twitter: a sketch for filtering signal from noise

John Battelle writes today in Twitter and the Ultimate Algorithm: Signal Over Noise (With Major Business Model Implications) that twitter might have the opportunity to make a giant leap forward as a social utility if it were only to apply algorithms to the feed. I concur. Twitter is indeed noisy — to a degree, that many of its users are suffering “feed fatigue.”

So let’s take on the challenges twitter would have to conquer in the design of truly effective algorithmic feed filters.

First, twitter “content” is not just information, or content pure and simple. It’s also social activity. So, multiple algorithms would be required to extract relevance from tweets.

It strikes me that these algorithms might fall into two broad categories: content relevance and social relevance. Content is relevant on twitter generically, as trending news and information, often cited, agreed with, and linked to.

Generic news is social news on twitter — a form of signal having no individual relevance other than that it is news.

Individually relevant news is news that I am interested in, given standing thematic interests. (I am interested in climate change. When a climate change story gains traction on twitter, it has both generic news relevance (because it is trending), it has vertical news interest, and is personally interesting to me. This last factor may be gleaned from the user’s past tweet topics, links, hashtags, and accounts followed.

  • Trending news: measured by appearance of news in tweets, and linked to as references, across multiple social circles. In other words, having the status of “general news interest.”
  • Vertical news: measured by repetition of news tweets and linked references but qualified by retweeting and @naming activity associated with a social circle.
  • Social circles: identified by persistent social referencing in @naming, multiple @naming, retweeting activity and distinguished by a baseline volume of personal tweets, etiquette and gestural tweets, and confirmed by twitter social graph mutual follows. (Members of the circle provide each other with sustained attention. A system might store data on persistent membership in these local or social graphs.)
  • Strong tie conversation: tweets “exchanged” between users who have, historically, a habit of addressing and replying to one another. These tweets manifest a relationship, online/offline it doesn’t matter, and can be characterized as ongoing open state of talk between two or more users.
  • Gestural interpersonal actions: tweets that are designed to solicit a response in the form of an @reply, an @name cite, follow back, listing, or hashtag participation (eg #FF). These are gestural tweets in which the semantic claims of the tweet are subordinate to the interpersonal display and soliciting or reciprocal attention and recognition.
  • Interpersonal connections: these tweets indicate initiating and cementing a follow connection on twitter, and may be identified by their use of etiquette confirmed by mutual follows and even listing.
  • Social gestures: these are tweets that appeal for acknowledgment and reciprocity by members of a perceived social circle, as when a tweet @names several members of an existing social circle (this could be looked up using the social circle algorithm). Such tweets might indicate relevance of a social circle’s activity — as happens when vertical news topics gain traction, and followers wish to recognize it and possibly become included.
  • Individual self-talk: tweeting activity whose semantic references are personal, which does not @name or address itself to others in particular, and sometimes making use of common “cultural” hashtags and expressive idiomatic hashtags.

These are some, and certainly not all, of the distinctions that algorithmic filtering of tweets might attempt to capture. Meaning, signal, or relevance on twitter necessarily refers to both content and social interaction, for twitter is a tool that serves both the publishing and distribution of content, and the maintenance of social talk among connected users.

I did not address influence here, which is where Klout continues to mine for relevance. Influence would be relevant to signal in suggesting and recommending users and topics. Some of the distinctions above would apply.

As far as advertising goes, the benefits of advertising into the feed increase if user tweeting styles and habits are taken into account. Here, a static or standing “influence” metric (such as a Klout score) would want to include a user’s audience responsivity. How likely are tweets and retweets to be distributed further, and by what number of influencers. This is because user activity alone is not a measure of a user’s value for commercial (or any) content distribution. Audience responsivity confirms a user’s potential value for this. Social capital is not an attribute — it is a relation (user to audience).

Related: Sharepocalypse, and why sharing is noisy

This entry was posted in Marketing, Twitter. Bookmark the permalink.
  • http://equinejointsupplements.blogspot.com mauricio

    The post is very Useful. Thank you, this advice will come in handy.

  • http://www.technologyactivist.com/ Communication Technology

    Great Post! Considerably, the article is actually the freshest on this notable topic. I agree with your conclusions and will thirstily look forward to your next updates.I will immediately grab your rss feed to stay abreast of any updates.

  • Pingback: Filtering Out the Hashtag Economy | Geoff Livingston's Blog