W-Shadow.com

In this post I’ll discuss some of the failings of existing RSS readers and suggest a few wistful and highly hypothetical improvements. The genre is moderate technobabble, the layot – web2.0 bulletpoint-y. A lot is left to your imagination to allow for personal interpretation (and other things that rhyme with “elation”). Would you like to know more?

Image by Kyle Wegner

What’s Wrong With Existing RSS Readers?

The current RSS reader applications are far from perfect. In fact, they’re little more than dumb aggregators. There are three main issues that they don’t address very well (or at all) :

• Huge amounts of information
  Manual categorization and a search function will only get you so far. It’s still easy to be overwhelmed if you subscribe to hundreds of feeds.
• Interestingness
  The user has to manually sift through mountains of data to find the salient tidbits. Most current applications don’t deal with this aspect at all.
• Timely-ness
  Some items are suitable to read at your own time, with a cup of tea in your hand. Other items, like news about a major event in your industry or a great deal at an auction site, should be seen immediately. Most existing RSS readers have no concept of “priority” or “importance”.

There’s also the whole “OMG social web!” aspect that’s generally lacking from current implementations (Google Reader being a partial exception), but I won’t deal with that here.

To The Shining Future

The web as a whole is slowly moving limping towards greater usability, but at the moment most sites are trying to take the easy way out by outsourcing “crowdsourcing” all the complex stuff (e.g. collaborative filtering and ranking). That makes sense as an initial approximation, but ultimately you can’t get very far by averaging opinions.

Lets assume we have time and money to spend on actual usability R&D, and possibly a weak AI handy. What improvements could we make to existing RSS readers?

• Use icons and colors to identify sources and represent common concepts or topics. Assign them automatically, possibly multiple icons per item. Make the glyphs degrade gracefully – use basic shapes recognizable at a glance. This will make “scanning” a list easier, not to mention how useful it will be when we all have retina displays and use HUD-type interfaces.
• Rewrite texts to use familiar words for easy comprehension and unfamiliar terms for newsness value. Determine familiar terms by examining existing communications : email, RSS, search history…
• Use machine learning to determine which items are more likely to be judged as interesting by the user. Possible criteria : how long the user spent reading the item (relative to the length of the text), whether they clicked any links, etc. Current RSS filters must be configured manually. Make them smart and adaptive instead (we already have adaptive spam filters). Reduce the time required to teach/personalize the filter by seeding it with settings used by people with similar interests. See the previous item for a hint on how to find them.
• Experiment with other modalities. Instead of plain text, try : shape, color, movement, structure (hierarchy), sound. Haptics, when they become available (sometime in the next decade most likely).
• Offer intelligent summaries, or at least personalized keyword extraction (topic detection). In the current framework, automatically categorize/label new items based on how the user has categorized existing items in the past.
• Remember that clients are part of the cloud. Are your servers too feeble to handle all of the above? Some personalization and filtering can be done offline, and that’s a good thing – you shouldn’t need to give away all of your info to online megacorps.