Crypto Sentiment Analysis

Inspiration

We were inspired to combine our interest in blockchain, investing, and cryptocurrency with our eagerness to dive into the field of data science through this project.

What it does

The program asks for a cryptocurrency, from there using the Twitter API, Twitter uses the 100 most recent tweets regarding the subject and from there a “magnitude” value is generated by each tweet, which then in turn impacts an overall value known as connotation positively or negatively depending on how well or poor a subject is talked about. The magnitude is calculated by a combination of the value of a tweet which includes its metrics such as likes and retweets, and the user who tweeted the tweet’s value, which accounts for if the user is verified and/or has over 100,000 followers. From there, the sentiment is analyzed through a complex analysis of different keywords involved in a particular tweet to see if the tweet is processed as positive, neutral, or negative. This in turn either increases or decreases the connotation.

How we built it

The tools used in creating this project were many tools available in the field of Data Science, including Jupyter Notebook, the Visual Studio Code Editor, and the Twitter API. The primary libraries we used included Tweepy (pronounced Twee - Pie) for directly communicating with the Twitter API, and Textblob for processing the textual data and conducting sentiment analysis.

Challenges we ran into

Due to the strict 36 hour time constraint and external factors, time was a huge restraint thus our connotation algorithm, although mostly accurate, can be EXTREMELY volatile and spike randomly. This is due to the constraint that for people like ourselves who have a free Twitter Developer account, only 100 tweets can be viewed at a time hence our connotation value being so sporadic.

Accomplishments that we're proud of

Being able to use the sentiment analysis and NLP to obtain the polarity of a tweet and return whether it was a positive, negative, or neutral tweet and being able to find the metrics of likes, retweets, followers, etc. in order to calculate the value for the tweet was a big achievement.

What we learned

We learned to use the tweepy and textblob libraries in order to access and pull data from the Twitter API in order to algorithmically calculate the popularity of a tweet. We also figured out how to map out our algorithm in order to calculate the value of a tweet.

What's next for Crypto Popularity Predictor

In the future, we plan on building upon this project in several ways, including a spam tweet detector, using an API from an existing cryptocurrency exchange or platform to find up and coming crypto currencies, improve our algorithm for calculating the connotation, and accounting for a popular slang term known as “ratio” which signifies a reply having significantly more likes than the original tweet authors likes, reducing their credibility and the connotation of the original tweet.

Built With

• colab
• data-analysis
• jupyternotebooks
• natural-language-processing
• python
• sentiment-analysis
• textblob
• tweepy
• vscode