Figure 1 is an accurate representation of the data. The goal is to, starting from a random document, take a random number of steps to another document. The navigation of the path would be weighted by how many inbound edges the metadata object in question has, as well as the weight of the individual edge between the document and the metadata object.
Essentially in your diagram, if I start at "document 4", and I take a "random" walk to a metadata object, I will of course go to "metadata 2" (the only link possible from document 4). Then, say the weight of the (document 3)->(metadata 2) edge is 0.0001, and the weight of the (document 2)->(metadata 2) edge is 0.721. In most cases I want to then navigate to hop to document 2, as it has a much higher edge weight - that's not to say that in 30,000 iterations I wouldn't occasionally hop to document 3, though! We then generate a random number to see if we continue walking. Say we do. Now that we're at document 2, we start the process over again. BUT, document 2 has two metadata properties. Metadata 1 has two edges associated with it, and metadata 2 has three edges associated with it. I want to weight the decision to move to the random metadata node to be weighted by the number of edges connecting to that metadata node. In this case, I want to generally weight LESS connected nodes more highly - so in this situation we'd slightly prefer hopping to metadata 1 instead of metadata 2. From metadata 1, we end up on document 1. We then generate a random number to see if we continue walking. Say we don't. Our weighted random walk might have moved from document 4 -> metadata 2 -> document 2 -> metadata 1 -> document 1 In the real graph the number of documents visited could be anywhere from 1->all of them, in theory Hopefully this explained what I'm trying to do a little better! :) If this is too complicated to do as a canned function in orient I think that's totally reasonable, someone else is working on writing a more direct Java interface which may give us somewhat better control over all this. Thanks, Josh On Wednesday, August 12, 2015 at 1:44:23 AM UTC-7, [email protected] wrote: > > Hi Josh, > I did not understand your first question, can you explain what you want to > achieve ? > What you mean by "centrality of the metadata" ? > Your structure is like that in Figure 1 or such as that in Figure 2, or > neither ? > > <https://lh3.googleusercontent.com/-MlpzxEFP5gU/VcsGwpzb2gI/AAAAAAAAAEY/OAFHx7BYuXA/s1600/Image1.png> > > <https://lh3.googleusercontent.com/-qHsB19V-5-s/VcsG8yrGRDI/AAAAAAAAAEg/1BJuZqz-b9Y/s1600/Image2.png> > Alessandro > -- --- You received this message because you are subscribed to the Google Groups "OrientDB" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
