Redefining Spam, in the age of Twitter

by Vid Luther on May 13, 2009

in blog, mysql, other, personal, php, web

For the past few months, I’ve been helping my friend develop and market Philtro .
We’ve gone through various iterations of the elevator pitch for it, and the one that seems to be kinda working, is: “It’s like a spam filter for your Twitter account.”

At SXSW, I got the opportunity to talk to Guy Kawasaki about this tool, and he said “There is no spam on twitter, if you don’t like it, don’t follow them”.

While that’s an easy way to handle spam, I also realized that the word Spam means different things to different people.

On Twitter, nothing is UCE. It’s easy to block the profiles with the attractive women, selling Blackberries, iPhones, and who want to chat with me in private with a webcam because 140charactersistoosmalltodiscusstheirdesires.

Philtro is not trying to eliminate viagra ads from your tweetstream. The aim is to help you make the most out of twitter, friendfeed, facebook without alienating people.

Example:

I follow Dan, because he’s a great guy to follow when it comes to design patterns, PHP and Macs. He’s got great insight into those topics, he’s also an avid fan of Twilight, Traveling pants, and underwater basket weaving. I am not.

I want to see what he has to say about the topics I share with him, but I don’t care about his latest Twilight fanclub meetings.

This is where Philtro comes in, it figures out the topics I care about based on the training I give it. “Training” means marking 50 unique
tweets as “Thumbs up ” or “Thumbs down”.

So, Philtro gets rid of uninteresting tweets to you. It’s not a spam filter in the traditional sense of the word, it’s not recommending Twitter users to you (yet…). I can’t think of a word for it, but the best analogy would be a chain of emails at work, to which you’re CC’d on, but the topic is something you have no say on. You can’t remove yourself from other people’s REPLY ALL button, so you’re just stuck deleting emails all day.

If you want to get into the private beta faster, use my super duper special link .

It’s not Unsolicited, and the Tweeter is not “Junk”, what is it?

  • Good question, Brad, and great answer Paul....I was wondering about that too!
  • @Brad: That's a great question. The short answer: don't worry, we err on the side of caution by only rating items that contain content that you've explicitly seen before. If a tweet comes through with content that's completely new to you (and, by extension, you've never rated before), we'll show it to you.

    The way we see it, your incoming tweets fall into a couple of categories: (1) those that you care about, (2) those you don't care about and (3) everything else.

    You're describing something that initially falls into #3. Once you rate that particular item, it'll then fall into #1 or #2.
  • Brad,
    We're working on a recommendation and compatibility engine which will allow you to find other people talking about things you do seem to be interested in. This could lead to you going through their stream and deciding you like something else that they talked about. It's not perfect yet, but your concern about not expanding the horizons is duly noted. We'll see what we can do, or not do :)
  • Brad
    @Nico, I also thought 'Interest Filter' from reading the explanation above.

    The idea does sound useful, however it seems like it might be 'building a silo' and only allowing things you explicitly say are interesting to you. What about things that are new to you, and don't know if they're interesting? Might this be overzealous in some instances?
  • It's sort of a filter within a filter - so while you don't want to shut out Dan completely, you do want to filter out his random ramb lings. And this filtering algorithm is not exactly like Netflix or Amazon because those focus on "what other people bought", instaed of just on YOU. I think this could easily be valuable, particularly since Twitter has not come close to saturation.
  • It sounds like an interest filter, and it would be great for those of us following "celebrity" Twitterers (of one kind or another).

    Great idea!
  • @Paul I'll be talking with Vid about it!
  • @Nan: If you come up with better ideas for the tagline, I'm all ears. :)

    @Toby: Unfortunately, I can't share much about the algorithms -- I'm sure you understand. :) That being said, a standard Bayes approach simply won't work too well given the type of problem we're trying to solve.
  • That's exactly why i find tweeter completely useless. I just could not care less about people's dogs and trips to the beach etc.

    Its just stupid to follow more than 20 active people as amounts of garbage are insane.

    I support your idea to give people some control over the junk they get. Who knows maybe some day (in the far future) ill even create account on tweeter through your system :- )
  • Hi!

    Sounds interesting. What kind of learning algorithm are you using? The typical bayes approach or something different?

    Regards,
    Toby
  • Oh yeah, nice FB integration, too. :)
  • After reading this I better understand Philtro. As catchy as "a spam filter for twitter" sounds, I'm not sure it's still the right tagline. Yes, I know I agreed with it on the streets of Austin. It's more of a smart filter for twitter since it shows you what you want and gets rid of the rest. Sounds cool!
blog comments powered by Disqus

Previous post:

Next post: