쟁점 사안에 대한 핵심 어휘 판별 및 시각화를 통한 언론사별 논조 분석

Abstract

Media outlets regularly publish articles on the same issue using various tones that are distinct to each media company. To discover how one company’s tone is different from those of other outlets is presented in news articles, we designed a text analytics framework based on the weight scores of words used in politics and editorial sections from four major domestic newspaper companies. In our experiment, we selected five controversial political issues and collected related newspaper articles reported within a specified period. Then, we preprocessed these articles, such as tokenizing and part-of-speech tagging, an open-source Korean morpheme analyzer. The weights of the words are computed on the basis of the frequency-based CRED TF-IDF and scaled F-score. In addition, we constructed a neural network classifier to categorize the publisher of each article correctly on the basis of an attention mechanism to find highly contributive words for publisher discrimination. Lastly, we analyzed the differences in tones by visualizing keywords to provide an intuitive understanding.