A systematic review to identify the effects of tea by integrating an intelligence-based hybrid text mining and topic model

摘要

An emerging problem regarding tea and health has been pyramidally addressed as research concerns from limited literature reviews to identify an important and interesting challenge. Although past studies activated for various topics and diverse purposes of tea and health, a gap for using hybrid intelligence-based techniques to discover useful information from literature analysis exists in the exploration of curative effects on tea against fatal diseases other than Western medicine therapies. This study is motivated to bridge this gap by solving this research issue for healthcare applications between tea and health. Thus, this study proposes a hybrid method of an intelligent/objective text mining technique and topic modeling principally by latent Dirichlet allocation with VEM method and Gibbs sampling as along with measurements for three evaluation metrics for model performance from published articles. In the experiment materials, this study sets conditions to collect 2109 journal articles from the Web of Science from 2010 to 2017. We divided this into three datasets that each corresponded to the three periods to differentiate discrepancies in future trends. This study contributes eight beneficial directions as follows: (1) From a technical view, the figure of VEM's perplexity has a screen plot, but Gibbs sampling is smooth and good; and, interestingly, the greater the number of topics, the lower the perplexity is; (2) in empirical results, the terms for primary topics are tea and tea compounds, and secondary topics are associated with terms for issues regarding tea and health; (3) this study yields five research findings with key empirical evidence that tea has a natural and important preventive impact on treating diseases, especially cancer; (4) as to any managerial implications, early preventions and treatments by greater tea consumption as a valuable healthcare activity with medicinal purposes; (5) regarding the novelty of this research, this study fills the gap in a hybrid knowledge-based objective text mining and topic modeling technique than past researchers have as regards tea and health issues that were only based on traditional content analysis methods; (6) for this study's strengths, it achieved manpower cost reductions and relative objectivity because the objective LDA method is rarely used for topic modeling when compared to past studies; (7) for the research significance, the proposed method benefits efficient analysis from massive amounts of extant data for exploring latent information, accelerates research processes quickly, improves understanding for new hypotheses, and identifies key questions for further research; and (8) for conclusive research importance, this study offers new rationales for medical application and discovers differentiations and gaps for studying research trends.