The BTC_sentiment_analysis.ipynb notebook, shows how to download and preprocess 100.000 Bitcoin (BTC) tweets from 2022-11-01 to 2022-11-15 to perform sentiment analysis to gather market intelligence. The aim of the analysis is to understand what are people’s opinions about BTC tweets.
Sentiment analysis consists of converting a text into a score that estimates its sentiment (e.g., demand for a given product) and can be deemed as a classification task (e.g., to extract subjectivity and sentiment polarity from text data).
Web scraping is done using snscrape (https://github.com/JustAnotherArchivist/snscrape).
Filtering for English language is done using langdetect (https://github.com/Mimino666/langdetect).
Sentiment analysis is done using spacytextblob (https://github.com/SamEdwardes/spacytextblob).
Scraped data looked like this:
After preprocessing, including removal of symbols, URLs, and non-ASCII characters, and removal of non-English Tweets, data looked like this:
Finally, sentiment analysis is carried out. Before analyzing the content of the tweets, Tweets underwent further preprocessing, including lemmatization (using the WordNetLemmatizer).
Scores look like this:
Most common words in positive Tweets:
Most common words in negative Tweets:
Conclusion: According to this analysis, with a mean sentiment of 0.07 (SD = 0.22) there is a slight but prevalent positive opinion in BTC tweets.