Standard Recommendation - An Example?

[Jeff Heidel]

I looked through what has already been said on various threads about standardRecommendation, and I just want to make sure I understand it.

An example graph:

In this example graph the 2-neighborhood of #1 is indicated by yellow nodes, and the valid nodes that standardRecommendation could return are ones in red text (7, 9, 10, 11).
standardRecommendation() would then determine which of these valid options had the lowest 2-edge sum and return the correct vertex.

Is this correct?

[Nishant George]

1) I believe one of the options we have is to allow ANY vertex in 2-nb that is not in 1-nb to be recommended... so I'd imagine that nodes 6 and 8 would be valid as well.

2) Did you mean the highest 2 edge sum? I think that's the idea.

--
Nishant George
Cornell University '11
Applied & Engineering Physics

[Jeff Heidel]

Yes, sorry, I did mean highest sum.

Isn't this algorithm supposed to be directed?
The PDF states that "The sum of all paths to v is greater than that of any other vertex which also meets the above requirements" where "we specify paths to v. The input graph is directed.".
Doesn't this mean that a directed path must exist from the node we would like to recommend to the target node, v? (this is why I didn't include nodes 6 and 8).


There is some controversy about this in other threads as well... it would be nice to have TA clarification on this subject.

[Nishant George]

Robert Escriva released a clarification on this topic via CMS. Search your email for that.

[Robert]

There was an email sent out clarifying this, I think. You can do
either directed or undirected paths. So undirected means any node a
distance of two is viable, whereas directed means only nodes with an
edge from the user to the intermediary, and an edge from the
intermediary to the possible recommendation are allowed.

On May 4, 9:57 pm, Jeff Heidel <jhei...@gmail.com> wrote:
> Yes, sorry, I did mean highest sum.
>
> Isn't this algorithm supposed to be directed?

> The PDF states that "The sum of all paths* to v* is greater than that of any

> other vertex which also meets the above requirements" where "we specify

> paths* to v*. The input graph is directed.".
> Doesn't this mean that a *directed *path must exist *from *the node we would
> like *to* recommend to the target node, v? (this is why I didn't include

[Jeff Heidel]

Found the email; that one got by me somehow.

Just a point of clarification: for the directed implementation, doesn't the clarification email state the opposite direction for edges linking the recommendee to the user (from recommendee to intermediate to user)?

And finally, is my example graph then correct as a directed example of standardRecommendation?

[Robert Escriva]

I guess it does. I'm sorry for this. I guess there are now three
acceptable strategies. You can have S->v->T, S<-v<-T, and S-v-T. Do
not mix the three. In testing, we will discover which you use and grade
appropriately.

-Robert