Human Values Estimation on Articles by Extracting and Concatenating of Opinion Expressions

Abstract

In this paper, we proposed a method for estimating the writer’s human values reflected in an editorial article. Suppose the editorial writers write each article based on one core value they consider the most important when composing editorials. The human values are reflected in the opinion sentences in the article. The proposed method extracts opinion sentences from an article, and then creates a concatenated text by combining opinion sentences in the article. The human value that appears most frequently in the concatenated text will be the estimated value of the whole article. An evaluation experiment was conducted using an editorial article corpus from the Mainichi Shimbun newspaper regarding the nuclear power plant after the Great East Japan Earthquake. Experimental results showed that the proposed method with extraction and concatenation of opinion sentences marked 87.5% accuracy in human values estimation, which was higher than the baseline accuracy of 62.5%.