Many of us are at South by Southwest. We’re picking up a lot of leads who want to do interesting pre-launch pilot projects with us. (Perhaps too many!) We’re also seeing a lot of performances by excellent bands you probably haven’t heard of yet, which is a very important thing to do at SXSW.
It is of course an Austin city ordinance that every tech company presenting at SXSW must have a Twitter demo.
So we have one. It’s what we’ve been using to show people how our system understands topics, and how it can do so using real-time data.
Basically, suppose you were to try to follow the hash tag #sxsw. It’s a mess. There are tweets going there faster than you can reasonably read them, and many of them are kind of crap of the form “retweet this for your chance to win a prize!”.
So our demo, sxsw.luminoso.com, does two things: it completely removes tweets from the stream that appear to be uninteresting or spammy, and out of the remaining tweets it shows you the ones that appear to correspond to selectable topic areas such as music, film, food, and (perhaps most relevantly) parties.
When watching another company’s SXSW Twitter demo, I noticed that they emphasized things that were being retweeted a lot — and that every one of them was one that we would want to filter out. The people retweeting weren’t participating in an authentic conversation. They were just being promised something in exchange for retweeting. This has no value to anyone else reading the tweet.
This demo of course works using our fuzzy semantic matching, not exact keywords. I was happy to see it quickly learn that “Spotify House” is referring to a music event, for example, with most of the tweets about it not including anything like the word “music”.
Now if only the demo could tell you which music to listen to. But actually, it seems to know to leave that up to NPR.