Computing precision and recall of LOCABAL about qrec

[Coder-Yu]

hi, this is mainly because LOCABAL is a method designed for the task of rating prediction. As a consequence, it is less likely to obtain a decent result on item recommendation. Empirically, if you really want to conduct item recommendation tasks with LOCABAL, I suggest that you should set a small iteration number, e.g. 3-5. However, I personally think that you should use methods designed for item recommendation such as BPR to compute precision, recall, etc.

[noemi-mauro]

Thank you for your answer.
I was looking at the code of locabal and I found a commented else.
if self.W.has_key(user):
self.P[u] += self.lRate * (self.W[user]*error * q - self.regU * p)
self.Q[i] += self.lRate * (self.W[user]*error * p - self.regI * q)
#else:
self.P[u] += self.lRate * (error * q - self.regU * p)
self.Q[i] += self.lRate * (error * p - self.regI * q)

I think that should not be commented or we update P and Q two times. What do you think about this?

[Coder-Yu]

sorry, I cannot remember the details of this paper. If you think you are right, just go ahead.