Departmental Bulletin Paper 新聞記事における偏向性の定量評価

小谷, 龍ノ介

It is necessary to select from a vast amount of information in order to obtain highly reliable information from the web media. Considering the objectivity, accuracy, fairness, etc. of the information to be selected, understanding of the bias of the media is required. In various media, it is difficult to understand at a glance how difference the bias is, and it is difficult to receive comprehensively and accurately and fairly information. In thisstudy, focusing on Web news, we aim to quantitatively visualize the bias of each newspaper company. In order to visualize the bias, "characteristic words" which can interpret the sentiment are extracted, and the way of handling each newspaper company for them is quantified. As the quantitative index, we used the sentiment value by the evaluation sentiment dictionary that takes into consideration the dependency of the characteristic word and the probability belonging to the newspaper publisher of each characteristic word by the topic model andlogistic regression. In the probability that each characteristic word belongs to a newspaper publisher, it is possible to quantitatively show the bias of a newspaper article that makes it easy to visually understand the relation of each newspaper publisher and each haracteristic word by hierarchical clustering.Key Words : Natural language processing, Topic model, Newspaper

Number of accesses :  

Other information