Multi-ion ionic liquids and a direct, reproducible, diversity-oriented way to make them

Abstract

Multi-ion ionic liquids featuring large numbers of distinct imidazolium cations can be easily and reproducibly prepared in a simple one-pot procedure. The method provides a dramatic improvement in efficiency over the almost universally used approach of mixing pre-existing ILs to make multi-ion systems.

Introducing CAD: the Contextual Abuse Dataset

“…The quality of annotations for abusive datasets has been widely critiqued, and inter-rater agreement scores are often remarkably low. Wulczyn et al (2017) report an Alpha of 0.45, Sanguinetti et al (2018) Kappas from k=0.37 for offence to k=0.54 for hate, Gomez et al (2020) report Kappa of 0.15 in the "MMH150" dataset of hateful memes, and Fortuna and Nunes (2018) report a Kappa of 0.17 for a text-only task. In a classification study of prejudice against East Asia, find that 27% of classification errors are due to annotation mistakes.…”

“…Harm includes physical violence, emotional abuse, social exclusion and harassment. This is one of the most harmful forms of hateful language (Marwick and Miller, 2014;Citron and Norton, 2011) yet usually it is part of an 'explicit' hate category (Zampieri et al, 2019;Wulczyn et al, 2017;Waseem and Hovy, 2016) and few datasets have treated it as a separate category, see Golbeck et al (2017), Anzovino et al (2018), and Hammer (2014) for exceptions.…”

“…Person-directed abuse includes serious character based attacks, such as accusing the person of lying, as well as aggression, insults and menacing language. Person-and Identity-directed forms of abuse are often addressed in separate taxonomies, although in some studies they have been merged into a more general 'toxic' category (Wulczyn et al, 2017;Golbeck et al, 2017).…”

“…Taxonomies of abuse Taxonomies vary in terms of the scope of abusive behaviours they cover. Some offer categories for abuse against both individuals and groups (Zampieri et al, 2020), others cover only abuse against identities Fortuna and Nunes, 2018;Kiela et al, 2020), against only a single identity, such as misogyny (Anzovino et al, 2018) or Islamophobia (Vidgen and Yasseri, 2019), or only abuse against individuals (Wulczyn et al, 2017). Some research distinguishes between content in different languages or taken from different platforms (Kumar et al, 2018).…”

“…They provide labels for (a) offence (yes/no), (b) whether a group is targeted, and (c) whether the abuse is intentional. Wulczyn et al (2017) identify different interpersonal abuse, including toxicity, aggression and attacks. Some taxonomies explicitly separate abuse from closely-related but non-abusive forms of online expression.…”