Github repo for Visual Framing of Science Conspiracy Videos: Integrating Machine Learning with Communication Theories to Study the Use of Color

Figure 2: Saturation comparison between conspiracy vs debunking videos

• to run t-test and generate boxplot, execute statistics test/t-test.R
• input: statistics test/latest_video_feature.csv

Table 1: Significant results on COVID-19 conspiracy-related videos: using first 10 seconds

• run t-test and calculate p-value: statistics test/t-test.R with input statistics test/latest_ten_seconds_features.csv

• run Benjamini-Hochberg test: statistics test/bh-test.R with input statistics test/latest_ten_seconds_features.csv

• calculate Cohen's d: statistics test/cohen.R with input statistics test/latest_ten_seconds_features.csv

Table 2: Significant results on COVID-19 conspiracy-related videos: using thumbnails

• run t-test and calculate p-value: statistics test/t-test.R with input statistics test/COVID19_thumbnails_low_aesthetics.xlsx

• run Benjamini-Hochberg test: statistics test/bh-test.R with input statistics test/COVID19_thumbnails_low_aesthetics.xlsx

• calculate Cohen's d: statistics test/cohen.R with input statistics test/COVID19_thumbnails_low_aesthetics.xlsx

Table 3: Significant results on climate change conspiracy-related videos

• run t-test and calculate p-value: statistics test/t-test.R with input statistics test/climate_feature.csv, statistics test/ten_seconds_featureclimate.csv, statistics test/Climatechange_thumbnails_low_aesthetics.csv,

• run Benjamini-Hochberg test: statistics test/bh-test.R with input statistics test/climate_feature.csv, statistics test/ten_seconds_featureclimate.csv, statistics test/Climatechange_thumbnails_low_aesthetics.csv,

• calculate Cohen's d: statistics test/cohen.R with input statistics test/climate_feature.csv, statistics test/ten_seconds_featureclimate.csv, statistics test/Climatechange_thumbnails_low_aesthetics.csv,

Table 4(a). Performance comparison in identifying conspiracy videos from correction videos.

You can find all the model setups and input data files under latest models/models-debunk.

• (1) .ipynb files are holders to train the models. Within each ipynb folder, you can find the corresponding models and their performance.
• (2) .h5 files are the output models.

Appendix IV. Correlation matrix of the visual features used in Model 4

Run statistics test/heatmap.ipynb to generate correlation matrix of the visual features with input latest models/handlabel_feature.csv.

Appendix V. Performance comparison in identifying conspiracy videos from normal videos.

You can find all the model setups and input data files under latest models/models-normal.

• (1) .ipynb files are holders to train the models. Within each ipynb folder, you can find the corresponding models and their performance.
• (2) .h5 files are the output models.