Exploring methods for determining the appropriate number of topics in LDA: Focusing on perplexity and harmonic mean method

Discovering fashion industry trends in the online news by applying text mining and time series regression analysis

“…The news agendas change the attitudes of information recipients [ 35 ], determining the level of public trust in the company [ 18 ] and the direction of the business management [ 26 , 45 , 46 ]. The study findings support prior research that used LDA topic modeling analysis with media report data [ 5 , 52 , 57 , 58 , 144 ]. It also confirmed that the importance of issues in the fashion industry’s major agendas could be highlighted through media reports.…”

“…Topic modeling is an unsupervised learning algorithm that gathers keywords and documents with similar topics based on words in large text data, if the major topics have a probabilistic distribution [ 51 ]. LDA topic modeling uses a probabilistic algorithm suitable for analyzing large quantities of digital data in science research [ 52 ]. Specifically, assuming a Dirichlet distribution in which a topic hierarchy exists between words appearing in a document, the distribution of words in each document is probabilistically calculated to create a model.…”

“…Choi and Lee [ 56 ] examined the current state of research on ethical consumption in fashion journals and derived the issues of animal protection, distribution, and convergence with science and technology. Furthermore, several studies [ 5 , 52 , 55 , 57 , 58 ] have demonstrated media empowerment through news report analysis using an LDA algorithm to extract key topics.…”

The Analysis of Research Trends and Public Awareness of Smart Farms using Text Mining

“…If there are very few topics, the redundancy may decrease and they are easy to interpret, but it is difficult to derive various keywords (Greene et al, 2014). Accordingly, previous studies (Park and Lee, 2019;Kwon and Kim, 2021;Lee and Yi, 2021;Park, 2021;Park et al, 2022) are determining the number of topics through perplexity, topic coherence, silhouette coefficient, and expert decision making. This study finally decided on the number of topics considering the silhouette coefficient inherent in NetMiner 4.0 and the results of previous studies.…”