Neo4j 的参照完整性

Question

I am working on a project that uses a graph database to hold click data for a search engine. The nodes can be search terms or urls, and the edges hold a weight attribute, and a percentage of times that search led to someone clicking that URL.

Number of times the URL was clicked / Number of times term was searched

My issue is that when I update the edges, the percentage will be accurate, but if I later update the search term node and the searched count changes, the edge will no longer have the correct percentage. Is there a way in Neo4j to keep referential integrity? like a foreign key type thing?

Answer 1

Unfortunately no, neo4j doesn't support this. You can still do it, with one of two methods. I'll tell you what they both are, then make a recommendation.

Relative to your relational database, I don't think you're looking for a foreign key or "referential integrity" -- I think what you're looking for is more like a trigger. A trigger is like a function or procedure that executes when data changes. In your case, it'd probably be good to have trigger functions that re-calculated all of the weight percentages on incident edges.

Option 1 - The capable Max De Marzi has got you covered there with a description of how you can do triggers in neo4j. Spoiling the surprise, there's a TransactionEventHandler in the java API. When the right kind of transaction comes through, you can catch that and do extra stuff.

Option 2 - the server provides an extension/plugin mechanism so that you could write this on your own. This is a big hammer, it can do just about anything, but it's harder to wield, too.

I'd recommend you look into Max's post and the TransactionEventHandler . You might then implement public void afterCommit(TransactionData transactionData, Object o) . In that method, you'd check out the transaction data to see if it was something of interest (not all transactions would be of interest). If the transaction updated a search term node or searched count changes, then I'd go do your recomputation, fix your weights, and you should be good.

Answer 2

The following info might be helpful.

If you stored the number of clicks instead of the percentage, there is no way to get inconsistent data. For example:

(:Term {id: 1, nSearches: 123})-[:HAS_URL {weight: 2, nClicks: 17}]->(:Url {id: 2})

With this data model, you'd calculate the percentage whenever you needed it.

For example, to find the 10 terms that have the highest percentage of visits to a specific URL:

MATCH (term:Term)-[r:HAS_URL]->(url:Url {id: 2})
RETURN url, term
ORDER BY r.nClicks/term.nSearches DESC
LIMIT 10;

But notice that the inverse query (find the 10 URLs that have the highest percentage of visits from a specific term) does not even require that you calculate the percentage! This is because in this case the percentages all have the same denominator. So, you can just use nClicks for sorting:

MATCH (term:Term {id: 1})-[r:HAS_URL]->(url:Url)
RETURN term, url
ORDER BY r.nClicks DESC
LIMIT 10;