Hello, my name is Alejandro, I am a PhD candidate in the Department of Political Science at the University of Michigan. Prior to this, I studied Political Science and Economics at Santa Clara University, where I graduated with honors (2013). After graduation, I attended the ICPSR Summer Program (2013) at University of Michigan, focusing on game theory. My current Work focuses on the development/validation of machine learning models for social science research.
Research into online political civility began in the mid 90's, but keeping up with the scale of content is difficult, given the variety of social media platforms today. I make things easier by speeding up the coding of politically relevant content. I train my computer to classify tweets, memes, and news content based on patterns in text and image data.
( 01 )
University of Michigan
PhD in Political Science (2021)
Current machine learning methods employ classifiers capable of automated text classification, but what if data takes dual forms, like text and image? I use neural networks to build a bimodal classifier: a machine-learned model trained on text and image data. I use this model to analyze Twitter data from the 2016 U.S. Presidential Election.
MS in Statistics (2020)
Coursework includes: maximum likelihood estimation, optimization algorithms, machine learning, and artificial neural networks. I have scraped and analyzed content from Wikipedia, Twitter, Facebook, New York Times.com, and FoxNews.com. The bulk of my work now resides in the field of automated bimodal content analysis.
Santa Clara University
BS in Political Science and Economics
My honors thesis examines Simone de Beauvoir's The Ethics of Ambiguity; I use her commentary on art and tragedy to derive an "existential aesthetic." I then use this framework to criticize the art of Andy Warhol.
Papers & Presentations
The Violent Aesthetic: A Philosophical Discussion of Neural Networks
This paper is concept art for computational social science. There are no benchmarks and plenty of sampling issues.
I offer an “aesthetic defense” of neural networks. This class of machine learning algorithms provide an elegant solution to the ugly data problem in mass media studies. I argue that machine learning applications, like facial recognition, deserve a normative discussion from social scientists.
Pixels as Data: ConvNets for Automated Image Analysis
Methods for automated content analysis focus exclusively on text while ignoring the visual cues that create additional, latent meanings. My paper addresses the issue of automated image analysis with specific focus on detecting racial cues in the media.
Stochastic Gradient Descent for Support Vector Machines
Stochastic gradient descent is a technique used to optimize a function with millions of parameters. The idea is to estimate the parameters that offer the lowest amount of error. We take a guess, find the error, and take a step towards reducing the error.