Understanding the Internet's Opinions on Formula 1
by Givarrel Veivel
PROBLEM STATEMENT: On the internet, it's much easier to pay attention to either toxicity & hate, or notions that only reflect our own (confirmation bias). This project looks at the data more objectively to REALLY unveil Reddit & Twitter's sentiment on different Formula 1 drivers, while also trying to make sense of the different factors of opinions in F1.
This is a TEXT CLASSIFICATION and OPINION MINING project, where data is retrieved from replies under official @F1 tweets (and possibly Reddit comments in the future). Each tweet will be classified based on topic (the subject driver or team), and then we will label the tweet's sentiment (positive vs negative opinion).
(1) Pull replies from specified tweet -> (2) Label opinion/sentiment -> (3) Clean text content of tweets (for bag of words model)-> (4) Evaluate & train model -> (5) Test model on unlabeled data
First I use train/twitterer.py to pull my train/test data. This is done by pulling tweets that are subreplies of my target tweet, although with the method I use it is limited to retrieving recent ones. Second, with the help of train/sentiment_labeler.py, I label my data: negative, positive, or neutral. Then I deploy model.ipynb to clean and evaluate the model, before using it to predict the sentiment of my test data.
The biggest limitation is the amount of data I have and can obtain. I would need to try a different method, or perhaps use a different platform, to try and obtain more training & testing data.
- Obtain metadata (likes, retweets, etc)
- Gather more training data (!!!)
- Implement RNN model