Analisis Sentimen Berbasis Aspek Pada Ulasan Pelanggan Restoran Menggunakan Algoritma Support Vector Machine (Studi Kasus: Depot Bamara)

Abstract

Bahasa Indonesia

Ulasan pelanggan berperan penting dalam evaluasi produk dan layanan restoran. Ulasan pelanggan membantu restoran mengidentifikasi kekuatan dan kelemahannya. Depot Bamara adalah rumah makan yang belum efektif dalam memanfaatkan ulasan pelanggan karena kurangnya teknologi dan sumber daya manusia yang ahli dalam menerapkannya. Salah satu solusi yang dapat diimplementasikan adalah menggunakan analisis sentimen berbasis aspek dengan bantuan pembelajaran mesin. Analisis sentimen berbasis aspek memberikan pengetahuan yang lebih berfokus pada setiap aspek restoran, dan penggunaan pembelajaran mesin memungkinkan pemanfaatan ulasan dengan sumber daya manusia minimal. Penelitian menggunakan data ulasan pelanggan tahun 2021-2022 sebanyak 1029 ulasan melalui web scraping di situs Google Review. Kata-kata dalam ulasan diberi bobot menggunakan teknik pembobotan frequency inverse document frequency (TF-IDF) dan diklasifikasi menggunakan algoritma support vector machine (SVM). Model klasifikasi diuji dengan teknik k-fold cross validation, menghasilkan rata-rata accuracy sebesar 92%, menunjukkan performa yang baik. Pengujian juga dilakukan pada masing-masing aspek, menujukkan hasil klasifikasi pada aspek harga memiliki rata-rata accuracy tertinggi, sebesar 96%. Data ulasan dianalisis menggunakan root cause analysis (RCA). Hasil RCA menunjukkan bahwa akar masalah ulasan negatif adalah tidak adanya opsi kepedasan, SOP yang belum ada, pelatihan staf yang belum rutin, dan kurangnya ruang parkir dan penataan tempat duduk saat makan siang. Hasil RCA digunakan untuk membuat rekomendasi setelah berdiskusi dengan stakeholder. Abstract Customer reviews play an important role in the evaluation of restaurant products and services. Customer reviews can help restaurants identify their strengths and weaknesses. Depot Bamara is a restaurant that has not been effective in utilizing customer reviews due to the lack of technology and human resources who are experts in implementing it. One solution that can be implemented is using aspect-based sentiment analysis with the help of machine learning. Aspect-based sentiment analysis provides more focused knowledge on each aspect of the restaurant, and the use of machine learning allows the utilization of reviews with minimal human resources. The research used customer review data in the years 2021-2022 as many as 1029 reviews through web scraping on the Google Review site. Terms found in the reviews were weighted using the frequency inverse document frequency (TF-IDF) weighting technique and classified using the support vector machine (SVM) algorithm. The classification model was tested with k-fold cross validation, resulting in an average accuracy of 92%, indicating good performance. Tests were also carried out on each aspect, showing the classification results on the price aspect had the highest average accuracy, at 96%. The review data was analyzed using root cause analysis (RCA). The RCA results showed that the root causes of negative reviews were the absence of spiciness options, the lack of SOPs, the lack of regular staff training, and the lack of parking spaces and seating arrangements during lunch. The RCA results were used to make recommendations after discussions with stakeholders.

English

Ulasan pelanggan berperan penting dalam evaluasi produk dan layanan restoran. Ulasan pelanggan membantu restoran mengidentifikasi kekuatan dan kelemahannya. Depot Bamara adalah rumah makan yang belum efektif dalam memanfaatkan ulasan pelanggan karena kurangnya teknologi dan sumber daya manusia yang ahli dalam menerapkannya. Salah satu solusi yang dapat diimplementasikan adalah menggunakan analisis sentimen berbasis aspek dengan bantuan pembelajaran mesin. Analisis sentimen berbasis aspek memberikan pengetahuan yang lebih berfokus pada setiap aspek restoran, dan penggunaan pembelajaran mesin memungkinkan pemanfaatan ulasan dengan sumber daya manusia minimal. Penelitian menggunakan data ulasan pelanggan tahun 2021-2022 sebanyak 1029 ulasan melalui web scraping di situs Google Review. Kata-kata dalam ulasan diberi bobot menggunakan teknik pembobotan frequency inverse document frequency (TF-IDF) dan diklasifikasi menggunakan algoritma support vector machine (SVM). Model klasifikasi diuji dengan teknik k-fold cross validation, menghasilkan rata-rata accuracy sebesar 92%, menunjukkan performa yang baik. Pengujian juga dilakukan pada masing-masing aspek, menujukkan hasil klasifikasi pada aspek harga memiliki rata-rata accuracy tertinggi, sebesar 96%. Data ulasan dianalisis menggunakan root cause analysis (RCA). Hasil RCA menunjukkan bahwa akar masalah ulasan negatif adalah tidak adanya opsi kepedasan, SOP yang belum ada, pelatihan staf yang belum rutin, dan kurangnya ruang parkir dan penataan tempat duduk saat makan siang. Hasil RCA digunakan untuk membuat rekomendasi setelah berdiskusi dengan stakeholder. Abstract Customer reviews play an important role in the evaluation of restaurant products and services. Customer reviews can help restaurants identify their strengths and weaknesses. Depot Bamara is a restaurant that has not been effective in utilizing customer reviews due to the lack of technology and human resources who are experts in implementing it. One solution that can be implemented is using aspect-based sentiment analysis with the help of machine learning. Aspect-based sentiment analysis provides more focused knowledge on each aspect of the restaurant, and the use of machine learning allows the utilization of reviews with minimal human resources. The research used customer review data in the years 2021-2022 as many as 1029 reviews through web scraping on the Google Review site. Terms found in the reviews were weighted using the frequency inverse document frequency (TF-IDF) weighting technique and classified using the support vector machine (SVM) algorithm. The classification model was tested with k-fold cross validation, resulting in an average accuracy of 92%, indicating good performance. Tests were also carried out on each aspect, showing the classification results on the price aspect had the highest average accuracy, at 96%. The review data was analyzed using root cause analysis (RCA). The RCA results showed that the root causes of negative reviews were the absence of spiciness options, the lack of SOPs, the lack of regular staff training, and the lack of parking spaces and seating arrangements during lunch. The RCA results were used to make recommendations after discussions with stakeholders.