Yuda Septiawan, Juliansyah Adi Putra, Adimas Aglasia, Danang Ade Muktiawan, Meiliza

Deteksi Sentimen Ujaran Kebencian dalam Tweet Media Sosial X dalam Konteks RKUHP Menggunakan Algoritma KNN

Introduction

Deteksi sentimen ujaran kebencian dalam tweet media sosial x dalam konteks rkuhp menggunakan algoritma knn. Deteksi sentimen ujaran kebencian di tweet terkait RKUHP menggunakan algoritma KNN. Studi ini menganalisis 703 tweet berbahasa Indonesia, mencapai akurasi 56,74% dalam klasifikasi sentimen.

Abstract

Penyebaran ujaran kebencian di platform media sosial, terutama dalam konteks undang-undang kontroversial seperti Rancangan Kitab Undang-Undang Hukum Pidana (RKUHP), menimbulkan tantangan kritis dalam menjaga diskursus publik yang konstruktif. Studi ini bertujuan untuk mengidentifikasi dan mengklasifikasikan sentimen dalam tweet yang berkaitan dengan RKUHP menggunakan algoritma K-Nearest Neighbor (KNN). Sebuah dataset berisi 703 tweet dikumpulkan melalui API X dengan kata kunci “RKUHP”, mencerminkan puncak percakapan online pada Juni 2023. Setelah disaring untuk konten Bahasa Indonesia, tweet diproses melalui pembersihan (penghapusan tanda baca, tag, URL), normalisasi, dan terjemahan ke dalam Bahasa Inggris. Penandaan sentimen dilakukan menggunakan TextBlob, dan kemudian diverifikasi oleh ahli linguistik untuk meningkatkan akurasi label. Term Frequency–Inverse Document Frequency (TF–IDF) diterapkan untuk ekstraksi fitur, dan cosine similarity digunakan sebagai metrik jarak. Beberapa nilai K (3, 5, 7, 9) diuji, dengan K = 3 menghasilkan akurasi tertinggi sebesar 56,74%. Evaluasi menunjukkan bahwa KNN dapat mendeteksi sentimen terkait ujaran kebencian dalam tweet Indonesia secara moderat, meskipun masih ada keterbatasan dalam menangani sarkasme, netralitas, dan ambiguitas terjemahan. Hasil ini mendukung upaya untuk meningkatkan deteksi ujaran kebencian otomatis dengan mengintegrasikan embeddings kontekstual, mengatasi ketidakseimbangan kelas, dan memanfaatkan pembelajaran ensembel.

Review

This manuscript addresses a highly relevant and timely topic: the detection of hate speech sentiment on social media, specifically Twitter (now X), within the sensitive context of Indonesia's controversial RKUHP. The study's focus on maintaining constructive public discourse in the face of pervasive hate speech is commendable and aligns with critical needs in social computing and natural language processing. The authors propose using the K-Nearest Neighbor (KNN) algorithm to classify sentiment in RKUHP-related tweets, offering an initial exploration into automated hate speech detection in a challenging linguistic and socio-political environment. Methodologically, the paper outlines a clear approach, beginning with the collection of 703 Indonesian tweets via the X API during a peak discussion period related to RKUHP. The preprocessing steps, including cleaning, normalization, and importantly, translation to English, are detailed. The use of TextBlob for initial sentiment tagging, subsequently verified by linguistic experts, is a commendable step to enhance label accuracy and mitigate reliance solely on automated tools. Feature extraction using TF-IDF coupled with cosine similarity for distance calculation provides a standard framework for text classification, and the systematic testing of various K values demonstrates a structured experimental design. While the study provides a foundational effort, the reported accuracy of 56.74% for sentiment detection is moderate and indicates significant limitations that warrant further scrutiny. This performance level suggests that the current KNN model, particularly after translating nuanced Indonesian content to English for TextBlob processing, struggles considerably with the complexities of hate speech. The authors themselves acknowledge critical challenges such as sarcasm, handling neutrality, and the inherent ambiguities introduced by machine translation—all factors that could severely degrade performance in such a culturally and contextually rich domain. For this system to be practically viable in identifying hate speech, substantial improvements are necessary, and the suggested future directions, including the integration of contextual embeddings, addressing class imbalance, and leveraging ensemble learning, are crucial steps towards achieving a more robust and reliable detection capability.

Full Text

Comments

You need to be logged in to post a comment.