-
Notifications
You must be signed in to change notification settings - Fork 0
/
topic-modeling.txt
28 lines (18 loc) · 1013 Bytes
/
topic-modeling.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
A Document is a mixture of topics.
We know the corpus and the number of topics. But, we don't have idea about the actual topics and its distribution in the document.
Topic modeling -
-> Text clustering problem, where documents and words are clustered simultaneously.
-> Approaches -
PLSA (Probabilistic Latent Semantic Analysis)
LDA (Latent Dirichlet Allocation) (Better and Popular)
Generative models and LDA
We take words from the topic model to 'generate' a document. (Generation)
We take a doc and find the probability distribution in a model (Inference or Estimation)
The topic models can be more than one. (Mixture model)
LDA (A generative model)-
-> Choose the length of the document D
-> Choose the mixture of topics for the doc
-> Use topic's multinomial distribution to output words to fill the topic's quota
-> Pre-processing text must be done (Tokenize, Normalize, Stop word removal, Stemming)
-> Convert tokenized documents into Document Term matrix
-> Build LDA models on top of it