How Algorithms discern our mood from what we write online | Hacker News Reader

How Algorithms discern our mood from what we write online

sinapticas.com

Transcription of zoom calls and other "recorded for quality assurance" support calls are already being fed into sentiment analysis to determine customer satisfaction.

Employees are evaluated based on positive caller sentiment.

Cheer pressure is now endemic and enforced by machine.

Another data point to be sold, bought, and resold?

Just another point on ad performance marketing: it's only for those foolish enough to believe it without putting in any critical thought.

You might think the first step in sentiment analysis would be teaching the computer to understand what humans are saying. But that’s one thing that computer scientists cannot do; understanding language is one of the most notoriously difficult problems in artificial intelligence. Yet there are abundant clues to the emotions behind a written text, which computers can recognize even without understanding the meaning of the words.

Without parsing meaning, the best case seems to be fairly shallow sentiments based on equally shallow norms and trends.

I am not saying this has no value. In a "is good or bad" sense, sure.

The piece goes on to talk about gauging mental health...

That scares me more than a little. When ones scope of common expression lies solid in the bell curve, maybe work of this kind can get us some basic sentiments.

What happens when ones expression lies outside that curve some?

Nothing good.

A while back I had written, "go ahead, cut your nose off to spite your face" in response to a Twitter discussion gone south, as so many do.

The humans in the discussion picked up on it, and we stepped back and picked the thread up from a better place and carried onto an even better place.

A couple humans did not get it and some computer somewhere decided I was advocating self harm.

In my appeal, I cited the phrase as colloquial speech, and dropped a Wiki link that spoke to all that.

Took about 10 minutes for another human to assess things and undo the moderation action.

Comics often talk about how a fairly significant percentage of people simply do not understand satire. "The Colbert Report" stands as an exemplary example of that dynamic in play. In my own circles, I know people who absolutely did not grok the show was poking fun at them politically. It was profound to me.

Moderating various communities has been telling too. Some of us are adept at inferring possible meaning from the written word. Most that are skilled, are also shockingly unaware that is not the meaning! Just a plausible one.

How many times have you been told, "no, this is what you meant" by someone else absolutely sure they are a superior authority regarding your own intent?

Me? Too many times to the point where I will assert myself as the final authority on my intent before something like that gets out of hand. That should go without saying. A given.

That it must be?

No way will software get anywhere healthy on sentiments, and seems a fantastically bad mental health idea. (However well intended it may be)

People have real trouble, frankly.

Reply on news.ycombinator.com