Exploring impacts of media characteristics on message content using text mining

Abstract

Background/Objectives: The purpose of this study is to empirically investigate the semantic similarity of documents posted on different forms of media about specific social issues.

Methods/Statistical analysis: Online text data were collected from personal blogs and Internet news published on a major Korean portal site, NAVER. To collect text data from online media, the study used R programming language for web crawling. We examined what effects medium characteristics had on the content of conveyed messages by using a keyword extraction method based on TF-IDF, which is a text mining method, and the cosine similarity measurement method.

Findings: The results of this study demonstrate that there were differences in the major keywords extracted from messages conveyed by the three forms of media, but the similarity between keyword-to-keyword matrices extracted from the media was confirmed by a Mantel test, and there were statistically significant degrees of similarity among these matrices. We were therefore able to discover similarities of message content conveyed by each medium.

Improvements/Applications: For this study, we used only blog and news data published on a single Korean portal site. The text data better be collected from variety of channels in future studies.