15-12-2025, 02:14 PM
(This post was last modified: 15-12-2025, 02:33 PM by nablator.
Edit Reason: n
)
(15-12-2025, 11:49 AM)dashstofsk wrote: You are not allowed to view links. Register or Login to view.Brighter colours show where there is better correlation.
How do you define correlation? I tried one method as far as possible from LSA similarity to see what happens.
The idea is to test various methods to find out what makes the studied property appear (a "very strong textual correlation across conjoint bifolia" in Q13 and Q20). LSA similarity is a black box, it does not help understand what matters exactly and what does not: word order or proximity? Sub-word tokens (BPE segmentation or prefix-stem-suffix decomposition by some grammar)? Glyph bigram/trigram statistics?
