Bitcoin price prediction using sentiment analysis

Svrha ovoga rada je utvrđivanje uzročno-posljedičnih veza između sentimenta s društvenih mreža i drugih povezanih bihevioralnih čimbenika i kretanja cijene bitcoina. Također, jedan je od temeljnih ciljeva postizanje veće razine točnosti predviđanja cijene bitcoina uključivanjem sentimenta u prediktivne modele. Sukladno tome, provedena je analiza sentimenta objava s Twittera korištenjem modula VADER na uzorku od 1,64 milijuna objava. Prosječni dnevni polaritet sentimenta korišten je kao input za izgradnju više različitih modela, zajedno s prosječnim dnevnim brojem objava i brojem pretraživanja pojma bitcoin na tražilici Google i Wikipediji. S ciljem predviđanja budućih cijena, korišteni su VAR, Prophet i LSTM model, te je provedena komparacija s obzirom na postignutu točnost modela.The main aim of this paper is to find and explore the causal relationship between social media sentiment and other related behavioral factors and bitcoin price movements. Additionally, one of the main goals was to achieve better accuracy in bitcoin price prediction with the inclusion of sentiment in predictive models. Therefore, Twitter sentiment analysis was conducted on 1,64 million tweets, with the use of VADER lexicon-based text mining library. The average daily sentiment polarity was used as an input for different models, together with the average daily tweet volume, Google search and Wikipedia search volumes. VAR, Prophet and LSTM models were used with the aim of predicting future bitcoin prices, and their performance was compared based on achieved accuracy