Inflow
How does information flow through the wildfire that is the internet? A lone man sets out to find the answers…
Part 1: Fires
As a Californian native I have seen a few wildfires. These monstrous fires move through forests at seemingly supersonic speeds, seeping into every breathable corner, and destroying everything in their wake. At their worst they destroy innocent lives while even at their best they can destroy our century old beloved redwood trees.
Information on the internet is a wildfire.
Over the past few years I’ve become increasingly intrigued by how viral stories are spread on the internet. Like wildfires, some stories move through the internet at light-speed, oozing into every corner of the Web, and trampling all stories in their way. At their worst, malicious stories can destroy lives and supplant century-long facts.
At the beginning of this year I started a project series, Inflow, to allow people to interact with the internet in new ways and see overarching trends and patterns on the Web. My hope is that these projects will open people’s eyes to the living and breathing fire that is the internet.
Inflow is very much still a work in project, but here is where I am today.
Part 2: How did I get here?
If Alice in Wonderland were written today, Alice would be falling through the rabbit hole that is YouTube recommendations. From one video to the next to the next YouTube recommendations can transport us from one reality to another.
YouTube’s dark realities include videos about vaccines causing autism, climate change being a hoax, and the Earth being flat.
Technically, the YouTube recommendation system is a cold and lifeless algorithm that outputs videos “related” to a selected video. Inflow challenges this perspective and instead presents the YouTube recommendation algorithm as a living being with a personality.
In “Inflow: YouTube Recommendations,” a user selects a video to be shown in the center. The user is shown the top three recommended videos given by YouTube for that initial video and when the user clicks a node the top three recommendations for the selected node/video are shown. Thus we create recursive patterns of recommended videos.
By interacting with “Inflow: YouTube Recommendations” I want people to notice the YouTube algorithm’s quirks and tendencies. For instance, the demo above shows how one might fall into the YouTube rabbit hole that is climate change hoax videos. Additionally, I visually communicate that the algorithm has “life” by creating fluid motions and bounces. This visualization provides a rare holistic broad view of the content YouTube recommends to us.
I hope that visualizations like “Inflow: YouTube Recommendations” can help people become more aware of the content served to them by supposedly lifeless algorithms and the role these algorithms play in pushing people towards extreme beliefs.
Part 3: Tell me a story
“Tell me a story,” I mutter as I pull down on the NYT app to refresh its headlines for the tenth time that day. If you’re like me, you may have an obsessive interest in keeping track of news.
Late one November night I decided I never wanted to miss a story again. I wrote some scripts to continuously collect headlines and article descriptions (along with metadata like date, publisher, etc) from the 22 most “influential” political new organizations in the United States (note: I am exclusively collecting articles categorized as “politics” by their publisher).
100,000+ headlines, some infrastructure-building, and six months later I have some interesting stories to tell:
“Trump” count grouped by organization, date
Headline count by date
Headline count by organization
Candidate count by date
These visualizations provide insight into the media we consume and the stories they choose to cover. I continue to collect this data and I hope when I release it users will be able to notice broad differences between media storytelling.
Part 4: The next chapter
I am actively developing new projects related to Inflow. Using machine learning I hope to identify patterns in the data (like topic modeling). I hope to combine the user experience from “Inflow: YouTube Recommendations” with my headline data to create a new way of browsing news. I plan on using anomaly detection to detect spikes in keywords and identify why those spikes may have occurred. All these projects are driven by a common desire to learn how we’re told information.
There is a problem on the internet: stories spread faster and farther than we can keep up. I hope, in some small way, that Inflow sheds light on this problem.
We’ve got to notice the fire before we can stop it.
Amar Ramachandran is an incoming junior studying Computer Science at the University of Michigan.
Should you be interested in the data I have collected, or otherwise just want to say hi, I can be found at amarjayr.com
