I am new to topic modelling, so kindly bear me if my question is silly.
I am trying to calculate perplexity after applying LSA. i am aware that LSA returns negative values, so i followed the steps stated in coccaro to find the propability of each word in the vocabulary(which i am still not clear whether must i calculate for the whole vocabulary or word by word in the given test document).
Now how do i calculate perplexity?
How must i find the document propability ? must i find the sum of propabilities of individual words in the document and call it as document propability. what if the words are repeated.
I found the following stackoverflow question which talks about PLSA.From that , perplexity is given as 2^−∑(i=1 to N)(1/N)log2q(xi)
How can i estimate log q(xi) in that equation for LSA.