[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83212-en":3,"doc-seo-83212-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},83212,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Understanding Interpretation Difficulty in Harmful Online Communication Insights from Cybercrime Communities","Harmful online communication often relies on slang, coded terms, abbreviations, and community-specific expressions, which complicate message interpretation. An exploratory study investigates interpretation difficulty in Discord chats tied to cybercrime communities. Reference interpretations are built from purposefully selected difficult messages and reviewed by an expert, then used to assess human and large language model (LLM) interpretations under varying context conditions. Results show humans benefit substantially from external knowledge and extended context, while local context alone is often insufficient. For LLMs, larger models and local context improve interpretation, and qualitative error analysis yields a preliminary classification of causes for interpretation failure, supporting evidence-integration over message-level classification.","Understanding Interpretation Difficulty in Harmful Online Communication: Insights from Cybercrime Communities  \nTomohiro Okatsu, Naoki Takada, Yin Min Pa Pa, Katsunari Yoshioka, Tatsunori Mori  \nYokohama National University, Japan {mori-tomohiro-bm, [takada-naoki-xs}@ynu.jp](takada-naoki-xs}@ynu.jp)[ ](takada-naoki-xs}@ynu.jp){yinminn-papa-jp, yoshioka, [tmori}@ynu.ac.jp](tmori}@ynu.ac.jp)  \narXiv :2607 .07277v 1 [ cs .CL] 8 Jul 2026  \nAbstract  \nHarmful online communication often contains slang, coded terms, abbreviations, and community-specific expressions, which make messages difficult to interpret. This paper presents an exploratory study of interpretation difficulty in Discord chats related to cybercrime.  \nWe construct reference interpretations of purposefully selected difficult messages, which were reviewed by an expert. We then use them to evaluate human and large language model (LLM) interpretations under different context conditions. The results show that local context alone is often insufficient for humans, while external knowledge and extended conversational context substantially improve human interpretation. For LLMs, local context also improves interpretation, and the larger model performs better. We further conduct a qualitative error analysis and propose a preliminary classification of factors that make harmful chats difficult to interpret. These findings suggest that harmfulcontent analysis should treat interpretation asan evidence-integration problem, rather than as message-level classification alone.  \nWarning: This paper contains examples and discussions of harmful online content.  \n1 Introduction  \nCybercrime and other forms of harmful behavior on online platforms have become serious global problems (INTERPOL, 2026 ; Lukoši¯ute˙ et al., 2026) . Platforms such as Discord are increasingly used not only for ordinary communication but also for harmful activities, including malware distribution, credential theft, phishing, doxxing, and illicit trading (Sophos, 2021 ; Zscaler ThreatLabz, 2021 ; Heslep and Berge, 2024 ; Intel 471, 2024) .  \nPrior work has studied automated methods for detecting harmful online behavior, including grooming, cyberbullying, and hate speech (Vogt et al., 2021 ; Murnion et al., 2018) . However, harmful intent is often not expressed directly. Instead,  \nusers may rely on slang, coded terms, abbreviations, and community-specific expressions (Magu and Luo, 2018 ; Yuan et al., 2018 ; Hughes et al., 2023) . Such expressions make harmful messages difficult not only to detect, but also to interpret.  \nRecent large language models (LLMs) have shown strong abilities in analyzing informal and context-dependent expressions (Brown et al., 2020 ; Ziems et al., 2024), which may make them useful tools for supporting the interpretation of difficult online messages. Nevertheless, rapidly changing slang and domain-specific expressions remain challenging for LLMs (Sun et al., 2024 ; Mei et al., 2024 ; Zhang et al., 2024) . It is still unclear what makes harmful messages difficult to interpret and whether humans and LLMs struggle with the same types of difficulty.  \nIn this paper, we conduct an exploratory study of interpretation difficulty in cybercrime-related Discord chats. Using difficult messages, we compare human and LLM interpretations under various context conditions and analyze recurring sources of interpretation failure. Based on this analysis, we propose a preliminary classification of factors that make harmful chats difficult to interpret.  \nThe research questions of this paper are as follows:  \n• RQ1: How do humans and LLMs interpret difficult harmful messages under different context conditions?  \n• RQ2: What factors make harmful chats difficult to interpret?  \nOur contribution can be summarized as follows:  \n• We evaluate human and LLM interpretations of difficult cybercrime-related Discord messages under various context conditions.  \n• We classify recurring factors that make harmful chats d","cbCaiooRlbDAiMTZ","https://ap.wps.com/l/cbCaiooRlbDAiMTZ","pdf",231706,4,1,15,"English","en",105,"# Introduction\n## Crime and Harmful Communication on Online Platforms\n## Automatic Analysis of Harmful Online Conversations","[{\"question\":\"What makes harmful online messages difficult to interpret in cybercrime-related Discord chats?\",\"answer\":\"Messages often include slang, coded terms, abbreviations, and community-specific expressions that are hard to decode without the right context and knowledge.\"},{\"question\":\"How were reference interpretations created and used in the study?\",\"answer\":\"Purposefully selected difficult messages were assigned reference interpretations that were reviewed by an expert, then used to evaluate both human and LLM interpretations under different context conditions.\"},{\"question\":\"What context factors most improve interpretation for humans and LLMs?\",\"answer\":\"For humans, external knowledge and extended conversational context substantially improve interpretation, while local context alone is often insufficient. For LLMs, local context also helps, and larger models perform better.\"}]",1784185983,38,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"understanding-interpretation-difficulty-in-harmful-online-communication-insights-from-cybercrime-communities","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":20},"https://docshare.wps.com/document/understanding-interpretation-difficulty-in-harmful-online-communication-insights-from-cybercrime-communities/83212/",{"url":52,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What makes harmful online messages difficult to interpret in cybercrime-related Discord chats?","Question",{"text":75,"@type":76},"Messages often include slang, coded terms, abbreviations, and community-specific expressions that are hard to decode without the right context and knowledge.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How were reference interpretations created and used in the study?",{"text":80,"@type":76},"Purposefully selected difficult messages were assigned reference interpretations that were reviewed by an expert, then used to evaluate both human and LLM interpretations under different context conditions.",{"name":82,"@type":73,"acceptedAnswer":83},"What context factors most improve interpretation for humans and LLMs?",{"text":84,"@type":76},"For humans, external knowledge and extended conversational context substantially improve interpretation, while local context alone is often insufficient. For LLMs, local context also helps, and larger models perform better.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]