This paper examines the relationship between the sentiment derived from social media and the cryptocurrency price movement. To achieve this objective, the study combines two natural language processing tools, specifically a document-level sentiment analysis and an aspectbased leved analysis, which is complemented by topic modelling. The dataset has been collected from Twitter using Python and it consists of tweets related to Bitcoin from 2017 to 2021. The research reveals that the aggregate daily tweets conveyed preponderently positive sentiments. Moreover, the correlation coefficient between sentiment polarity and prices is close to 0, indicating the lack of a connection between the two variables. The aspect-based sentiment analysis provides slightly better results than the document-level one, however, it is still not able to explain how does the prevailing sentiment of a topic explain the changes in Bitcoin price. There was not enough evidence to confirm a significant relationship between the sentiment derived from Twitter and Bitcoin price movements. Several limitations could have affected the accuracy of the models: lack of multimedia content processing, use of slang and sarcasm.
|Educations||MSc in Applied Economics and Finance, (Graduate Programme) Final Thesis|
|Number of pages||116|
|Supervisors||Raghava Rao Mukkamala|