News provided by mass media is required its accuracy, fairness and objectiveness. However, it is often biased by “hidden” thought of each media. In this paper, we challenged to detect such biases among newspaper articles provided three Japanese publishers; Asahi, Mainichi and Yomiuri newspapers. We collected online articles of those papers and Wikipedia as a supplemental data, and (1) divided them into set of each single words using morphological analysis, (2) obtained vector representation of each word by the word2vec methodology, (3) adjusted vector orientation based on the value registered on the Semantic orientation dictionary, and (4) compressed the vector dimension with t-SNE (t-distributed Stochastic Neighbor Embedding) and PCA(Principal Component Analysis) for data visualization. We objectively confirmed from our results that usage of words in news articles has not small variety among media.