Feature Importance on Text Analysis for a Novel Indonesian Movie Recommender System

With the increasing use of streaming services such as Netflix, Disney Plus, and Prime Video, a movie recommender system is an important tool for service providers and customers. However, there is no specific recommendation system for Indonesian films. Our research aims to improve the performance of the Indonesian film recommendation system with text analysis and feature importance. The first step is to get the Internet movie database (IMDb) Indonesian Movies dataset from Kaggle. We then conduct a text analysis that utilizes several stages, one of which is the bag of words (BoW) process. We apply feature importance to re-optimize the dataset. The following step compares three state-of-the-art classification models, SVM, naïve Bayes, and logistic regression, in predicting the Indonesian movie recommendation system. The metric we use to measure the performance of each model is the accuracy obtained by the cross-validation method. The test results show that out of 12 scenarios involving text analysis, feature Importance, and three prediction models for movie recommendation systems, the model with the best performance is naïve Bayes using text analysis and feature importance. The accuracy is 0.87 using K-fold cross-validation, with K utilized at 10.