I recently asked my son if he was “up for going skiing.”

He replied that he was “down for that.”

Down is the new up, as we all know.

Language changes over time which is one of the challenges we face when we build AI models to score articles for sentiment.

Clients are increasingly looking at sentiment as a way to detect opportunities. It helps focus their attention and assess the general mood about a company.

But the scores are only useful if they are accurate. That means models need to be re-trained on a regular basis.

Bloomberg engineers have built one model for news articles about companies and a separate one for social media because the content is so different.

Bloomberg Terminal clients can chart social sentiment by running the GT function. Type {FB US <Equity> GT <go>} and toggle the range to one month.

The blue barchart shows the explosion of Twitter posts after Meta reported disappointing earnings. The barchart at the bottom scores each tweet as positive (green) or negative (red).

It’s not surprising that articles about Meta scored negative. Sentiment is not a magic number that tells you something you don’t know.

What the score does is provide a measurement, a quantifiable way for investors to make meaningful comparisons with previous periods or other companies.

That’s something even my son would be down with.