Team: 13
School: Eldorado High
Area of Science: Economics
Interim: Theo Waitkus
Nicholas Pasch
Tejas Kandath
Nathaniel Wolff
Supercomputing Challenge
9 December 2018
Problem Definition
Our project is making a sentiment analysis model that takes in text from news outlets about current events and business, and predicts how the stock market will behave. If we are successful in finding a significant correlation between stock market behavior and the sentiment as a result of news and commentary, we will have effectively demonstrated that markets are ruled by psychology and not conventional, objective indicators.
Problem Solution
We’ll be creating a neural network model on tensorflow and train it to predict stock prices given text from news and media sources about business and current events happening in the world. For example, when Elon Musk appeared on the Joe Rogan Podcast and was filmed smoking fat doinks, Tesla stock went down significantly as a result. However, to what degree is that a result of what Elon did? We’ll be trying to predict situations like this by using a sophisticated machine learning model. Sentiment analysis has been used before effectively on other problems, but based off of our research, we do not believe it has ever been integrated with another neural network for this application.
Current Progress
We have extensively researched the subject, and plan to start coding very soon. We have read multiple books on machine learning in python and tensorflow. We will be using python with tensorflow, for fast model training, on cloud servers. Tensorflow is an open source machine learning framework that trains on the cloud. We plan on writing and refining our code over the next month and then training our model. In addition, we have read various studies by renowned computer engineers like Andrej Karpathy who designed the recurrent neural network algorithm we plan to use for sentiment classification.
Expected Results
We aren’t quite sure what the results of our project will be as this study is designed to test the veracity in the behavioral economics approach in (as close to) an objective manner. Sentiment analysis is tried and tested, however we aren’t sure how strong the correlation between commentary and stocks evaluation is. Current algorithmic-trading bots completely dominate the day-trading scene (in the interval between seconds and minutes) however ours will find trends in the interval between days and weeks, giving it a chance to succeed if a correlation can be found.
Works Cited
“Aspect Extraction for Opinion Mining with a Deep Convolutional Neural Network.†NeuroImage, Academic Press, 17 June 2016, www.sciencedirect.com/science/article/abs/pii/S0950705116301721.
Liu, Bing. Video Game History, www.cs.uic.edu/~liub/FBS/sentiment-analysis.html.
“Sentiment Analysis Systems Case Study.†Paragon Poll, paragonpoll.com/sentiment-analysis-systems-case-study/.
“A Sentimental Education: Sentiment Analysis Using Subjectivity Bo Pang and Lillian Lee Proceedings of ACL, Pp. 271--278, 2004.†Object Recognition, www.cs.cornell.edu/home/llee/papers/cutsent.home.html.
“Machine Learning Algorithmsâ€.Packt Birmingham Mumbai. Giuseppe, Bonaccorso.July 2017.Packt Publishing
An Introduction to Biometric Recognition - IEEE Journals & Magazine, Wiley-IEEE Press, ieeexplore.ieee.org/document/7371033.
Team Members:
Nick Pasch
Tejas Kandath
William Waitkus
Nate Wolff
Sponsoring Teacher: Gary Bodman