Authors:
(1) Utkucan Balci, University of Binghamton, United States;
(2) Michael Sirivianos, University of Cyprus, Cyprus, Cyprus;
(3) Jeremy Blackburn, University of Binghamton, United States.
Connection table
Summary and 1 Introduction
2 Background and related work
3 data
3.1 Definition of Tankie subtitles
3.2 Defining Ideology Sub -Series and 3.3 Post Collection
4 user base analysis and 4.1 graphic structure and community detection
4.2 Community Growth
4.3 User transitions over time
5 Content Analysis and 5.1 What are the tanks talking about?
5.2 Who are the witnesses talking about?
5.3 False alignment analysis
5.4 toxicity analysis
5.5 Impact Analysis
5.6 Lemmygrad Analysis
6 DISCUSSION AND CONCLUSION AND 6.1 LIMITING
6.2 Inferences and future studies and references
A data
B Nations
C False alignment analysis
D field analysis
5.4 toxicity analysis
In this section, we use Perspective API models to compare the online behavior of tanks and other remote communities.
Perspective API. Perspective API [92] commonly used [9, 12, 26] Vehicle to measure toxicity. Although it has limitations, for example, there are prejudice and performance problems when encountering speech models that are not trained, it provides good measure for comparison between online communities on a scale. API provides six production model: 1) toxicity, 2) serious_toxicity, 3) Insult, 4) Identity_attack, 5) threat and 6) profanity (see. [91] For all details about models). Hoseini et al. [57]. To be a base line for comparisons, describes more than 36 m posts of 0.5% of Reddit broadcasts during the timeline of the data set.
Consequences. Figure 5 shows cumulative distribution functions (CDFs) for each model in the remote cluster. Our analysis shows that tanks have higher scores for all perspective API models (except R/Alltheleft) from other Farless communities. In addition, all remote communities have higher perspective API scores than the basic reddit example.
In particular, tanks have high scores for both β₯ 0.5 score and Identity_Attack and threat (ie β₯ 0.8 points) and the second highest scores for remaining models. Table 5 shows that tanks have almost twice as high as the average of other remote communities.
We confirm that the score distributions for each model are significantly different between tanks and other left communities using the 2-for example, using the Benjamini-Hochberg method for multiple tests π <0.01 for all π <0.01). These results show that tanks tend to make poles with higher levels of toxicity, insult and blasphemy than other remote communities except R/Alltheleft. Tanks also tend to do tasks with more identity attacks and threats than all other remote communities.
Then, we remove the naming beings in the publications of the tanks by removing the named beings that appear less than 100 floors. In Table 6, we present the existence of the first 10, which are listed by some of the publications that talk about assets and have high points in all perspective models. For all models except Identity_attack, the most mentioned assets are primarily related to the US (eg, American, Yankee, Charlotte, Qanon),
Public figures and politicians (eg John Oliver, Elon Musk, Kyle Rittenhouse, Anthony Blinken, Erdogan, Alex Jones, Joe Rogan) or countries/nationalities (eg Chechen, Brasil, Iraqis). For the Identity_Attack model, the most common beings are typically religious or ethnic groups. Witnesses are primarily aimed at the Muslims and Jews, 41.80% of these groups and 34.03% of these groups talk about these groups with high identity_attack scores; The highest rate of high identity_attack scores of any remote community we analyze for Muslims and Jews. Moreover, we observe that tanks attacked the identities of Asians, Arabs, Hindus, Mexicans, Africans and whites in more than 20% messages talking about these identities.
Package. Our analysis shows that tanks have the highest scores for identity_attack and threat among the left -wing communities, and that they have the highest scores for toxicity, serious_toxicity, profanity and insults behind R/Alltheleft. Although our findings show that US politics is not the main focus of tanks, they still express strong views on US -related events, conspiracy theories, politicians and public figures. Finally, we observe that tanks often aim in their duties of Muslims and Jews.
