Hi Jacob, thanks for answer. ) Followed your idea, found plenty of BranchSelector examples at https://github.com/neo4j/graphdb/tree/master/kernel/src/main/java/org/neo4j/kernel (link you've quoted somehow didn't work). I guess modified PostorderBreadthFirstSelector method, with gatherSourceIterator() sorting its queue by node weight, will fit in my model just right. )
By the way, I'm a bit confused about how to design this. I suppose the queue sorting algorithm should be based on properties of wrapper objects, not the ones of nodes. Does it mean it's better to apply something like... ArrayList<InterestNode> Interest.sort(ArrayList<InterestNode>), ... wrapping all the nodes into objects? Or will it suffice to design some helper method in Domain class, working directly with Nodes? (sorry if it looks a bit complicated: it has to be, I've just started to use Java :) ) About processing the weights outside of traversal: yes, that's definitely a sensible option, and may be we'll just go that way after all... But what I'm trying to do is to use the laziness of Traverser: fetch only the required quantity of matching nodes (best first), not all of them, and then opt out of its cycle. Again, thank you for your suggestions! -- iD On 23 June 2011 19:38, Jacob Hansson <[email protected]> wrote: > The order method indeed does decide what order the graph is traversed, you > can either use it with the predefined traversal orders like so: > > td.order(CommonBranchOrdering.PREORDER_DEPTH_FIRST) > > or you can pass it your own branch selector implementation, see > > https://github.com/neo4j/community/blob/master/kernel/src/main/java/org/neo4j/kernel/PreorderDepthFirstSelector.javafor > an example. > > An easier way to include weights would be to make the traversal only follow > LIKES relationships that are over a certain weight threshold. You can also process the weights outside of the traversal entirely. For > instance, you can calculate the average weight for each path returned by > the > traversal, sort the result by that, and only return the top results. > /jake > > On Thu, Jun 23, 2011 at 8:30 AM, Igor Dovgiy <[email protected]> wrote: > > > Hi all, > > > > I'm trying to implement a recommendations system app with Neo4j (1.4). > > Basically, I intend to have some 'User' nodes and some 'Interest' nodes > in > > my graph. Then, to find users with similar interests, traverse it by > > something like... > > > > TraversalDescription td = new Traversal.description() > > .relationships(RelTypes.LIKES) > > .evaluator(Evaluators.atDepth(2)); > > > > The real question is, though, whether I'll be able to choose what > Interest > > nodes to walk through first, depending on their weights (i.e., number of > > users linked to them)? As I understand, the order() method is designed > > right > > for this case, but sadly I was unable to find examples of its usage... > > > > Any help will be greatly appreciated! > > > > -- iD > > _______________________________________________ > > Neo4j mailing list > > [email protected] > > https://lists.neo4j.org/mailman/listinfo/user > > > > > > -- > Jacob Hansson > Phone: +46 (0) 763503395 > Twitter: @jakewins > _______________________________________________ > Neo4j mailing list > [email protected] > https://lists.neo4j.org/mailman/listinfo/user > -- -- iD _______________________________________________ Neo4j mailing list [email protected] https://lists.neo4j.org/mailman/listinfo/user
