Questions about last year quiz 4

Questions about last year quiz 4

by Deleted user -
Number of replies: 1

Hello,

After trying to finish the last year quiz 4 and comparing with the corrections, I have several questions that I don't really understand:

Question 8.1 : the question asks which the better system is where it is important to find all the documents that are relevant. In this case I think we should look at the "Recall" score. However since the two systems have the exactly same recall score, why the answer is S2?

Question 9.2 : in this question, I find out the whole method looks like k-means approach for a clustering problem. So I thought it was an unsupervised classification method. However in the correction, it is said to be a hybrid method. Why is it hybrid in this case?


Thank you in advance for your clarification,

Qi

In reply to Deleted user

Re: Questions about last year quiz 4

by Jean-Cédric Chappelier -

Question 8.1: indeed Recall is here important, so you look for the better system at high recall, which clearly is S2 (compare for instance precision scores for Recall 0.6; or look at this picture :

P-R curve for S1 and S2

Saying that "the two systems have same recall" is not really meaningfull as recall is increased incrementally on any system evaluated in order to plot the Precision-Recall curve...

Question 9.2: sure steps 4 to 6 look like K-means and are unsupervised, but steps 1 to 3 are indeed supervised (definition of topic t), thus the whole is hybrid.