In a case where thousands of documents from the same domain is fed into the standard LSA topic modeling, resulted topics are confusing. In some cases many keywords in each topic is being repeated in many topics.
I am new to topic modeling. A sample result is shown below (the keyword play is being repeated).
topic ratio keyword
0 0.178252012 play
0 0.13429748 like
0 0.129169756 get
1 -0.299574178 ending
1 -0.213000647 play
2 0.395652058 ending
2 -0.296851122 play
What could be the problem?
How to confirm the model is built correctly?
In English the word 'play' has many meanings. For example, it can be used as a verb or as a noun. Your LSA topic modelling has correctly distinguished between these two meanings! (Magic!) In topic 0 it is a verb=fun. In topic 1 it is a noun=dramatic work. In topic 2 it is probably dealing with the domain of sport, or schoolbell ringing or technical, dealing with recordings?
For example, WordWeb lists 17 noun meanings and 37 verb meanings for "play"! Now, how to teach your software to recognise the other 49 meanings?? What does a negative ratio means?
- Badges
- Science method
- Similar topics
- Teaching