Detecting Evergreen Articles: Comparing Word Embeddings for Hybrid Bidirectional GRU and LSTM Model on Indonesian News Portal

Faza, Ahmad (2025) Detecting Evergreen Articles: Comparing Word Embeddings for Hybrid Bidirectional GRU and LSTM Model on Indonesian News Portal. In: 2025 5th International Conference on Electronic and Electrical Engineering and Intelligent System (ICE3IS), 06-07 August 2025, Yogyakarya, Indonesia.

Text [Ahmad Faza - SI] Detecting Evergreen Articles Comparing Word Embeddings for Hybrid Bidirectional GRU and LSTM Model on Indonesian News Portal.pdf Download (2MB)

Abstract

Automatically detecting evergreen articles in Indonesian news media is a challenge and vital task for sustaining long-term SEO and audience engagement in media industry. This research propose a hybrid parallel deep learning model that hybrids bidirectional GRU and LSTM layers with FastText embeddings, and compare its performance against other word embedding methods, CNN-LSTM and IndoBERT. The experiments utilize a balanced dataset of 5,810 manually labeled article titles from Kompas.com, which underwent preprocessing via case folding, punctuation and number removal, and stemming. Three word embedding techniques, TFIDF, Word2Vec, and FastText, are employed, with each model trained under consistent settings. The results shows TF-IDF and Word2Vec yield acceptable performance, FastText enhances model robustness. The hybrid BiGRU-BiLSTM model attains a test accuracy of 92.51% with low loss, although IndoBERT attains superior performance with 95.09% accuracy. These findings highlight the effectiveness of word embeddings and hybrid architectures for evergreen article detection. Future work could incorporate attention mechanisms and ensemble methods to enhance classification, offering a promising direction for automated news analysis and supporting strategic decision-making in media organizations.

Item Type:	Conference or Workshop Item (Paper)
Creators:	Faza, Ahmad
Contributors:
Keywords:	CNN, Evergreen Articles, IndoBERT, RNN, Word Embeddings
Subjects:	000 Computer Science, Information and General Works > 000 Computer Science, Knowledge and Systems > 006 Special Computer Methods
Sustainable Development Goals:	Goal 16. Promote peaceful and inclusive societies for sustainable development, provide access to justice for all and build effective, accountable and inclusive institutions at all levels Goal 09. Build resilient infrastructure, promote inclusive and sustainable industrialization and foster innovation
Divisions:	Faculty of Engineering & Informatics > Information System
Date Deposited:	06 Oct 2026 09:42
URI:	https://kc.umn.ac.id/id/eprint/49651

Actions (login required)

View Item