One of the incredibly exciting features of the new Google Cloud Vision API is its ability to detect and categorize the base emotional state of the people in an image. Put another way, the Cloud Vision API examines the facial expression of each person in an image and assigns it a basic emotional state such as extremely happy or very sad, making it possible to rapidly triage the firehose of images surrounding an event to identify those images of people happy about it and those unhappy about the event.
GDELT today operates one of the largest deployments in the world of sentiment analysis, assessing upwards of 4.5 billion emotional indicators per day from global news coverage across 65 languages. Yet, to date GDELT has been limited to assessing the emotional content of the text of each news article – it has never been able to characterize the emotional response of people featured in the imagery accompanying the article. Today, through the Cloud Vision API, GDELT is taking the first steps towards realtime global image-based emotional assessment.
This is especially powerful in areas like understanding the characterization of political leaders (a newspaper choosing a very upbeat and positive image of a leader vs an image portraying the leader as downtrodden and exhausted), separating cheering masses of supporters from angry protesters, a press conference featuring a grim faced police caption versus a grinning and jubilant lotto winner, a devastated natural disaster victim versus someone overjoyed that none of their family members were injured, and so on.
For the first time GDELT is now able to combine text-based emotional assessment with the emotional responses of those featured in the imagery accompanying the article, bringing together the emotions of text with those of images. We are just taking the first steps towards this brave new world and the results in just our first few experiments have been nothing short of extraordinary. Stay tuned for more announcements over the coming weeks.