Unsupervised Sentimental Analysis in R

Question

How would you evaluate unsupervised sentimental analysis?
I am reading on evaluating sentimental analysis and learning that much of the classification models that are being used, the data has target/label variables. How would you evaluate a survey that is only text with no target/label variables? Right now, I am just giving the responses a sentimental score (positive/negative/neutral) but I am having trouble approaching validation with 1000+ responses. Any ideas/suggestions?

Erwan · Answer

In a truly unsupervised setting, the only possibility is to cluster the documents. Typically this would be done with topic modelling. Then the result could be evaluated by inspecting the words most associated with every topic, and assign a sentiment class to the whole topic/cluster, i.e. all the documents labelled with it.

Answered by Erwan on December 3, 2020

Unsupervised Sentimental Analysis in R

One Answer

Add your own answers!

Ask a Question