StockSentimentAnalysis

HACK TCNJ

Garrett Beatty, Taylor He, and Kwong Lee

OBJECTIVE:

Make some kind of software that predicts aggregate stock data by analyzing political news sources and their sentiments.

HOW IT WORKS:

1) Mined twitter posts from @realDonaldTrump using twitter’s API

2) Mined Bloomberg’s stock market API to get S&P500 data

3) Used NLP - TextBlob API - to generate a sentiment from text

4) Factor retweets and favorites to generate weighted sentiments

5) Used a LinearRegression model to predict the effect of trump’s next statement

6) Used the past 7 days’ tweets as an indicator of prediction:
over 100000 simulations, 69.4% accuracy for predicting 7 days

OUTLINE:

Software:

Use Bloomberg’s API to get stock data from a time range

Use IBM’s AlchemyData News API to get news from a time range

Connect stock data with political news with python

Use ML algorithm [Linear Regression] to predict stock behavior based on news articles

Electrical:

Design LED structure that displays predicted stock behavior (positive or negative)

Use of Processing 3.3 to read data and transfer to Arduino

Programmable scrolling LCD screen that displays stock stats using Arduino Uno

The LEDs are LIT.

THINGS WE DID WELL:

Identified good APIs to use

Quick decision making

Acceptable accuracy

Work ethic in the first 5 hours

Fast at gathering and organizing data from API

Mutual agreement of laziness

Kwong learned a lot (False)

CHALLENGES:

Importing files to the Arduino

Getting both stock and csv data to line up

Too much salt in chinese food (More than rassbaby for comparison)

Doing the project

Consistently doing work

Lack of motivation

Finding enough relevant memes