Hi, You can store properties on Relationships, just as you can on nodes. Where in the documentation does it say that you cannot do that? Because that is just plain wrong, it's always been possible. If documentation somewhere says that it isn't, that needs to be corrected with top priority.
Or do I completely misunderstand your question? Storing the Markov chain in Neo4j is not a problem[1], the problem you get into (that I think you've identified) is with *using* the Markov chain. What you want to do is choose a branch based on the probability that the word on the destination node follows the word in the source node. You could do this either by storing the links in a list with the number of occurrences of a link corresponding to it's probability or something "smarter". While you can have multiple relationships (of the same or different types) between the same two nodes in Neo4j, this gets you into problem since traversing the relationships is a linear operation, and you are only after One relationship (but chosen at random). What I did was to store the total number of *virtual * outgoing links on the source node and then the count of the number of occurrences stored on each relationship. I would use the number at the start node to generate a random number, then iterate through the relationships increasing a counter by the weight of that relationship in each step until that counter was greater than or equal to the random number I had generated, I would then return that relationship. Just my 2c, since I've worked on a similar problem before ;) Moving on to answering Oliver Rossels question about reification: No. Reification of relationships is not possible in Neo4j, this would induce a penalty in performance and isn't something you need in that many cases, which is why we have opted to not support it. You can however emulate it by introducing an intermediate node that you can draw the reificating relationship from, like so: NODE_ONE --REL_PART_ONE--> INTERMEDIATE --REL_PART_TWO--> NODE_TWO Where the two relationships could have either the same type, or different types depending on how your model is structured. The performance from this would be pretty good actually since the INTERMEDIATE node would only have one incoming relationship of it's main relationship type and one outgoing. This is what we do in the RDF layers on top of Neo4j. Happy Hacking, /Tobias [1] I've done similar toy examples for a web site I made that is a parody on Swedish news sites and generate fake "news" with the same word probability as Swedish news, by using the feeds from these news sites as training for a Markov chain. -- Tobias Ivarsson <[email protected]> Hacker, Neo Technology www.neotechnology.com Cellphone: +46 706 534857 _______________________________________________ Neo mailing list [email protected] https://lists.neo4j.org/mailman/listinfo/user
