Reddiment

• 

  Sentiment score in March 2018, all below 0.5 (negative)

• 

  Sentiment scores and the Crypto Market

Inspiration

We using the tools in Machine learning tools to build a sentiment analysis model based on text data from different blogs.

What it does

It accurately shows the sentiment score of the world in daily basis.

How I built it

We first clean the data from Reddit blogs, and train our model on text data, such Amazon reviews. Then we build the Machine Learning model and do cross validation on our hand labeled blog data in crypto market. The last step is to put this model on our platform that keeps updating news and blog text data to get sentiment outputs.

Challenges I ran into

Because all the blogs on Reddit are unlabeled, which means we cannot just train our ML model with these data. So we must hand label some of them to generate our own test data, and train our model with some other smartly chosen data sets.

Accomplishments that I'm proud of

We made this model works with 54 hours, and we wont the fist place in the Seattle Startup Weekend Blockchain edition in 2018.

What I learned

Don't try too many new stuffs in hackathon or startup weekend, and just focus on the minimum viable product first. Also the pitch and presentation is very crucial, since if audience cannot understand what we built, then all the efforts are useless.

What's next for Reddiment

We will make the API of our product online, and get some feedback from the market first.

Built With

• python-sk-learn
• python-turicreate
• sci-kit-learn
• shiny-r
• sql