Economic news using LSTM and GRU models for text summarization in deep learning

This study innovatively addresses the field of text summarization in Natural Language Processing, focusing specifically on the Thai language. In a departure from the limited existing text summarization models for Thai, the researcher employs an approach, leveraging the maximum matching algorithm and Thai Character Cluster (TCC) from the PyThaiNLP library (version 2.2.4) to group Thai words according to the dictionary. The experimental setup utilizes a model incorporating Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models, with a unique exploration of varying maximum input text sizes—namely, 150, 500, and 700 words. Real-world data from Thai economic news articles, sourced from ThaiSum, consisting of 2,000 articles in the Thai language, forms the basis of the study. The evaluation framework employs the ROUGE scores (ROUGE-1, ROUGE-2, ROUGE-L) to assess efficiency. In the performance analysis for LSTM, the LSTM model, with a maximum input text of 500 words, emerges as the standout performer, securing the highest ROUGE-1, ROUGE-2, and ROUGE-L scores. Notably, the LSTM model achieves the highest ROUGE-1 recall (R) at 20.7. For the part of GRU, the GRU model, with a maximum input text of 700 words, also demonstrates robust performance, attaining the highest ROUGE-1, ROUGE-2, and ROUGE-L scores, with the highest ROUGE-1 recall (R) at 26.1. This research marks an advancement in text summarization for the Thai language, introducing innovative methods and models that contribute to the sparse landscape of existing approaches. The integration of the maximum matching algorithm, TCC, and the LSTM and GRU models, coupled with the meticulous evaluation using real Thai economic news data, positions this study as a valuable and pioneering contribution in the domain of Natural Language Processing.