on that, they can delete comments and rearrange the display order.
In Yahoo! News comments, in addition to human patrols by a specialized team, Yahoo! uses AI developed in-house to delete an average of about 20,000 inappropriate posts such as slander and slander per day. It is said that it is being implemented.
In addition, AI scores constructive comments that meet the conditions such as "objective and presenting grounds if necessary" and "providing new ideas, solutions, and insights". From 2018, it introduced a unique technology "constructive comment ranking model" that displays the ones with the highest scores.
This constructive comment ranking model is based on  remarks that arouse controversy based on one's own opinions,  remarks that are objective and provide evidence if necessary, and  new ways of thinking. Comments that meet one of four conditions: a solution, an insightful statement, and  a rare story about an article are given a high score.
This time, it is the technology of this "constructive comment ranking model" that will be provided to the outside free of charge. Posting service providers will introduce a large amount of learning data and calculations required for AI development by introducing the AI technology that is used to counter comments on Yahoo! News, which has an average of 330,000 or more posts per day. It is said that it is possible to start measures for soundness of comments without spending initial investment such as costs.
For example, "improvement of display order" that uses the given score as the value of the comment display order, "improvement of patrol" that prioritizes comments with a low given score and targets for patrol, and comments with a high score It will be possible to utilize comments with scores in various ways, such as "recommendations" that prioritize the display of many articles.