0

After performing the SVD of a term-document matrix, and getting a reduced rank matrix, various sources have stated the following reduced query vector formula. It seems easy to see how its derived.

However, in this link, the query vector is calculated as centroid of the corresponding reduced term vectors. I tried to see if the two were the same, but the results were different.

What is the difference between the two and what are the pros/cons of using either?

Aarav Pandya
  • 15
  • 3
  • I'm not an expert on LSA, but based on what you said above, why don't you pick the first option? If it is used by "various sources", it seems there is more support for that one. The link you included is just one source and it seems to be a tutorial question for a course. Of course, if you're enrolled in that course, I suggest you ask the professor. If you're not and you just came across it with Google, it seems to make more sense to stick to the one with the "various sources". – Ray Nov 19 '18 at 14:48
  • Suppose you use the second option ... what would you cite? Another university's tutorial exercise and a link to this stackoverflow question? You should consider the possibility that those exercises should be seen in the context of the course -- perhaps in the lecture, the professor said s/he made some simplification for the students. I'm not so sure you should take it too seriously as long as your "various sources" are reputable. (I assume they are research publications?) – Ray Nov 19 '18 at 14:50
  • Yes one of the sources is the irbook by Christopher Manning. That is what i am going to do. I am curious since almost everyone follows the prior formula, but this particular tutorial uses something much simpler. I am not enrolled in this course and i dont have any direct means of communication with the author. I certainly will try to contact him if i dont get a good enough answer. – Aarav Pandya Nov 19 '18 at 15:05
  • In my humble opinion, I'd stick with what is published, either in books or in research publications. Tutorials are sometimes simplified to make it similar to other topics in the class -- a lecturer has the freedom to make such a decision and not follow the field in order to get the message across to his/her students. Likewise, if you're not enrolled in the course, I'm not so sure if the lecturer feels any obligation to help you. Good luck! (PS: You might want to see if the lecturer has published in the area; it might be based on prior work of his/her's.) – Ray Nov 20 '18 at 04:46

0 Answers0