참고문헌의 유무가 LDA를 사용한 토픽 모델링 결과에 미치는 영향

초록

This study tried to investigate the necessity of the reference removal procedure during the text pre-processing in topic modeling by comparing the topic modeling results using the text with reference and the text without reference. The use of full-text article for the topic modeling may cause extra work on removing the subsidiary part that is not directly related to the content of the article, because it is not able to selectively extract the certain part of the text that researcher wants to include in the analysis has not yet been developed. Therefore, manual work by the researcher is required to remove the subsidiary information and it takes a great amount of labor and time to perform the text pre-processing rather than the actual topic modeling analysis process. Thus, this study compared the topic modeling result of the text containing reference and of the text without reference. The research articles has been published from three major sports management and marketing journals from 2009 to 2018 were collected. JSM, ESMQ, and SMQ have been selected based on their reputation and status in the academic field. Five sport management/marketing experts were hired to evaluate the quality of the topic modeling result. The result indicated that the text containing reference produced a better topic modeling result than the text without reference. The relevance between the articles which were used to generate certain topic was higher when the text containing reference. It can be concluded that the quality of the topics is also high when the relevance of the articles is high. Based on this result, reference removal procedure is not necessary for the topic modeling using research articles. It can reduce a considerable amount of time and labor that the researcher needs to spend on pre-processing.

키워드