ANALISIS SENTIMEN BERBASIS ASPEK MENGGUNAKAN SUPERVISED LEARNING PADA SOSIAL MEDIA E-GOVERNMENT (STUDI KASUS DIREKTORAT JENDERAL IMIGRASI KEMENTERIAN HUKUM DAN HAM REPUBLIK INDONESIA)
Fathiyarizq Mahendra Putra, Dr, Sigit Priyanta, S.Si., M.Kom
2025 | Tesis | S2 Ilmu Komputer
The government in the central sector, ministries/institutions and local governments, create social media accounts to be able to get closer to citizens, as well as respond to them. social media accounts to be able to get closer to citizens, as well as respond to public input and complaints. input and complaints from the public, but it is quite difficult to understand the substance and aspects of opinions about services provided by netizens, in accordance with the performance and aspects of performance and service or just another opinion. just another opinion. This research proposes an algorithm model for sentiment analysis and opinion aspects in assessing the policies and services of the Directorate General of Immigration of the Ministry of Law and Human Rights of the Republic of Indonesia through user opinions on Twitter. The focus of the research is to develop the best model from several supervised learning algorithms for aspect and sentiment classification using a dataset of tweets from service users and public opinion.
The design of this research consists of data collection and data filtering, data preprocessing, Word Embedding, and data splitting using K-Fold Cross Validation. In collecting and labeling there is an imbalance of data between labels so SMOTE is needed to handle this problem, classification stages with supervised learning algorithms, and evaluation testing. Results demonstrate that the Word2Vec technique, particularly Skip-gram, outperformed CBOW. For the "service" aspect, Random Forest (Skip-gram) achieved 98.36?curacy, 99.79% precision, 97.78% recall, and 98.78?-Score. K-Nearest Neighbor yielded 62.167?curacy, 62.051% precision, 98.641% recall, and 76.18?-Score. The "policy" aspect, analyzed using SVM, generated 77.417?curacy, with interpretation limitations due to data imbalance.
The study underscores the critical importance of preprocessing techniques like stemming, data balancing through SMOTE, and feature extraction using Skip-gram in enhancing classification model performance on unstructured social media datasets, particularly for public service opinion analysis.
Kata Kunci : Analisis Sentimen, aspect-based sentiment, machine learning, supervised learning