Can Twitter Sentimental Analysis Predict Bitcoin Price Fluctuation?

Programatically obtained sentiment has been the focus of many academic researches during the past few years, especially that social networks and micro-blogging sites have remarkably grown since their inception in the beginning of the millennium. The users of social networks have exceeded 2 billion in the year 2015, as per data presented by Statista.com. Interestingly enough, sentiment analysis has been utilized to formulate useful insights across various industries, especially in online marketing and analysis of currencies, stock, commodities…etc.

Bitcoin price has skyrocketed during the past few months, which was reflected by a threefold rise in the number of bitcoin related searches on Google, since the beginning of 2017. By design, bitcoin is a decentralized cryptocurrency that represents a free market; thus, the public’s perception has a great impact on bitcoin’s value.

A recently published paper examined the public’s perception of bitcoin’s value, via analyzing around 2.27 million bitcoin related tweets, in order to monitor fluctuation of sentiment that could signal change in price in the near future. The analysis of sentiments of snippets and opinions obtained from Twitter regarding bitcoin, aimed at answering two main questions;

1. Is there a relationship between the public’s sentiment on twitter and fluctuations of bitcoin price?
2. Can a naive bitcoin price prediction model, which is based on changes in the public’s sentiment, provide better than random accuracy?

 

Twitter Sentimental Analysis:

The study analyzed sentiment, as well as bitcoin price, on a short term basis, ignoring how sentiment, as reflected by posts on micro-blogging sites such as Twitter, can impact macro-trends in a cryptocurrency market such as bitcoin’s. The study monitored sentiment within a short term window 24 hours. Sentiment was classified as per the most naive binary way of positive or negative. A Lexicon based approach was used. A lexicon represents a group of features, such as words and their matching sentiment classification. Lexicon based approaches are commonly used means for sentimental analysis, where a piece of text, e.g. from a tweet, is compared to lexicons and assigned sentiment classifications. Even though Lexicons are relatively hard to create, once created they require minimal resources to use. Well formulated Lexicons can yield high accuracy results.

 

Results of the Study:

The study presented a naive sentiment prediction model, which relied on the intensity of sentiment’s fluctuations across short term intervals. The model proved that the most precise aggregated period to make price predictions over, was approximately one hour, signaling a bitcoin price change within a 4 hour period into the future.

The price prediction model evaluation proved that accumulating tweet sentiments over a period of 30 minutes, with four shifts forwards, along with a 2.2.% sentiment change threshold, yielded an accuracy of 83%.

Despite the fact that the used prediction model produced an accuracy of 83%, the overall number of predictions, were too few to be appropriate for venturing into reliable prediction model conclusions.

This study could be further built upon by other researchers via working on the lexicon, by improving the classifier via addition of domain specific lexicons that can pinpoint cryptocurrency related terms and produce more representative sentiments, which would definitely improve the accuracy of price predictions.