Ekstraksi Aspek untuk Pemilihan Topik Spesifikreview Film Menggunakan Rule-Based Aspect Extraction

Abstract

The growth rate in the usage of subscription video-on-demand platform in Indonesia estimated to reach 26.5%, around 13 million users in 2020. This growth is utilized by media streaming platform service providers to attract potential buyers/consumers. Each streaming media platform shows the advantages they have to attract consumers, one way is by showing a variety of exclusive films or tv shows that are only available on that platform. But that alone is not enough, as many as 93% of consumers say that reviews influence their decision to buy a product. Review text is a text about the author's judgment and opinion that contains sentiments and aspects of a product/service. With the large number of product choices and reviews that continue to increase, results in wasting a lot of potential customer’s time to read reviews and make decisions in buying a product. Long review can be summarized into a brief review consisting of aspects (opinion’s targets) and sentiments obtained from the opinion sentences in the review. The purpose of this study is to produce aspects, categories of these aspects and the sentiment of each reviews to make it easier for potential customers to make decisions in buying a product. The aspect extraction method is carried out by utilizing opinion lexicon and the rules of grammatical structure from Universal Dependencies resulting in 48438 pairs of opinions and aspects. Aspect categorization using the K-Means and Gaussian Mixture Model algorithm, both produces 13 aspect categories. Sentiment Analysis using TextBlob produces 68% accuracy, meanwhile VADER produces 98% accuracy.