Results
Results
Embedded Files
Figure 1: Heatmap
Figure 1: Heatmap
A simple heatmap provides an illustration of the rates of occurrence of rumorous, non-rumorous, and opinionated content for different keywords.
Fig. 1: A heatmap of posts on LINE that contained any of six political keywords, categorized by truth label. Posts containing one or more instance of either the English and/or Chinese version of the keyword were selected.
For the keyword Lai, the last name of the candidate for the left-leaning Democratic People's Party, the volume of content is much higher than for other terms. Worth considering is that the volume of content matching these categories that contains the keyword Hou, the surname of the China-friendly party's candidate, is nearly insignificant.
Figure 2: Evolution of Volume of Keyword-Containing Posts
Figure 2: Evolution of Volume of Keyword-Containing Posts
Time-series graphs, using a rolling average window of 45 days, illustrate the evolving incidence of keyword-containing posts created between the initialization of the dataset and the final date of the dataset.
Fig. 2: The number of posts for each term are plotted across time. In order, the graphs displayed are for [environment, president, KMT, Hou, Lai, illegitimate]. Each curve on a graph describes a different truth-type for a single keyword. Three of the control terms exhibit noise with a consistent spike between 2021 and 2022. The surname keyword Hou experiences a large increase in traffic until election day (Jan. 13, 2024), and immediately drops off.
Whereas the control terms exhibit generic noise, the growth in volume of content displaying the terms "Lai" and "illegitimate" display similar patterns; both are relatively unused until the period immediately following his election win. To further explore this connection, Figure 3 displays a comparison of the rates of co-incidence of the two terms.
Figure 3: Co-Incidence of "Lai" and "illegitimate"
Figure 3: Co-Incidence of "Lai" and "illegitimate"
A time-series graph illustrates the incidence of keyword-containing posts from the initialization of the dataset until the final date of the dataset. A rolling average window of 14 days is used.
Fig. 3: The number of posts matching each criterion is mapped to an according curve across time. The behaviors of the curve for the solo keyword "illegitimate" and the aggregate curve for the two terms "Lai" and "illegitimate" follow in very close step with each other during the period after Lai's election win.
The nearly identical behaviors of the curves mentioned above imply that posts including the solo keyword "illegitimate" are likely to include the keyword "Lai" as well.
These results would be best explored in several more exhibits. I apologize for the brevity and am excited to continue researching the subject with various levels of laxity over the summer and into my M.S. year here at UCSC.
Page updated
Report abuse