Published on July 12, 2019.
As a continuation of the previous chapter, this passage argues that an artificial intelligence system trained on National Diet proceedings quantified the uniformity of Japanese Communist Party lawmakers and the diversity of Democratic Party lawmakers, and in that sense even accurately anticipated the later split of the Democratic Party.
It also introduces an attempt to apply the same method to newspaper editorials in order to make visible the political bias embedded in Japan’s public discourse.

2019-07-12
Conversely, the fact that the opinions of lawmakers belonging to the Democratic Party were diverse can also be said to mean that it was a party prone to splitting, and that prediction proved strikingly accurate in the split of the Democratic Party.
What follows is a continuation of the previous chapter.　
When the classification system obtained by machine-learning the proceedings of the National Diet from 1999 to 2008 was made to determine, in a five-way choice, which party each lawmaker belonged to—the Liberal Democratic Party, Komeito, the Democratic Party (as it then was), the Social Democratic Party, or the Japanese Communist Party—the highest rate of correct answers was for the Japanese Communist Party at 93%, while conversely the lowest rate of correct answers was for the Democratic Party at 65%.　
In other words, it can be interpreted that because the opinions of lawmakers in the Japanese Communist Party were the most uniform, the artificial intelligence found it easiest to identify them as “Japanese Communist Party,” whereas because the opinions of lawmakers in the Democratic Party were the most diverse and their common points were difficult to discern, the rate of correct answers also fell.
The fact that the opinions of lawmakers belonging to the Japanese Communist Party were uniform can also be said to mean that that party had the strongest tendency toward groupthink.
Conversely, the fact that the opinions of lawmakers belonging to the Democratic Party were diverse can also be said to mean that it was a party prone to splitting, and that prediction proved strikingly accurate in the split of the Democratic Party.　
The Liberal Democratic Party had the next-lowest correct answer rate after the Democratic Party, at 70%.
Based on this objective data, the criticism that the Liberal Democratic Party is characterized by groupthink is not correct.　
Based on these research results, I considered whether this same system might be able to measure the political bias of discourse in society.
That is because it can numerically indicate which party’s lawmakers’ statements various kinds of discourse most closely resemble.
The first thing undertaken was an application to newspaper editorials.
When the five newspapers—the Asahi Shimbun, Mainichi Shimbun, Nihon Keizai Shimbun, Yomiuri Shimbun, and Sankei Shimbun—were examined, the result was a natural one in line with general perceptions: the Asahi Shimbun had the highest degree of similarity to the three opposition parties of the time, namely the Democratic Party, the Social Democratic Party, and the Japanese Communist Party, while conversely the Sankei Shimbun had the lowest.
However, in all five newspapers, the degree of similarity to the opposition parties exceeded that to the ruling parties.
Since newspaper editorials are basically often written from a critical stance toward something, it is thought that such a result was obtained.
This article will continue.