4 ms·
Interesting, but this is a really naive (and outdated) NLP technique for analyzing sentiment. The researchers are injecting an enormous amount of bias by choosi
by snikolic 15y ago
Interesting, but this is a really naive (and outdated) NLP technique for analyzing sentiment. The researchers are injecting an enormous amount of bias by choosing apriori what words are considered positive and negative - they'll also miss bad spellings, abbrevs, sarcasm, etc. I worked for a company doing similar analyses (but with different techniques), and I've heard of margins of error as bad as 40% for this type of analysis. You might be better off flipping a coin.
- civilian 15y agoYet, if it was coin-flipping, shouln'td we get wildly different results for each of the 7 days, since they were assumingly analyzed separately?
- sev 15y agoWhen a statistical experiment is flawed, anything goes.
- snikolic 15y agoFlawed results can correlate with flawed results. If we measure Monday incorrectly, and we have the same flaw in how we measure Tuesday, is it a surprise that the results for both days might look similar? All we've proven is that people use similar words at similar times of days - not necessarily anything about their sentiment or emotion. The coin flipping comment is a joke. These are obviously not random results. But that doesn't mean they're accurate or meaningful, either.
- pork 15y agoGreat comment. I just wanted to add that "not random" != "interesting".