Latent Semantic Indexing (LSI) and Hierarchical Dirichlet Process (HDP) Models on News Data
Arif Ridho Lubis, Santi Prayudani, Yulia Fatmi, Okvi Nugroho
Abstract
News has become a very important need in modern society. Almost every level of society needs information such as news. Online news gets the attention of writers because there are a lot of characteristics of the data available and can be used for analysis so as to get insights and current trends, such as on 4 news portals that are most frequently accessed by the public such as detik.com, okezone.com, kompas.corn, and tribunnews.com. Related research in this field about Trend Analysis and Topic modeling on news media data in Indonesia then discusses news data collection methods and data collection techniques news pre-processing text, and 2 Topic Modeling algorithms applied in this research, namely Latent Semantic Indexing (LSI) and Hierarchical Dirichlet Process (HDP). In this study, the author has succeeded in extracting topics from news data from media in Indonesia. Various topics show the effectiveness of a learning model producing topics from the LSI and HDP models. This research has implemented natural language preprocessing (NLP) to process news data so that it can be processed and used as data from modeling a topic. This study finds topics based on the model of coherence and LSI and HDP contained in the news data with a total of 29,437. The accuracy obtained is 87% compared to the model from LDA-KNN which can determine the topic with an accuracy of 72%. Hence, the Latent Semantic Indexing (LSI) model and Hierarchical Dirichlet Process (HDP) have better accuracy than the LDA-KNN model.