Hacker News Reader

Comment by HopenHeyHi | Hacker News Reader

  As the presence of hate words does not guarantee that the text is hateful, we utilize the Perspective API, a family of models used for detecting toxic content, to further filter our data

Yeah. Running a random list of words you don't like through a blackbox is some garbage-in garbage-out voodoo. It hardly matters what random words bots were able to generate on Twitter, what might matter is how many humans actually saw and interacted with any of it.

  The spike in March, 2022 coincides with the Canada convoy protests.

Proceeds to show literally the opposite. There is no measurable permanent spike of anything at all in their own graph!

These sort of "studies" will hurt the credibility of researches longer term.

View on HN

I think you misread the chart?

> While baseline tweets remain stable throughout the year (with an exception in early 2022 coinciding with the Canada convoy protests (Soucy 2022)), hateful tweets increase dramatically following Musk’s takeover.

So the claim is that total tweets increase in early 2022. Total tweets in Figure 2 is the top orange line. There's a noticable spike approximately in April and another one around August. (That's going by my faulty eyes, so their claim that it's in March makes sense to me.)

Was that caused by the trucker protests? No idea. But there are lots of interesting looking spikes in this chart.

I think the chart is either extremely poorly made or intentionally misleading?

> There's a noticable spike approximately in April and another one around August

> Was that caused by the trucker protests? No idea. But there are lots of interesting looking spikes in this chart.

Precisely. They simply marked two occasions they don't like. Might as well correlate with mercury being in retrograde.

What is the baseline? The mean? The year over year? Look at the numbers about 100+-50 "bad" tweets on a given month.. this isn't a serious effort, I am embarrassed on their behalf.

I think you're misunderstanding the claim about the trucker convoy. They just said more people started tweeting around then.

The chart seems reasonably clear to me.

They define baseline pretty clearly.

> To ensure fluctuations in hate speech are not reflective of fluctuations in overall user activity, we also sample a base- line set of tweets collected during the same time intervals. We used the keyword “thing,” taken from a set of the most commonly used words (Geoffrey Leech 2001).

> I think you're misunderstanding the claim about the trucker convoy. They just said more people started tweeting around then.

OK. Why? A lot of things happened around then. Why mention the trucker convoy specifically?

> They define baseline pretty clearly.

How would a chart look like zoomed out to multiple years?

> OK. Why? A lot of things happened around then. Why mention the trucker convoy specifically?

Presumably they believed it was the correct explanation for the anomaly in the chart?

> How would a chart look like zoomed out to multiple years?

That's an interesting question but it isn't the one this paper is written about, which is a specific event.

May I ask if you're seeking to understand this paper or to criticize it? Because it kinda feels like when your objections were addressed you came up with some more of them?

> Presumably they believed it was the correct explanation for the anomaly in the chart?

On what basis? How is it not an arbitrary event? There is absolutely nothing else in the paper about this event.

> May I ask if you're seeking to understand this paper or to criticize it? Because it kinda feels like when your objections were addressed you came up with some more of them?

Is that not what you're supposed to do with papers? Is this a study or marching orders?

It's not about the trucker convoy, no. On what basis? There's a citation. (Soucy 2022) If you're interested, you can check it out.

It's not "marching orders", certainly. But reading to find how you might dismiss it is different from reading it critically. Like, it just strains my ability to believe you're reading the paper in good faith when you imply they haven't defined the term baseline, when it's simple to find the definition just by searching.

> But reading to find how you might dismiss it is different from reading it critically.

How so?

You're free to believe whatever you wish about me, only the actual issues raised are what matters, not the person raising them.

> It's not about the trucker convoy, no.

This is far more important than whether or not I personally rub you the wrong way. It means the paper is not legitimate.

> when it's simple to find the definition just by searching

I didn't just search, I actually read it. Stare at the paragraph again:

> Figure 2: Overall hate speech on Twitter increased after Elon Musk bought the platform. Number of hateful tweets sam- pled during each week in 2022, compared to a baseline of tweets collected during the same time period. The spike in March, 2022 coincides with the Canada convoy protests.

Look at the chart and this description and tell me you don't understand the problem here.

The difference is that if you make up your mind about the paper before you've even read it, a.) You've written a self fulfilling prophecy, b.) Your criticism will be shallow, and c.) You're not going to understand the paper.

The problem isn't that I have any animus towards you (which I don't). Frankly the problem is that you're misrepresenting what's in the paper. If you say you've read it, you shouldn't be saying it's misleading and justifying that by things that are trivially checked by searching.

I'm doing my best to presume good faith and to presume that you weren't doing this intentionally. But you're not really giving me a lot to work with in that regard.

I've explained quite clearly why I think it is misleading without misrepresenting anything. Some of my points you've conceded others you've decided to ignore and make it about me.

If you read it as well and deem this paper of good quality I suppose we can just agree to disagree and move on what can I say. I believe in following the truth wherever it may lead not fitting data to politically motivated conclusions which this work is very clearly doing, sad state of affairs.

I share this value of pursuing the truth without favor or prejudice; I think you should consider whether that was really what you went about doing. When I read your responses here, I honestly do get the impression you read this document with an agenda. When I pushed back on that, it didn't really seem like you denied it as much as justified it. That's not a problem for me, I do my best not to be invested in the opinions of others, but from what you're saying, maybe it should be a problem for you.

Best of luck, stranger. See you around.

Sorry you got that impression, that is between you and yourself as it were. I don't have anything to justify, as once again, my one and only agenda is as you say pursuing the truth without favor or prejudice.

It does not matter to me in which direction the paper is politically slanted one iota, or what conclusions it so desperately seeks. The ends don't justify the means, sorry.

It matters very much to me that it isn't anything like science, and worse, that it masquerades. You can dance around this all you want.

Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.

If you wouldn't mind reviewing https://news.ycombinator.com/newsguidelines.html and taking the intended spirit of the site more to heart, we'd be grateful.

I am familiar with the guidelines and the intended spirit of the site and am totally on board.

Are you referring to something I posted in this thread as unsubstantive and flamebaity?

To be clear, nothing I wrote here was written intended as such.

Edit (since I can't reply): You've met me at a very strange time in my life, what can I say, I'll keep those notes in mind.

To be clear, nothing from this particular thread?

"You can dance around this all you want." was a flamebait swipe.

https://news.ycombinator.com/item?id=35563582 was unsubstantive flamebait.

https://news.ycombinator.com/item?id=35568205 was flamebait ("So... Schizophrenia?")

https://news.ycombinator.com/item?id=35567227 was particularly bad ("What on earth are you talking about?", "Have you been watching the dictator drunk again?", etc.)

You've been breaking the site guidelines in other places too (e.g. https://news.ycombinator.com/item?id=35534129).

Reply on news.ycombinator.com