Abstract

Over recent years, the data generated by social media websites have been increasing exponentially. Opinions, rumours, and news about the latest trends and technologies are generated in humongous quantity by people every day. These data are usually shared through various social media sites such as Twitter and Reddit.

The project aimed to use this information about the latest cryptocurrencies to predict the outcome of the cryptocurrency market. Multiple pre-processing techniques were experimented before concluding data preparation. TF-IDF Vectorizer was selected for feature extraction in the model built after thorough experimentations. An imbalance among the classes was detected in the dataset which was tackled with the tactful implementation of the Synthetic Minority Oversampling Technique (SMOTE). The C-Support Vector Classification algorithm was selected as the primary classifier after thorough testing and experimentations with several other state-of-the-art classifiers put up against each other. The model was further optimized with hyperparameter tuning.

The results of these experimentations were finally brought together in an interactive GUI dashboard. The dashboard allows the user to analyse the sentiment of the most recent data generated on social media sites using various visualization tools. It also allows for comparison with the live cryptocurrency market. The user may also input their own data into the dashboard to analyse the sentiment using a state-of-the-art SVC model.

Cryptocurrency Market Sentiment Analysis