Emotions expressed through twitter might help to predict how people think a about an asset.
What it does
The model takes twits sentiment analysis and historical prices into consideration to predict whether the price goes up or down in the next day.
How I built it
We use python to get time series data from Thomson Reuters and sentiment analysis score from twitter. The matlab code solve the SVM model and give the prediction.
Challenges I ran into
Twitter no longer allow public to access their twits which are older than 7 days through APIs. It is extremely hard to gather enough data for training purpose since twitter forbid other data set providers from uploading it too. The best way we could come up with is to use the sentiment analysis results from Quantopion. The data we got are one-dimension which meant we can neither apply our own sentiment analysis nor get richer-in-dimension tone analysis provided by Bluemix. It might have an impact on the performance of the model. However, with proper streaming APIs, this model could help investor better understand the market trends.
Accomplishments that I'm proud of
For to apply sentiment analysis, we learn to use API and coded a two-hidden-layers neural network for literally the first time. As non-CS major students, it was a very exciting thing to do. Most importantly, we make it till the end.
What I learned
A LOT! First time using Postman, APIs, and neural network. The intensive coding experience itself is extremely amazing!
What's next for Twit&StockPrices
Try to get a better way to get data resource and improve the model.