By The Numbers: When ZeroBot Gets Confused
One thing ZeroBot is taking a while to learn is how to name the stories it’s creating. It’s gotten to a point where, very often, the title is so, so close — it just needs that one little push. Think small tweaks like capitalizing a word, not including the venue name along with the event name, highlighting the wrong aspect of an event, etc. They’re all fixable problems that we’re gently coaxing ZeroBot to learn.
But sometimes, it’s not just little bloopers that are easily fixable. Sometimes, ZeroBot just plain gets confused. Here’s an example.
Hyperlocal Stories
We’ve been working on capturing hyperlocal stories in the markets we currently cover, not just concerts, sporting events, conferences and the like. We want to find stories like grand openings and niche meetups — stories that would be interesting for people who care about what’s going on in their communities. A new type of story that’s begun cropping up are from college campuses. We’ve captured really cool stories like commencements, greek fundraisers and intimate speaker events.
The Problem With Schools
One difficulty with college campuses we’ve seen ZeroBot have is how muddled stories can get. Even though there are separate events taking place, keywords can blanket all of the events — the name of the school, the name of the mascot, etc. ZeroBot sees everyone talking about what’s basically the same thing, and assumes they all belong in the same story.
“Georgetown University” is a beautiful story that encompasses everything that’s happening on campus leading up the end of the school year.
There’s a great mix of people getting excited for an upcoming track & field meet, celebrating Georgetown Day, teaching assistants hanging out, and reflecting on speakers who stopped by the night before.
It looks great! It’s a well-balanced, overall look at what’s happening on campus. Last night, though, was a different story:
Last night’s Georgetown University story featured three different events — three separate conversations with Speaker of the House Paul Ryan, rapper Wale and Venezuelan Congressman Angel Medina.
The second-largest trending word last night was “Hoya,” Georgetown’s mascot. But the top trending word was “Wale” — people were really, really hyped about hearing him speak. People were also excited to hear Rep. Ryan and Congressman Medina speak, but the overwhelming number of posts last night were from Wale’s conversation.
So what did ZeroBot do? He assumed everyone was talking about the same event, threw all the posts together, and named it “Wale.”
Storytelling Using Social Media
In one of the first “By The Numbers” we posted, we discussed how we taught ZeroBot how to differentiate between the hundreds of New York Fashion Week shows we were capturing, creating individual stories based on each designer’s show instead of lumping them all into one giant NYFW story.
By teaching ZeroBot how to find hyper-local stories, we’re realizing the differences in a much smaller area (college campus) may be a little bit harder to find than the differences in a larger area (an entire city). For now, what we can do is continue to expand ZeroBot’s learnings and increase the amount of data it culls from what its presented with. We’re excited to see how ZeroBot finds its way from confusion to clarity!
Feel free to reach out to us with any questions or comments at team@zeroslant.com.
