When does a compliment become sexist? Analysis and classification of ambivalent sexism using twitter data

By Akshita Jha, Radhika Mamidi. 2017

196 citations recorded by the source corpus.

Sexism is prevalent in today’s society, both offline and online, and poses a credible threat to social equality with respect to gender. According to ambivalent sexism theory (Glick and Fiske, 1996), it comes in two forms: Hostile and Benevolent. While hostile sexism is characterized by an explicitly negative attitude, benevolent sexism is more subtle. Previous works on computationally detecting sexism present online are restricted to identifying the hostile form. Our objective is to investigate the less pronounced form of sexism demonstrated online. We achieve this by creating and analyzing a dataset of tweets that exhibit benevolent sexism. By using Support Vector Machines (SVM), sequence-to-sequence models and FastText classifier, we classify tweets into ‘Hostile’, ‘Benevolent’ or ‘Others’ class depending on the kind of sexism they exhibit. We have been able to achieve an F1-score of 87.22% using FastText classifier. Our work helps analyze and understand the much prevalent ambivalent sexism in social media.

Subfields: Natural Language Processing, Safety, Robustness & Interpretability

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Natural Language Processing

In Safety, Robustness & Interpretability

Citations in the corpus

Fair Is Better than Sensational: Man Is to Doctor as Woman Is to Doctor (2019, background): “One is that they get easily propagated both in science itself (Jha and Mamidi 2017; Gebru et al. 2018; Mohammad et al. 2018; Hall Maudslay et al. 2019), also outside NLP and artiﬁcial intelligence (McQuillan 2018) and in popularized articles (Zou and Schiebinger 2018), where readers are usually in…”
Large Scale Crowdsourcing and Characterization of Twitter Abusive Behavior (2018, methodology): “A widely used and available dataset in this field, is the one in (Waseem and Hovy 2016), employed by several studies working on hate speech detection such as in (Badjatiya et al. 2017), (Gambäck and Sikdar 2017), (Park and Fung 2017), and (Jha and Mamidi 2017).”
Challenges for Toxic Comment Classification: An In-Depth Error Analysis (2018, background): “…classiﬁcation tasks, related work considers different aspects of toxic language, such as “racism” (Greevy and Smeaton, 2004; Waseem, 2016; Kwok and Wang, 2013) and “sexism” (Waseem and Hovy, 2016; Jha and Mamidi, 2017), or the severity of toxicity (David-son et al., 2017; Sharma et al., 2018).”
HONEST: Measuring Hurtful Sentence Completion in Language Models (2021, methodology): “In future work, we will investigate sentence completions with “benevolent sexism” categories (Jha and Mamidi, 2017), e.g., stereotypes like women are good at cooking or men are good at ruling .”
Finding Microaggressions in the Wild: A Case for Locating Elusive Phenomena in Social Media Posts (2019, positive): “These stereotypes may have inherently negative connotations (“lazy”), but may also be neutral (“liking pink”) or positive (“strong”), which complements recent work on benevolent sexism (Jha and Mamidi, 2017).”
Towards a Comprehensive Taxonomy and Large-Scale Annotated Corpus for Online Slur Usage (2020, methodology): “It contains 16k comments from Twitter, annotated according to the offense criteria of McIntosh (1988). Waseem (2016) is an extension of this corpus by 6,909 comments and it considers amateur as well as expert annotations.”
Effective hate-speech detection in Twitter data using recurrent neural networks (2018, background): “The work by Jha and Mamidi (2017) dealt with the classiﬁcation problem of tweets, but their interest was on sexism alone, which they distinguished into ‘Hostile’, ‘Benevolent’ or ‘Other’.”
A Unified Deep Learning Architecture for Abuse Detection (2018, methodology): “Some works try to detect more specific types of hateful behavior, such as racism [22, 24] or sexism [20].”
Directions in Abusive Language Training Data: Garbage In, Garbage Out (2020, background): “Common examples of group-directed abuse include sexism, which is included in datasets provided by [32, 43, 62, 101, 111] and racism, which is directly targeted in [99, 111].”
Understanding the Shades of Sexism in Popular TV Series (2019, background): “In order to analyze the inherent sexism, there has been much attention given in recognizing sexism in Tweets collected with certain hash-tags (Waseem and Hovy, 2016; Park et al., 2018; Jha and Mamidi, 2017) via machine-learning approaches.”
Automatically Inferring Gender Associations from Language (2019, background): “It is well-established that gender bias exists in language – for example, we see evidence of this given the prevalence of sexism in abusive language datasets (Waseem and Hovy, 2016; Jha and Mamidi, 2017).”
Bias in word embeddings (2020, background): “We traced sexist buzzwords under the notions of traditional sexism [32, 73], while we defined and searched sex related compliments given theories of benevolent sexism [9, 59].”
RtGender: A Corpus for Studying Differential Responses to Gender (2018, background): “Social issues like gender bias are often not just about hostility but also behaviors such as stereotype-reinforcing benevo-lence (Eagly and Mladinic, 1989; Glick and Fiske, 1996 Jha and Mamidi, 2017).”
TweetBLM: A Hate Speech Dataset and Analysis of Black Lives Matter-related Microblogs on Twitter (2020, background): “(Jha and Mamidi, 2017) proposed the datasets with classes belonging to sex-ist, racist, benevolent sexism, and hostile sexism.”
Directions in abusive language training data, a systematic review: Garbage in, garbage out (2020, background): “Of the four datasets that comprise 100% abusive content, three are synthetic datasets (reported in one publication [72]) and the other is an amendment to an existing dataset and only contains sexism [59].”
A Just and Comprehensive Strategy for Using NLP to Address Online Abuse (2019, background): “First, although recognized within the larger NLP abuse typology (Waseem et al., 2017), only a handful of approaches have attempted these problems, such as identifying benevolent sexism (Jha and Mamidi, 2017), and new methods must be developed to identify the implicit signals.”