Amazon, Pandora, TiVo, iTunes, Last.fm, Google Reader, the list goes on. Personalized recomendation services are a hit and for good reason. There's so much utter rubbish out there to be ploughed through it's just too much for even the most obsessive datarats among us to handle on our own. We need machine assistance to help us sift and sort to find those tasty nuggets we crave so badly.
But the trend has been to assume that we already know what we like and that we should simply have more like it. The much discussed APML is evidence that this line of thinking is still mainstream. The only way to adjust the settings of the suggestion engines in any of these systems is to allow them to profile you as you interact with them. But what if what you're interested in today doesn't directly relate to what you've been paying attention to in the past? How can you still make use of that attention information while actively directing future suggestions?
This sounds like a job for a Real World Scenario!
I'd never heard of bluegrass music until I was nearly done highschool. So I hit the net, scrounged (napster'd at the time) up some MP3's and had a listen. It was terrible. I discovered that I pretty much hate bluegrass. But let's say something like Last.fm was around then and it already had a nice APML like profile of what I had been listening to. If I were able to manipulate my attention profile by editing key concepts as tags and their attention values as sliders I could add a bluegrass tag, slide it up to maybe 50% and let the suggestion engine do it's job.
The end result would be somewhere between a saved search (or smart folder / playlist) and a profiled suggestion. I get all the benefits of cross referencing my recorded tastes with those of the Last.fm masses but with some added direction over where it takes me. Maybe if I'd listened to the kind of bluegrass tunes that other Tea Party, Weezer, and Cranberries fans had been listening to I might have loved it ...
Nah.
Now where'd I really love to see this functionality is in my news reader. Let's say I get it in my head that I want to read more about the 2008 Olympics. I haven't been reading many items in my news reader related to the topic, but if I were looking for feeds on the subject I'd likely be more interested in stories that relate to both the Olympics and Canada, which does come up often in my news reader. Those feeds that cross over could then be suggested first. Guided+Profiled personalized suggestions.
Finally once a facility was in place for something like this, I could see much use in creating "APML bookmarks" for collections of feeds (OPML). In this way I could have all my feeds together, but browse them under say the "Politics", or "Tech" attention profiles as needed. All the while each "bookmark" would grow on its own as it profiled me, but I could also be directing it as I went along.
Do you hear me intertoobs!? I need more control over where I'm spending my paltry attention budget.
But the trend has been to assume that we already know what we like and that we should simply have more like it. The much discussed APML is evidence that this line of thinking is still mainstream. The only way to adjust the settings of the suggestion engines in any of these systems is to allow them to profile you as you interact with them. But what if what you're interested in today doesn't directly relate to what you've been paying attention to in the past? How can you still make use of that attention information while actively directing future suggestions?
This sounds like a job for a Real World Scenario!
I'd never heard of bluegrass music until I was nearly done highschool. So I hit the net, scrounged (napster'd at the time) up some MP3's and had a listen. It was terrible. I discovered that I pretty much hate bluegrass. But let's say something like Last.fm was around then and it already had a nice APML like profile of what I had been listening to. If I were able to manipulate my attention profile by editing key concepts as tags and their attention values as sliders I could add a bluegrass tag, slide it up to maybe 50% and let the suggestion engine do it's job.
The end result would be somewhere between a saved search (or smart folder / playlist) and a profiled suggestion. I get all the benefits of cross referencing my recorded tastes with those of the Last.fm masses but with some added direction over where it takes me. Maybe if I'd listened to the kind of bluegrass tunes that other Tea Party, Weezer, and Cranberries fans had been listening to I might have loved it ...
Nah.
Now where'd I really love to see this functionality is in my news reader. Let's say I get it in my head that I want to read more about the 2008 Olympics. I haven't been reading many items in my news reader related to the topic, but if I were looking for feeds on the subject I'd likely be more interested in stories that relate to both the Olympics and Canada, which does come up often in my news reader. Those feeds that cross over could then be suggested first. Guided+Profiled personalized suggestions.
Finally once a facility was in place for something like this, I could see much use in creating "APML bookmarks" for collections of feeds (OPML). In this way I could have all my feeds together, but browse them under say the "Politics", or "Tech" attention profiles as needed. All the while each "bookmark" would grow on its own as it profiled me, but I could also be directing it as I went along.
Do you hear me intertoobs!? I need more control over where I'm spending my paltry attention budget.
3 comments:
The problem here is that the vast majority of web 2.0 cruft is geared at and written by hapless muppets. The biggest problem with things like most conventional RSS readers is that they want to be special and differentiate themselves from your conventional reading workload (ie, email). News clients were in this same camp years ago, until Netscape accidentally stumbled across one good idea and decided to integrate the two (synergy, bitches). Whether you're using NNTP or RSS, it doesn't make any sense to special-case the workload because you happen to be pulling it down via a different protocol. There's ultimately nothing new or innovative about half of these things.
Years ago when RSS was first starting to rear its ugly head it was obvious that there were two options. 1) Become a slave to GUI brain-damage. 2) Add RSS support to mutt. Obviously 1) wasn't very realistic, so that left 2) as the only viable option. What I ended up doing then, and what I still regularly do now is to just pull down the feeds I care about via a cron job, and then convert that on the fly to Maildir format. Regular notifications (ie, biff) work just fine with this, you can tag things you care about, bounce them along for people to look at (forwarding takes care of constructing a saner header than what is needed to trick the MUA), and most importantly, _filter_ based on keywords you care about. Anyone on a high volume mailing list (ie, l-k) is already used to filtering stuff out based on areas of interest, and neither RSS nor NNTP are particularly special in this regard. It's utterly mind-boggling that Gmail incorporates sane filtering to this extent, whilst Google reader is utterly devoid of anything even leaning towards usefulness. The tool I wrote for this was written in about 10 minutes 5 or 6 years ago, so it's not a difficult problem, even if userspace is a scary place to tread.
Now if only the people that wrote XML parsing libraries weren't the same idiots responsible for much of the current web 2.0 abortion. To paraphrase jwz, "Some people, when confronted with a problem, think 'I know, I'll use XML.' Now they have two problems." Ironically the original quote was aimed at regular expressions, which end up being a much cleaner way to do parsing than things like libxml/libxml2 and its spawn.
I've looked at a handful of desktop readers but it's just not useful for my needs. I use so many different computers in my travels I need an online solution. And no, tunneling home to grab a screen session isn't an option due to assclown firewalls at the cube.
Is there a web interface for mutt I could use on the road?
There's no reason to make things complicated. I usually use a combination of a local postfix server and offlineimap for synchronization, then just work with anything I've mangled in to Maildir format via replication for offline use. This tends to be the easiest solution, since no matter where I write mail or fetch updates, it's auto-replicated as soon as I have some form of connectivity. Admittedly I do have a centralized SMTP server where all of my decentralized postfix instances relay through, but things can happily queue up whilst offline.
With regards to firewalls, corkscrew works wonders. If you require a web frontend and don't wish to tunnel, then there are things like squirrelmail and so on that you could also use, though I certainly don't recommend it. An alternate option is simply to have gmail do your IMAP synchronization for you so you can sidestep the need for direct port access directly from your client side.
Post a Comment