Predicting Person Perception

Students

2020
Engineering and Applied Sciences

Faculty

Assistant Professor of Psychology

Project Summary

The goal of my project was to predict how people rate others in terms of likeability. I did this using a few tools: word vectors, machine learning, and survey data. Word vectors are multi-dimensional vectors of numbers used to represent concepts. For example, the concepts “red” and “pink” have very similar word vectors, while “black” has a vector that is a bit different and “rock” has a vector much different from the previous three. Google had already used neural networks to create a massive dataset of these vectors, so I made use of that. Machine learning algorithms analyze existing data that has been classified and try to find a pattern, so that they can predict the classification of future data. In this case, I was trying to train the machine learning algorithms on the existing data of ratings of likeability for famous people combined with these famous people’s word vectors. I created a survey and had 30 participants rate 500 famous people on how much they liked them, or likeability. Then, I found the word vectors for these 500 famous people, combined them with the likeability rating, and trained the machine learning algorithms on the resulting data. The prediction performance was highly accurate.

 

I also compared this prediction with prediction using dimensions of likeability proposed in psychology studies: warmth and competence. I created a survey to determine warmth and competence ratings for the 500 famous people, and ran the machine learning algorithms on that. The prediction performance was quite similar, which shows that the method of word vectors gives strong predictions.

 

I ended up helping some of the graduate students working with Professor Bhatia with their projects as well, creating surveys for them and contributing ideas.

 

I learned many useful tools through my research experience. I got to know firsthand how research in the field of psychology is carried out. I got to be a part of meetings where we analyzed other research papers, and so I got a better idea of the overall field. I gained some machine learning experience and got to work with word vector embeddings, which are a new topic people are looking into. I learned how to create and conduct surveys online. Finally, I got to work with some very bright individuals who taught me to think critically.