Aspect and Opinion Extraction of Indonesian Lipsticks Product Reviews using Conditional Random Field (CRF)

Aspect and opinion extraction is a key process in several downstream applications, such as market analysis. In this paper, we modeled aspect and opinion extraction of lipsticks product reviews in Indonesia using Conditional Random Field (CRF) method. The dataset contains review text written not only in standard and colloquial Indonesian languages but also standard and colloquial English, labeled by BIO format notation. The experimental results show that this is a challenging task, described by the average F1 score on 10 aspect opinion labels is 44.1% with accuracy of 81.8%. The results on the baseline method, HMM, show lower F1 and accuracy scores. The errors found in the results have mostly occurred when the same words have different meanings. This error has a percentage of 75% of the total result. Second largest error caused by unknown words with the percentage of 14% of the total result.