Analisis Sentimen Berbasis Aspek Pada Ulasan Video Game A Space for the Unbound Menggunakan Latent Dirichlet Allocation Dan Support Vector Machine

Gian Abisha Pratama Abidano, Dr. techn. Khabib Mustofa, S.Si., M.Kom.; Novera Istiqomah, S.T., M.T., Ph.D.

2026 | Skripsi | ILMU KOMPUTER

Perkembangan industri video game lokal Indonesia mencetak sejarah global melalui game naratif non-horor berjudul "A Space for the Unbound" karya Mojiken Studio yang berhasil menembus nominasi Games for Impact pada The Game Awards 2023. Kesuksesan ini menghasilkan akumulasi volume ulasan pemain global yang masif dan kaya akan opini campuran pada platform Steam. Oleh karena itu, penelitian ini melakukan analisis sentimen berbasis aspek guna memetakan persepsi dan kritik pengguna tersebut secara terstruktur.

Penelitian ini menerapkan metode hibrida Machine Learning untuk melakukan Analisis Sentimen Berbasis Aspek pada tingkat kalimat dari ulasan berbahasa Inggris game "A Space for the Unbound" dari platform Steam. Metode Latent Dirichlet Allocation (LDA) digunakan untuk mencari jumlah klaster topik optimal guna mengekstraksi aspek-aspek utama pada game. Selanjutnya, model Support Vector Machine (SVM) dengan Linear Kernel diimplementasikan untuk mengklasifikasikan sentimen positif dan negatif pada tiap aspek yang telah ditemukan. 

Analisis pada 9.889 kalimat ulasan menunjukkan karakteristik ketidakseimbangan kelas yang signifikan, dengan dominasi mutlak sentimen Positif sebanyak 9.372 kalimat (94,7%) berbanding 517 kalimat bersentimen Negatif (5,3%). Ketimpangan data ini selaras dengan tingginya apresiasi komunitas global di Steam yang memberikan predikat ulasan "Overwhelmingly Positive" (97%). Pengujian model hibrida LDA dan SVM menghasilkan nilai akurasi global sebesar 88,91%, dengan performa klasifikasi yang dipengaruhi oleh ketimpangan kelas, menghasilkan F1-score sebesar 0,94 untuk kelas positif dan 0,30 untuk kelas negatif. Penelitian ini berhasil menyimpulkan representasi sentimen pemain global secara otomatis serta menyediakan wawasan berbasis data mengenai kekuatan utama game, sekaligus mendeteksi area minor yang memerlukan optimalisasi bagi pengembang game.

The development of the Indonesian local video game industry has marked a global milestone through a non-horror narrative game titled "A Space for the Unbound" by Mojiken Studio, which successfully secured a nomination for Games for Impact at The Game Awards 2023. This success has generated a massive volume of player reviews on the Steam platform, yielding rich and mixed opinions from global players. Consequently, this study implements an aspect-based sentiment analysis approach to structurally map out user perceptions and criticisms.

This study applies a hybrid machine learning method to conduct Aspect-Based Sentiment Analysis at the sentence level on English user reviews of "A Space for the Unbound" extracted from the Steam platform. The Latent Dirichlet Allocation (LDA) method was utilized to determine the optimal number of topic clusters to extract the core aspects of the game. Furthermore, a Support Vector Machine (SVM) model with a Linear Kernel was implemented to classify positive and negative sentiments for each identified aspect.  

Analysis of 9,889 review sentences revealed a significant class imbalance characteristic, heavily dominated by the majority class with 9,372 positive sentences (94.7%) against 517 negative sentences (5.3%). This data disparity aligns with the high appreciation from the global Steam community, which rates the game as "Overwhelmingly Positive" (97%). Testing of the hybrid LDA-SVM model yielded a global accuracy of 88.91%, with the classification performance being inherently affected by the data imbalance, resulting in an F1-score of 0.94 for the positive class and 0.30 for the negative class. This research successfully concludes the automated representation of global player sentiments and provides data-driven insights regarding the game's core strengths, while simultaneously detecting minor areas that require optimization for game developers.

Kata Kunci : Sentimen, Aspek, Latent Dirichlet Allocation (LDA), Support Vector Machine (SVM), Ulasan Game, A Space for the Unbound