Advancements in Semantic Expansion Techniques for Short Text Classification and Hate Speech Detection
Ari Muzakir, Kusworo Adi, Retno Kusumaningrum · Ingénierie des systèmes d information · 2023
Traditional text classification methodologies, which primarily rely on document context and word frequency, often fall short in handling the linguistic complexities of the Indonesian language, such as colloquialism and informal language usage.This study presents a comprehensive semantic expansion-based framework to address these challenges in detecting hate speech within Indonesian social media commentary.Our framework leverages trusted knowledge bases, WordNet and Kateglo, to alleviate ambiguity in short texts.The BERT word insertion model is employed for semantic similarity calculation, followed by the application of a CNN deep learning model for hate speech classification.This approach effectively enhances semantic understanding and accurately classifies hate speech.The study also highlights future trajectories in semantic expansion for short text classification, encouraging further research to implement the proposed framework as an automated detection system.