Penentuan Aspek Implisit dengan Ekstraksi Knowledge Berbasis Rule pada Ulasan Bahasa Indonesia (Determination of Implicit Aspects with Rule Based Knowledge Extraction in Indonesian Reviews)

Abstract

Determination of implicit aspects is one of the important things in opinion sentences. This study proposes a new approach for developing rule-based knowledge by forming a relation between opinion words with aspect categories. The relationship is obtained from the combination of rules, based on Opinion Word Similarity (OWS). Evaluation for rule-based knowledge extraction is in the form of threshold values of frequency and confidence to produce the best precision, recall, and f-measure values. The knowledge extraction consists of two phases: training phase and testing phase. The training phase is described as the process to extract rule-based knowledge. The testing phase is described as the process to obtain the implicit aspects of opinion sentences by referring to rule-based knowledge. To extract rule-based knowledge on user reviews, it is necessary to identify opinion sentences with explicit aspects and get pairs of aspects and words of opinion with rules generated from regular expressions. The evaluation res ult of rule-based knowledge with confidence using OWS showed better results compared to rule-based knowledge without using OWS. By using OWS, precision value increased by 0.25%, recall value increased by 1.15%, and precision value increased by 0.83%.