PENERAPAN ALGORITMA MACHINE LEARNING CONDITIONAL RANDOM FIELD DAN MAXIMUM ENTROPY PADA KATEGORISASI ASPEK TEMPAT WISATA DI BANDUNG

• Main Author: Ibnu Sulistiyono, Balya
• Format: Final Project
• Language: Indonesia
• Subjects: Teknologi
• Online Access: https://digilib.itb.ac.id/gdl/view/64895
• Institution: Institut Teknologi Bandung

Bandung is a city with tons of attractive tourism. Visitor review regarding a tourist destination can be used as a feedback for tourist destination management. These reviews may include many aspects. ABSA or aspect-based sentiment analysis is a way to determine the polarity of an opinion in feature/aspect level. One of the important task in ABSA is aspect extraction and categorization. Therefore, a model that may perform this task is needed, along with its most optimal hyperparameter. For aspect extraction with default hyperparameter, the highest F1-score average is achieved by averaged perceptron algorithm (avg. F1=0.4743) and the lowest is reached by adaptive regularization of weighted vector (avg. F1=0.3727). Moreover, the fastest fitting and scoring time average is achieved by adaptive regularization of weighted vector (avg. time=3.3518) and the slowest is achieved by L-BFGS algorithm (avg. time=4.4818). For aspect categorization with default hyperparameter, the highest F1-score average is achieved by SAG solver (avg. F1=0.5210) and the lowest is reached by Newton-CG and L-BFGS (avg. F1=0.4270). Moreover, the fastest fitting and scoring time average is achieved by SAGA (avg. time=2.4693) and the slowest is achieved by SAG (avg. time=5.3123). For aspect extraction with default hyperparameter, the highest F1-score average is achieved by algorithm=l2sgd, c2=0.1, calibration_eta=2=0.2, calibration_rate=1.5, max_iteration=2000 (avg. F1= 0.50098). Moreover, the fastest fitting and scoring time average is achieved by algorithm=arow, gamma=0.01, max_iteration=50, variance=0.1 (avg. time= 2.529). For aspect categorization with default hyperparameter, the highest F1-score average is achieved by solver=Liblinear, penalty=l1, C=1000, max_iteration=200 (avg. F1= 0.66238). Moreover, the fastest fitting and scoring time average is achieved by solver=Liblinear, penalty=l1, C=1000, mac_iteration=150 (avg. time= 3.215).