Authors: (1) UTKUCAN BALCI, Binghamton University, United States; (2) MICHAEL SIRIVIANOS, Cyprus University of Technology, Cyprus; (3) JEREMY BLACKBURN, Binghamton University, United States. Authors: Authors: (1) UTKUCAN BALCI, Binghamton University, United States; (2) MICHAEL SIRIVIANOS, Cyprus University of Technology, Cyprus; (3) JEREMY BLACKBURN, Binghamton University, United States. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background & Related Work 2 Background & Related Work 3 Data 3 Data 3.1 Identifying Tankie Subreddits 3.1 Identifying Tankie Subreddits 3.2 Identifying Ideology Subreddits and 3.3 Post Collection 3.2 Identifying Ideology Subreddits and 3.3 Post Collection 4 User-Base Analysis and 4.1 Graph Construction & Community Detection 4 User-Base Analysis and 4.1 Graph Construction & Community Detection 4.2 Community Growth 4.2 Community Growth 4.3 User Migrations Over Time 4.3 User Migrations Over Time 5 Content Analysis and 5.1 What do tankies talk about? 5 Content Analysis and 5.1 What do tankies talk about? 5.2 Who are tankies talking about? 5.2 Who are tankies talking about? 5.3 Misalignment Analysis 5.3 Misalignment Analysis 5.4 Toxicity Analysis 5.4 Toxicity Analysis 5.5 Domain Analysis 5.5 Domain Analysis 5.6 Lemmygrad Analysis 5.6 Lemmygrad Analysis 6 Discussion & Conclusion and 6.1 Limitations 6 Discussion & Conclusion and 6.1 Limitations 6.2 Implications & future work, and References 6.2 Implications & future work, and References A DATA A DATA B NAMED ENTITIES B NAMED ENTITIES C MISALIGNMENT ANALYSIS C MISALIGNMENT ANALYSIS D DOMAIN ANALYSIS D DOMAIN ANALYSIS 5 CONTENT ANALYSIS In this section, we examine the focus of tankies’ discussions and compare them with other far-left communities using a variety of techniques. We first use topic analysis to understand what tankies are talking about. We then use named entity recognition to determine who tankies are talking about. Next, we compare tankies and other far-left communities with respect to language misalignment and conceptual homomorphisms. 5.1 What do tankies talk about? We begin by determining what tankies talk about via topic modeling. We leverage BERTopic [45], a technique that uses pre-trained, transformer language models to create document embeddings, applies dimensionality reduction with UMAP [80], hierarchically clusters these embeddings with HDBSCAN [79], and finally creates topic representations with class-based TF-IDF. Unlike traditional topic modeling techniques (e.g., LDA), BERTopic derives topics based on documents rather than words, taking into account the context and semantics of words. Previous research [36] found that it performs better in terms of topic separation than LDA and another embedding-based topic modeling approach, Top2Vec [11]. After removing hyperlinks, and posts from “Automoderator” (a bot which handles a variety of automated moderation tasks), we train our model using the default parameters. Due to the large number of clusters produced by the underlying technique of BERTopic, following [114], we hierarchically reduce these clusters to a range of 20-80 (incrementing by 10 at each step) and select the model with the highest coherence score; 50 in our case. To analyze and compare the discussions of tankies’ with other far-left communities, we introduce a metric called Relative Topic Prevalence Rank (RTPR), which is calculated by ranking each community within a topic relative to the percentage of posts in that fall within in that topic. A higher RTPR means that a given community focuses on a particular topic more than another community does, relatively speaking. By using RTPR, we can compare and interpret the discussions of tankies within far-left communities, highlighting their unique focus areas and preferences. This metric allows us to better understand how tankies differ from other far-left groups in their discourse and ideological leanings. Results. Table 2 presents the topics discussed by tankies ordered by the number of posts in each topic, as well as tankies’ RTPR for each topic. We find that the most popular topic among tankies is the Uyghur (a Turkic ethnic group from Xinjiang) Genocide, accounting for 1.11% of all tankies posts. In addition, we see that topics related to Marxist/Leninist ideologies (socialism, communism, fascism, and capitalism) and two AES states (North Korea and USSR) in the top 10. When we look at the RTPRs, we find that they generally focus more on recent, state-level political events compared to other far-left communities. As we can see from the table, tankies have RTPR = 1 for the topics related to Uyghur Genocide (#1), North Korea (#4), Russian invasion on Ukraine (#24), China-US conflicts (#29), Israel-Palestine conflict (#37), communism (#2), fascism (#8), Stalin (#17). Results On the other hand, tankies rank in the bottom-half when it comes to focusing social problems. We find tankies have RTPRs between 4-7 on topics related to police (#20), climate change (#42), healthcare (#44), housing (#45), guns (#46), unions (#48), and taxes (#49). Additionally, our analysis reveals that tankies have RTPRs 2-3 on topics related to USSR (#9), tankies (#25), and Cuba (#31). However, even though topics related to socialism and capitalism are among the top 10 most popular topics of tankies, they have RTPRs 6 and 5 for these topics, respectively. These results suggest that tankies prioritize discussions of state-level political events over core ideological issues and social problems that have long concerned the far-left community. Takeaways. Our findings reveal that the predominant topic of discussion among tankies centers on the Uyghur genocide, which happens in an AES country, China. Tankies prioritize discussions of state-level political events over social issues, as indicated by their higher relative topic prevalence ranks (RTPRs) for topics, e.g., Uyghur Genocide, Russian invasion on Ukraine, and Israel-Palestine conflict. This is further supported by their lower RTPRs for social issues, e.g., police, climate change, healthcare, housing, gun control, unions, and taxes compared to other far-left communities. Takeaways This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv
