Thai Sentence Core Graph for Thai NLP
Nuchjarin Krukaset, Watcharapong Krukaset, Chouvalit Khancome · 2024
This research introduces a new data structure called “Thai Sentence Core Graph (TSCG).,” designed using a directional graph to serve as the core for managing Thai sentence patterns in natural language processing. The structure is specifically designed to handle tasks such as word segmentation, sentence segmentation, sentence generation, and checking sentence structure patterns according to Thai grammar rules. The graph is constructed through a thorough analysis of the Thai language, allowing it to summarize 20 fundamental sentence patterns. The theoretical resulting graph is simple, flexible, and efficient comprising only 6 nodes. The TSCG data structure can be applied to various tasks in Thai natural language processing, including word segmentation, sentence segmentation, sentence generation, and checking sentence patterns based on grammar rules. The performance evaluation through the development of computer programs for comparing the efficiency of Thai word segmentation, sentence segmentation, and sentence generation algorithms, designed by the research team, revealed that this new approach can achieve 100%accuracy in both word and sentence segmentation, equivalent to the performance of the existing PyThaiNLP library. Furthermore, it demonstrated superior handling of spaces and punctuation marks, especially when compared to the conventional methods employed by PyThaiNLP. Notably, when assessing the performance of Thai sentence generation using the TSCG, it outperformed the traditional N-gram model used in PyThaiNLP, with significant improvements in accuracy for sentences of lengths 2, 3, and 4 words/phrases, reaching accuracies of 53.33%., 34.67%, and 28.00%, respectively.