Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

LLM-based sentiment analysis of Hacker News posts between Jan 2020 and June 2023

outerbounds.com

Post reply on HN
Print view View original

21–30 of 79 posts

Re: LLM-based sentiment analysis of Hacker News posts between Jan 2020 and June 2023

#22

Post by Sleaker »

Why is everything only plotted between 4 and 8 if the scale of the least liked topic should be 0 and most liked should be 9. Also 4.5 is the midpoint, but 4 is displayed as bright red and 6 is a muted gray blue, why? This makes no sense except to be psychologically disingenuous.

And no 5s? What is even going on in that LLM?

Re: LLM-based sentiment analysis of Hacker News posts between Jan 2020 and June 2023

#23

Post by EarthLaunch »

> Tokens Don't Lie

> But how do people feel about these topics

I find it notable that tokens don't necessarily express people's feelings. Put another way, tokens aren't how people feel, they're how they write.

Samstave mentioned in this thread that Twitter is a 'global sentiment engine'. I'm sure that's literally true. Sentiment measurement is only accurate to the degree that people are expressing their real feelings via tokens. I can imagine various psychological and political reasons for a discrepancy.

If you did sentiment analysis of publicly known writings of North Korean administrators, would that represent their feelings?

I think the interplay with free speech is interesting here: In a setting where people feel socially and legally safe to express their true opinion, sentiment analysis will be more accurate.

Re: LLM-based sentiment analysis of Hacker News posts between Jan 2020 and June 2023

#25

Post by throwawaymaths »

antihipocrat wrote: post #18
tantalor wrote: post #7

Is this just using LLM to be cool? How does pure LLM with basic "In the scale between 0-10 ..." prompt stack up against traditional, battle-tested sentiment analysis tools? Gemini suggests NLTK and spaCy https://www.nltk.org/ https://spacy.io/

I'm wondering how their LLM parsing 250 mil words in 9 hours compares with performance of traditional sentiment analysis. Also, many exisiting sentiment analysis tools have a lot of research behind them that can be referenced when interpreting the results (known confounds etc). I don't think there is yet an equivalent for the LLM approach

And yet, it's so much easier to deploy an LLM, either through a service or on prem.

Re: LLM-based sentiment analysis of Hacker News posts between Jan 2020 and June 2023

#27

Post by tantalor »

throwawaymaths wrote: post #25
antihipocrat wrote: post #18

Earlier quoted context omitted.

I'm wondering how their LLM parsing 250 mil words in 9 hours compares with performance of traditional sentiment analysis. Also, many exisiting sentiment analysis tools have a lot of research behind them that can be referenced when interpreting the results (known confounds etc). I don't think there is yet an equivalent for the LLM approach

And yet, it's so much easier to deploy an LLM, either through a service or on prem.

It's easier to do a lot of things. That doesn't make it better.

Re: LLM-based sentiment analysis of Hacker News posts between Jan 2020 and June 2023

#29

Post by anonu »

ysofunny wrote: post #28
anonu wrote: post #2

Great analysis. How is divisiveness actually calculated?

the most divisive topic seems to be "gnome" with 0.82 on the divisiveness scale that's really "hacker", a worthy first place

more like h4x0r
Post reply on HN