Hierarchical Topic Mining and Multi-label Classification on Online News in Bahasa
Ratih Nur Esti Anggraini, Hana Machmudah, Riyanarto Sarno · 2023
The proliferation of online news media has resulted in a surge of online news. Consequently, properly categorizing news based on their respective topics becomes crucial, enabling readers to analyze and comprehend the information more easily. In this study, the researchers conducted topic mining using the Joint Spherical Tree and Text Embedding (JOSH) method, a technique that utilizes spherical space for topic modeling. This method involves embedding a tree structure into the spherical space, with each category being surrounded by its representative terms in the same direction within that space. The resulting topic hierarchy was then employed for multi-label classification of Bahasa news texts acquired through web scraping from various news sites. The most successful topic mining outcomes exhibited a topic coherence value of 0.7882. Subsequently, a Long Short Term Memory (LSTM) model was developed for multi-label classification, yielding good results, including a hamming loss value of 0.9229 and a recall score of 0.9982.