Today's social media landscape is flooded with unfiltered content, which can range from hate speech to cyberbullying and cyberstalking. As a result, locating and eliminating such toxic language presents a significant challenge and is an active current research area. In this paper we focus on detecting hate speech against women, i.e. misogyny, exploiting a “prompt-based learning” paradigm with the aim of providing a first assessment of recent developed LLM (OpenAI's GPT-3.5-turbo). We experiment with a benchmark dataset of Reddit posts and evaluate different prompts types w.r.t. response stability, classification accuracy and inter-annotator agreement. Our experiments show that zero-shot detection GPT capabilities - against human annotations - outperform supervised baselines on our evaluation dataset and that ensembling different prompts possibly further improve the accuracy up to 91%. We also found that responses to specific prompts is quite stable, while slightly more variation and less agreement is observed when asking the questions in different ways.

Can LLMs assist humans in assessing online misogyny? Experiments with GPT-3.5

Morbidoni C.;Sarra A.
2023-01-01

Abstract

Today's social media landscape is flooded with unfiltered content, which can range from hate speech to cyberbullying and cyberstalking. As a result, locating and eliminating such toxic language presents a significant challenge and is an active current research area. In this paper we focus on detecting hate speech against women, i.e. misogyny, exploiting a “prompt-based learning” paradigm with the aim of providing a first assessment of recent developed LLM (OpenAI's GPT-3.5-turbo). We experiment with a benchmark dataset of Reddit posts and evaluate different prompts types w.r.t. response stability, classification accuracy and inter-annotator agreement. Our experiments show that zero-shot detection GPT capabilities - against human annotations - outperform supervised baselines on our evaluation dataset and that ensembling different prompts possibly further improve the accuracy up to 91%. We also found that responses to specific prompts is quite stable, while slightly more variation and less agreement is observed when asking the questions in different ways.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11564/822522
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 0
  • ???jsp.display-item.citation.isi??? ND
social impact