Linguistic Patterns for Code Word Resilient Hate Speech Identification.

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Cybersecurity

Background:

• The rise of online activities has led to an increase in hate speech, necessitating advanced automatic detection methods.
• Current methods often rely on keyword dictionaries and supervised classification, which are ineffective against evolving extremist code words.
• Code words, with benign everyday meanings, are strategically used to evade detection, posing a significant challenge.

Purpose of the Study:

• To develop an improved automatic detection method for online hate speech.
• To address the challenge of code words with dual meanings used by extremist communities.
• To enhance the accuracy and generalizability of hate speech detection systems.

Main Methods:

• Introduction of a word embedding model to learn the hidden meanings of code words.
• Development of a classifier that utilizes linguistic patterns to mitigate the impact of individual words.
• Evaluation of the proposed method across three diverse datasets to ensure generalizability.

Main Results:

• The word embedding model successfully learned hidden hate speech meanings.
• The linguistic pattern-based classifier demonstrated superior performance compared to baseline methods.
• Empirical results confirmed the effectiveness and generalizability of the proposed approach.

Conclusions:

• The proposed method, leveraging word embeddings and linguistic patterns, effectively detects hate speech, even when code words are employed.
• This approach offers a more robust solution than traditional keyword-based methods for identifying nuanced online hate speech.
• The findings enable more comprehensive analysis of hate speech expressions in digital communication.