Trendline

Figuring out what's going to be big in the future is hard.

Trendline is a tool that tries to predict future success of current tech by relying on the powers of machine learning and induction.

The code:

At the heart of it, Trendline uses:

1) The Python Reddit API Wrapper (PRAW) to query for submissions that relate to a topic

2) Pytrends API to gauge how interest in a topic is developing over time via Google Trends.

3) WikiBot to traverse Wikipedia pages.

4) US Patent Office data and Beautiful Soup to parse patent records.

5) TextBlob to perform sentiment analysis on selected text.

6) TensorFlow for the actual functional approximation.

Trendline diagram

Performance:

We ran a dense three layer neural net on a training set of 15 samples. (Yes, we're aware this is pitifully small.) On our test set of 11 samples, we reached an accuracy of ~81.8%.

Applications:

This is all heavily speculative, but the early results for Trendline look (perhaps) promising. Being able to leverage existing public datasets (Wikipedia, etc.) to create predictions is always exciting, especially considering our early success and the potential .

Future Tasks:

1) Acquire waaaay more data. (Additional features, more things for our model to consider, etc.)

2) Run more training iterations and update our model.

3) Test our model to see if generalization outside of the technology sector occurs.

Built With

Share this project:
×

Updates