Neo4j is neat and underutilized. I think rglullis is just saying, if that's why espeed used it, he could have saved three paragraphs of explanation by just saying that.
I find elegance in simplicity. If you don't need the tabular features of a relational database, why not use a graph DB for relational data? It's like a key value store with directly connected relationships.
This is so much simpler than having to deal with creating tables, DDLs, and migrations:
>>> g = Graph()
>>> james = g.vertices.create(name="James")
>>> julie = g.vertices.create(name="Julie")
>>> g.edges.create(james, "knows", julie)
Futhermore, there's no impedance mismatch so your code is cleaner.But regardless, this is Hacker News where people explore, build, and share new things. I hope we're not moving to a place where that gets admonished.
* Nodes/Vertices: Person, Topic, Entry
* Relationships/Edges: Tagged, Authored
person -- authored --> entry
entry -- tagged --> topic
Here's the base model: https://github.com/espeed/lightbulb/blob/master/lightbulb/mo...The comment system is a separate, generic component.
Marko, the guy who created Gremlin (https://github.com/tinkerpop/gremlin/wiki), just released a Gremlin Tree step (https://github.com/tinkerpop/gremlin/wiki/Tree-Pattern), which makes it really easy to build a threaded comment tree in one quick shot.
If you have a large blog or content system, you can use likes and views to make content recommendations using a basic Gremlin collaborative filtering query:
// calculate basic collaborative filtering for vertex with user_id
def rank_items(user_id) {
m = [:]
g.v(user_id).out('likes').in('likes').out('likes').groupCount(m)
m.sort{a,b -> a.value <=> b.value}
return m.values()
}
If you use Facebook, Twitter, or GitHub to authenticate users you can easily incorporate friends and followers into the recommendations.