Rene,
you mean the fork that Mattias has been providing you with for handling
millions of relationship types? It is possible to take that into master,
however, I think Johan said we need to test and make sure all cases are
covered. Don't have an ETA on that ATM. I think you could raise an issue
and
@ qcon: I can't find the twitter videos on
http://qconlondon.com/london-2011/ rick could you please point to further
rescources?
@ aseem: I gave a detailed answer to your questions on
http://www.rene-pickhardt.de/graphity-an-efficient-graph-model-for-retrieving-the-top-k-news-feeds-for-users-in-so
A couple of things bothered me about Rene's approach.
1. Custom Neo4j that could handle lots of relationship types. Not 100% sure
where this fits?
2. Performance drops like a bomb when K (the number of items retrieved)
increases.
So it kind of works to get the top 15 items... A twitter stream co
Subject: Re: [Neo4j] Activity Streams and Twitter Sample App
This is very interesting -- thanks Peter for the link, and thanks
maxdemarzi for starting this conversation.
In our social network -- which has extremely little load, it's just in beta
-- we currently use basically (a), and it works
This is very interesting -- thanks Peter for the link, and thanks
maxdemarzi for starting this conversation.
In our social network -- which has extremely little load, it's just in beta
-- we currently use basically (a), and it works just fine. We use Cypher to
do the sorting/trimming on the server
You might even be interested in Rene Pickards work on a full solution
(albeit with some write-time overhead), see
http://www.rene-pickhardt.de/graphity-an-efficient-graph-model-for-retrieving-the-top-k-news-feeds-for-users-in-social-networks/
Cheers,
/peter neubauer
GTalk: neubauer.peter
Sk
I had not considered imperfect solutions, and in some activity stream
scenarios a sampling of the last few messages could work. The sample would
have to be taken from all Person nodes because if we sample from the Tweets
in general and we encounter a "chatty" person node early on, it would take
up
one more solution. set a sampling_ratio, say 10:
Person1.outgoing(:follows).outgoing(:tweeted).depth(2).filter("position.length()==
2;") .prune("position.returnedNodesCount() > 100 * sampling_ratio")
then do a sort based on timestamp.
the goal is not to get the perfect result but *good enough* ones
Came up with another possibility:
G) Store the Latest 100 tweeted relationship ids with dates as a property of
the User Node, and a custom Breadth First Traversal that evaluates the list
of every follower's latest 100 tweets before deciding which relationships to
follow.
--
View this message in c
Andreas Ronge created a new sample app called kvitter @
https://github.com/andreasronge/kvitter .
This got me thinking about the Twitter clone done in Redis @
http://redis.io/topics/twitter-clone
If you scroll down 2/3's of the way down you'll read this piece:
"After we create a post we obtain t
10 matches
Mail list logo
