Comparative Studies of Detecting Abusive Language on Twitter

Abstract

The context-dependent nature of online aggression makes annotating large collections of data extremely difficult. Previously studied datasets in abusive language detection have been insufficient in size to efficiently train deep learning models. Recently, Hate and Abusive Speech on Twitter, a dataset much greater in size and reliability, has been released. However, this dataset has not been comprehensively studied to its potential. In this paper, we conduct the first comparative study of various learning models on Hate and Abusive Speech on Twitter, and discuss the possibility of using additional features and context data for improvements. Experimental results show that bidirectional GRU networks trained on word-level features, with Latent Topic Clustering modules, is the most accurate model scoring 0.805 F1.

Challenges and frontiers in abusive content detection

“…Many definitions pre-empt the effects of abusive language. For instance, Lee et al describe abusive language as 'any type of insult, vulgarity, or profanity that debases the target; it also can be anything that causes aggravation' (Lee, Yoon, & Jung, 2018). Similarly, in Wulczyn et al's dataset of over 100,000 Wikipedia comments, 'toxicity' is defined in relation to how likely it is to make individuals leave a discussion (Wulczyn et al, 2017).…”

“…This can lead to considerable degradations in the quality of datasets over time. For instance, Founta et al shared a dataset of 80,000 tweets but soon after this was reduced to 70,000 (Founta et al, 2018;Lee et al, 2018). This not only decreases the quantity of data, reducing variety, but also the class distribution changes.…”

Stop PropagHate at SemEval-2019 Tasks 5 and 6: Are abusive language classification results reproducible?

“…Nevertheless, the it remained not updated at the moment of the submission of this paper. Another work could also not replicate this results (Lee et al, 2018). After circumventing some of the aforementioned problems of the original code, we explained our specific version and used it to enter the shared task.…”

Izboljšanje zaznave sovražnega in zlonamernega govora s pomočjo slovarja besed

“…Zraven naštetim učnim algoritmomčlanka [5] v poglavju 2 smoše implementirali nevronsko mrežo MLP in tradicionalni model "Bagging"klasifikator. Nevronsko mrežo smo izbrali zaradi dobrih rezultatov v sorodnihčlankih.…”

“…Prav tako je do razlik v rezultatih enakih testiranih učnih metod kot včlanku [5], prišlo zaradi tega, ker nismo mogli pridobiti vseh tvitov iz baze, ker so upravljalci Twitterja med temže izbrisali nekatere zlonamerne in sovražne tvite.…”