不少Facebook的用戶也會在自己的平台上發佈各式各樣的圖片,包括生活照片、惡搞圖片,甚至一些含有惡意文字的圖片。由於每天也有超過一億的圖片,Facebook的管理人員未能逐一審查,便推出了人工智能Rosetta去偵測惡意的圖片文字。
Rosetta運用機器學習(machine learning)令圖片上的文字轉成可供機器閱讀的文字。即使圖片上的文字不是英文或拉丁字母,例如阿拉伯語和印度語,也能有效地轉換。現時Rosetta每天更可以審查超過10億的圖片和影片,有助Facebook 管理自己的平台。
相信Facebook在不久的將來也能利用此功能去分析用戶圖片上的文字,從而去了解用戶的喜好、興趣,使商家在製定廣告時,可以更精準地鎖定廣告受眾。
Facebook’s Rosetta AI can detect texts from images
With over 300 million images being uploaded on Facebook everyday, Facebook is now deploying a latest AI system, Rosetta, to read texts images in order to improve overall user experience on the platform.
Rosetta understands texts that appear in images in different forms, across languages and media. To this date, the AI model is able to understand the context combining text and graphics in over a million of images and videos posted on Facebook in languages other than English and Arabic letters. In a long run, this technology will be helpful in optimizing search quality in graphics and monitor post feeds that violate Facebook content policies.
It is believed that Facebook will as well apply this function to understand users’ habits and interests within its platform in the future. It is also useful for companies to target the right audiences in advertising campaigns for an optimal ad outcome.
Comments