Detail Cantuman

Analisis sentimen merupakan salah satu cabang penting dalam Natural Language Processing (NLP) untuk mengidentifikasi opini dan emosi dalam teks. Penelitian ini membahas penerapan Multi-Channel Convolutional Neural Networks (MCCNN) dalam analisis sentimen ulasan buku horor berbahasa Indonesia dari platform Goodreads. Dataset terdiri dari 3.333 ulasan yang kemudian disaring menjadi 3.041 setelah validasi anotasi menggunakan Cohen’s Kappa. MC-CNN dirancang dengan tiga channel input: word embedding, character-level, dan bigram word, sehingga dapat menangkap makna semantik maupun pola struktural teks. Kinerja MC-CNN dibandingkan dengan Single-Channel CNN (SC-CNN), Support Vector Machine (SVM), dan Naive Bayes menggunakan metrik accuracy, precision, recall, dan F1-score. Hasil menunjukkan bahwa MC-CNN memperoleh accuracy rata-rata 78,02% dan F1-score 0,7768 melalui cross-validation, lebih unggul dibandingkan SVM (76,10%), SC-CNN (74-75%), dan Naive Bayes (73,95%). Temuan ini menegaskan keandalan MC-CNN dalam menangani teks ulasan dengan konteks ambigu, khususnya pada genre horor di mana kata-kata berkonotasi negatif dapat bermakna positif. Penelitian ini berkontribusi pada pengembangan metode analisis sentimen berbasis deep learning dalam bahasa Indonesia dan membuka peluang untuk pengembangan lebih lanjut menggunakan model dengan mekanisme attention atau arsitektur Transformer.

Sentiment analysis is an essential branch of Natural Language Processing (NLP) aimed at identifying opinions and emotions in text. This study applies a Multi-Channel Convolutional Neural Network (MC-CNN) to sentiment analysis of Indonesian horror book reviews collected from Goodreads. The dataset consists of 3,333 reviews, refined to 3,041 after annotation validation using Cohen’s Kappa. The MC-CNN model integrates three input channels, word embedding, characterlevel, and bigram word, to capture both semantic and structural aspects of the text. Its performance was compared to Single-Channel CNN (SC-CNN), Support Vector Machine (SVM), and Naive Bayes, evaluated with accuracy, precision, recall, and F1-score. Results indicate that MC-CNN achieved an average accuracy of 78.02% and F1-score of 0.7768 through cross-validation, outperforming SVM (76.10%), SC-CNN (74–75%), and Naive Bayes (73.95%). These findings highlight the robustness of MC-CNN in handling ambiguous review contexts, especially in the horror genre where negative expressions may convey positive experiences. This study contributes to advancing deep learning-based sentiment analysis for the Indonesian language and provides opportunities for future research involving attention mechanisms or Transformer architectures.