pooled sequences cut

Questions, answers, discussions related to VideoLectures.Net Recommender System Challenge

pooled sequences cut

Postby Zenek » Fri May 06, 2011 4:37 pm

hello!

from description about how pooled sequences are created I understand that for counting frequencies there are taken in consideration only lectures after cut, from example

clicking stream
id1 -> id7 -> id2 -> id4 -> id5 -> id6 -> id3

triplet and counted friequencies
{id1, id4, id5} -> id6 : 1, id3 : 1


why there aren't taken all lectures that aren't in triplet?

if there will be triplet {id1, id7, id3}, there will be no frequencies counted? because we determine all unordered triplets, thanks!
Zenek
 
Posts: 3
Joined: Fri Apr 22, 2011 12:43 pm

Re: pooled sequences cut

Postby ninoaf2 » Mon May 09, 2011 3:30 pm

Dear contestant,

Yes, we only account lectures that are from the right hand side to particular triplet.
Once the triplet is formed, only lectures/IDs appearing after the lectures from that triplet, in the original viewing sequences, are taken into account when forming the right side of the pooled viewing sequence.
In this way we want to determine which lectures should be recommended after some particular
triplet has been viewed.

Best regards,
Organizers
ninoaf2
 
Posts: 25
Joined: Tue Mar 08, 2011 2:53 pm

Re: pooled sequences cut

Postby Zenek » Mon May 09, 2011 10:07 pm

thanks for reply, so the order of triplet does matter, and for above example with clicking stream id1 -> id7 -> id2 -> id4 -> id5 -> id6 -> id3 there can not be a triplet {id5, id1, id7}, because id5 is older in stream than others ids from triplet?
Zenek
 
Posts: 3
Joined: Fri Apr 22, 2011 12:43 pm

Re: pooled sequences cut

Postby ninoaf2 » Tue May 10, 2011 10:56 am

Dear contestant,

No, the order of lectures in the triplet (left side of the pooled viewing sequence) does not matter.
Order on the right matters and is obtained by scoring.
Pooled viewing sequences are constructs that only (statistically) approximate real viewing sequences.
Again - here is the process:
By cutting at some point all real viewing sequences - we get their left sides and right sides. Now choose some triplet of lectures
and count all their joint appearances on the left side of all viewing sequences (their order on the left side does not matter!).
This gives the frequency of the triplet. At the same time we count the appearance of all other lectures on the right side of the same
sequences in which triplet occurs. These gives us counts and then also the order of lectures on the right side of the pooled viewing sequence.
We keep only ten most frequently co-occurred lectures on the right side for the particular triplet (i.e. pooled viewing sequence).

If you have any more questions, feel free to ask.

Best regards,
Organizers
ninoaf2
 
Posts: 25
Joined: Tue Mar 08, 2011 2:53 pm

Re: pooled sequences cut

Postby Zenek » Tue May 17, 2011 11:20 pm

in evalutaion error measure MARp the lectures are ordered by score, what happend if all lectures have the same score?
Zenek
 
Posts: 3
Joined: Fri Apr 22, 2011 12:43 pm

Re: pooled sequences cut

Postby ninoaf2 » Thu May 19, 2011 11:22 am

Dear contestant,

"A special situation happens when there are more equally relevant items at the same rank (ties) at the cut-off length of the s list. In that case, any of these items is treated as relevant (true positive) in calculating Rp@z(r). " [From VL.Net Challenge site]

Best regards,
Organizers
ninoaf2
 
Posts: 25
Joined: Tue Mar 08, 2011 2:53 pm


Return to VideoLectures.Net Recommender System Challenge

Who is online

Users browsing this forum: No registered users and 1 guest

cron