Announce: News Reader with Lisp Based AI backend

[darkglyph]

Our News Reader, Storylens is now active. It uses a natural language processing component along with a learning system to personalize your news. All of the AI aspects were written in Common Lisp.

We have an HTML 5 web application, a native iPhone and iPad application along with a desktop app that runs on Mac, Linux, and Windows. We also have a mobile web app for non iOS smartphones.

For more information or to sign up, just visit storylens.com or download the apps from the App Store.

[Zach Beane]

Very interesting! Is there any information available about what tools
you used? Do you have anything written about your experience using
Common Lisp? Do you intend to share any CL software or libraries as a
result of the work you did for the commercial product?

Zach

[Pascal Costanza]

I don't understand from the description what this Storylens is actually
going to do, and I always hesitate to sign up for a service where I
don't know upfront what I can expect from it.

What exactly does "personalizing" mean? Does it filter news that I'm
likely not interested in? Does it get news for me without me asking to
subscribe to any RSS feed? What set of RSS feeds does it start with,
when it knows nothing about me yet?


Pascal

--
My website: http://p-cos.net
Common Lisp Document Repository: http://cdr.eurolisp.org
Closer to MOP & ContextL: http://common-lisp.net/project/closer/
The views expressed are my own, and not those of my employer.

[kenanb]

Testing now. first feedback: the shortcuts at the right and left side
of the window should have some alt-text/description, since if a user
skipped the guide quickly, he/she is left with no idea what those
buttons do (except the javascript:foo() links that appear in statusbar
when you mouse-over, but they are obviously not the best way to get
this information). search function should definitely start searching
when I hit RET, trying to RET and realizing it won't search unless I
push the search button every time disinclines me. otherwise I loved
the interface :)

Also I don't know if this is already implemented, but it should also
learn my interest from my search keywords, since even when it has zero
results matching my search or i don't find one that i am particularly
interested in among the search results, search keywords are great
hints.

[darkglyph]

Thanks very much for the feedback! I'll let the UI guys know about the
search facility issues and I'll see what I can do about getting those
searches as keyword training for the AI.

One item that we implemented in the desktop client but never made it
into the web client is a tag cloud. Basically it looks for topics in
all of your content and displays that as a tag cloud. When you click
on a link, it will do a search on that topic. I'll have to check to
see if that generates a training set.

Thanks again for your comments. Feel free to keep them coming. We love
the feedback. You can also just send me an email.

[darkglyph]

The system comes with a set of pre-determined feeds and these are
placed into a series of sections (we call these mixes). You can also
add your own feeds and create your own mixes. As you interact with
stories, the system reviews these and uses these to train a model. So
every user gets a specific model. We then use this model to score new
stories as we fetch them. We keep a queue of new stories for every
user. When you request a Mix via one of our UI's we return the stories
from your queue in the order you requested, either by score (our
predicted interest) or by popularity or date and time.

When you first sign up you get everything in date and time order until
we have some learning. As soon as we have some training we then start
using the model to do our predictions.

Thanks very much for your questions. FYI, We got started because we
were tired of sifting through a lot of stuff in most readers that we
were not interested in and wanted an automated system that would do
that for us. We are definitely looking for feedback and ways that we
can make the system better.

[nmancer]

On Oct 26, 9:08 am, darkglyph <darkgly...@gmail.com> wrote:
> On Oct 25, 4:54 pm, Pascal Costanza <p...@p-cos.net> wrote:
>
>
>
>
>
>
>
>
>
> > On 24/10/2011 18:39, darkglyph wrote:
>

> > > Our News Reader,Storylensis now active. It uses a natural language

> > > processing component along with a learning system to personalize your
> > > news. All of the AI aspects were written in Common Lisp.
>
> > > We have an HTML 5 web application, a native iPhone and iPad
> > > application along with a desktop app that runs on Mac, Linux, and
> > > Windows. We also have a mobile web app for non iOS smartphones.
>

> > > For more information or to sign up, just visitstorylens.com or

> > > download the apps from the App Store.
>

> > I don't understand from the description what thisStorylensis actually

From the web app there are tools to kick start and maintain your
mixes. We have an OPML import facility, we can also directly import
your Google reader feeds (though we currently don't maintain a sync
facility with the Google reader). There is also bookmarklets for
several browsers. These are available via the advanced tab:http://
www.storylens.com/advanced.html