【Author】
Cheuque Cerda, German; Reutter de La Maza, Juan
【Source】COMPANION OF THE WORLD WIDE WEB CONFERENCE (WWW 2019 )
【Abstract】The Bitcoin protocol and its underlying cryptocurrency have started to shape the way we view digital currency, and opened up a large list of new and interesting challenges. Amongst them, we focus on the question of how is the price of digital currencies affected, which is a natural question especially when considering the price roller-coaster we witnessed for bitcoin in 2017-2018. We work under the hypothesis that price is affected by the web footprint of influential people, we refer to them as crypto-influencers. In this paper we provide neural models for predicting bitcoin price. We compare what happens when the model is fed only with recent price history versus what happens when fed, in addition, with a measure of the positivity or negativity of the sayings of these influencers, measured through a sentiment analysis of their twitter posts. We show preliminary evidence that twitter data should indeed help to predict the price of bitcoin, even though the measures we use in this paper have a lot of room for refinement. In particular, we also discuss the challenges of measuring the correct sensation of these posts, and discuss the work that should help improving our discoveries even further.
【Keywords】Bitcoin; Recurrent Neural Networks; Twitter; Sentiment Analysis; Price Prediction
评论