Hi Aman, This is the classic friends-of-friends problem recast as friends-of-places. The hard thing about places (as opposed to most normal friends) is that they tend to be popular. While I might only reach hundreds of friends-of-friends (if I'm lucky!), being recommended people that have visited London will run into millions - hard to process efficiently and sadly prone to giving poor recommendations.
The question I'd ask there is: is it meaningful to be recommended friends solely based on places they've visited?There are thousands of people visiting London right now and it would be weird if they were recommended to me as friends, and not so convenient either since they live in other places. So, wouldn't it be better to include more sophisticated signals other than just "shared place"? For example, why not recommend me friends who have visited the same cities as me, and who live in my city. That way we can get together and reminisce about those wonderful places we've visited. Or recommend friends of friends that have visited certain cities that I have also. That way you're not only improving the likelihood of giving good recommendations, but you're narrowing the data set on which you're operating - win/win. I think you could elegantly solve this with pattern matching or Cypher (which is build on pattern matching). See: https://github.com/jimwebber/neo4j-tutorial/blob/master/src/koan/java/org/neo4j/tutorial/Koan08.java For an example of how we pattern match in the Doctor Who universe. Jim _______________________________________________ Neo4j mailing list User@lists.neo4j.org https://lists.neo4j.org/mailman/listinfo/user
