Project description:

The visit of U.S. House of Representatives Speaker Nancy Pelosi to Taiwan resulted in China conducting a military drill to intimidate the Taiwanese government. In addition to the military actions, we also found that there were many suspicious accounts on Facebook and Twitter spreading pro-China and anti-U.S. sentiments in an attempt to influence the Taiwanese public. However, through analyzing the number of shares and likes, it could be seen that pro-Taiwan and pro-U.S. sentiments were gaining more support and had a higher level of community engagement.

Impact reached:

China often uses information warfare to divide the Taiwanese public, and during Speaker Pelosi’s visit to Taiwan, China has launched another wave of information attacks. We have collected 830000 posts from Facebook and Twitter and analyzed which categories these posts belong to, such as support for China, support for Taiwan, support for the United States, etc.
After classifying these posts, we can see that some suspicious accounts are spreading the same message in many communities, which are suspected to be operated by China. By analyzing the behavior of these suspicious accounts, readers can be alerted when they see related information, so as not to be easily divided.

Techniques/technologies used:

We used Crowdtangle and Twitter API to collect 830000 posts related to Perlozzi, and then performed topic modeling analysis. We then manually recognized the themes based on the machine classification results, and classified all posts according to the theme keywords. Then, using R language analysis, we analyzed the spread and impact of these posts based on the classified data.

Context about the project:

Due to the vast amount of data, we used new techniques to perform topic modeling analysis in an attempt to classify these posts. However, even after machine learning, we still had to spend a lot of time manually confirming the accuracy of theme classification. At the same time, we had to race against time, which was our biggest challenge. Nevertheless, based on the experience of machine learning and analyzing text in the past, we were still able to successfully produce a report and analyze how suspected individuals manipulated public opinion.

What can other journalists learn from this project?

Taiwan has long faced information warfare attacks. In addition to analyzing the content of public opinion, this time we also attempted to find out where these opinions ferment in communities, as well as their level of influence and interaction. This will serve as reference data for future information warfare attacks.

Project links:

https://www.readr.tw/post/2912