Data comes from FakeNewsNet and was collected by the fact-checker Politifact.
Project files:
-
classifier-politifact.ipynb
- Models predicting whether in article is fake or real news. Best models achieve 90% accuracy and f1-score.
-
project1.Rmd
- unigram/bigram analysis of fake and real news articles, correlation analyses, sentiment analysis
-
project1.html
- knitted output file from project1.Rmd
-
politifact.ipynb
- script that provides summary statistics on real and fake news
- i.e average article length, average number of sentences, unique words, etc.
Data comes from Chicago Data Portal:
Project files:
-
project2.Rmd
- descriptive crime reported statistics from 2018, spatial analysis of reported crime / arrests and demographics, 311 lights out analysis and its relationship with reported crime rates
-
project2.html
- knitted output file from project2.Rmd
India, 2011
Data comes from
Project files:
-
project3.Rmd
- maps showing India Census demographics, night lights, and GDP by districts. Regression models to examine to correlation between GDP and night lights in a district.
-
project3.html
- knitted output file from project3.Rmd
Most of the research on misinformation focuses on English, and very little attention has been put to other languages, such as Spanish which is the third most spoken language in the world. This project explores misinformation and fake news in Spanish and Portuguese speaking countries.
We prepare the text in two fashions: one, keeping the text in the original language and running the textual analysis, and two, translating the text to English then running the analysis. Through this, we are able to examine how much of the narrative is βlost in translation.β
We found similar results with topic modeling, however disparate results with regard to sentiment scores, highlighting the care researchers should take with both translating texts and utilizing sentiment analysis.
Project files include the code and data I used. I prepared the text data and did analysis on the Spanish text. Spanish-speaking team members validated the results.