Sentiment Analysis | Interpret the output of Naive Bayes Classification

Moe
Moe New Altair Community Member
edited November 5 in Community Q&A
Hello Rapidminer Community, 

Many thanks in advance to anyone who helps :)

I´m currently analysing Employer Reviews from four different companies on the Employer Review Platform "Kununu" for my Bachelor Thesis. 

What I´m trying to do is finding different topics/words that reoccur in either positive or negative reviews. So for example if "Salary" "pay" "paid" occurs more often in negative reviews it could mean that a company has to update their salary structure to improve Employee Satisfaction.

I thought of using Naive Bayes to do that. What I basically did is, I labelled the reviews as either positive or negative, then I applied the process I attached to this comment. 

The Output of this process features a list of words and there likelyhoods to occur in either positive or negative texts (from my understanding of what NB does). For example I attached the likelyhoods of the word "Berufseinstieg" (which translates to "starting to work").

In the next step I would group similar words like (sticking to the example) "Salary" "pay" "Money" and interpret the likelyhoods to find factors that Employees tend to find great or bad. 

Could I use the output of NB in the way I described or would that be wrong because I am quite new to sentiment analysis and I could not find any papers with studies that did something like this.

Thanks a lot in Advance and have a great day





Tagged: