The whole article is quite funny, especially the lists of most used tankie words, or the branding of foreignpolicy as a left-wing news source.

  • millennialchaos@lemmygrad.ml
    link
    fedilink
    English
    arrow-up
    4
    ·
    1 year ago

    I did an in-depth ‘debunk’ of this study.

    I want to highlight the most egregious part of it, to me at least. Here’s an excerpt from my article:

    As we find later in Section 5.4, tankies have the most proportion of posts with high identity attack against Jews in the far-left community.

    ??? Let’s pull up that section quickly:

    The Perspective API [92] is a widely used [9, 12, 26] tool for measuring toxicity. Although it has limitations, e.g., there are issues of bias and questions of performance when encountering conversation patterns that it was not trained on, at scale it provides a decent measure for comparison between online communities.

    They used an API tool to analyze comments on the tankie subreddits. They specifically mention that it has limitations if it wasn’t trained on certain conversation patterns. The Perspective website doesn’t mention it being trained on Reddit comments or comments in leftist communities. This is junk science, of course.

    Finally, we observe that tankies frequently target Muslims and Jews in their posts.

    I’m not about to dig too deep into the way this API determines what constitutes an Identity Attack, since this study doesn’t even attempt to elaborate on it, but I’m going to assume that if it detects ‘hateful words’ in the same comment as a ‘named entity’ like Jew or Muslim, it just assumes the comment is attacking that entity.

    Here’s the problem. A comment like this:

    “Zionists are pieces of shit for assuming all Jews support Israel”

    or this:

    “Implying that the US gives a fuck about Muslims when they criticize China is delusional”

    would likely be considered by this bot to be an attack against Jews or Muslims. Curiously, this report doesn’t provide a single shred of evidence of these attacks on Jews or Muslims. But, in the ‘C.1 Qualitative Validation’ section, they do give some examples of the toxic comments that this bot identified. Not a single one is specifically about Jews or Muslims.

    Here’s two examples:

    To me, boarding schools serve as schools for potential terrorists, and China’s approach seems more humane than the US’s

    and

    Zionism equates to Fascism.

    Neither comment is an Identity Attack against Muslims or Jews. The first is talking specifically about the small portion of Uyghurs that China has identified as being radicalized, not all Muslims. The second is about Zionism, which as this study pointed out, does not mean all Jews. Neither one contains the word ‘Jew’, or ‘Muslim’, anyway.

    Hmm, I wonder why they omitted that. Because the truth doesn’t fit the ‘tankie bad’ narrative they are pushing? This is research misconduct, pure and simple, and this singular example of evidentiary omission should cause any non-tankies reading this study to dismiss it in its entirety. But of course, it won’t.