by Evan Sixtin
An open access, peer-reviewed research paper published on May 12, 2017, by researchers from Korea University and Kangnam University in South Korea has explained a method for predicting the fluctuation in the bitcoin price and transactions based on user opinions posted on online forums.
The paper, entitled “When Bitcoin encounters information in an online forum: Using text mining to analyze user opinions and predict value fluctuation” was written by Young Bin Kim, Jurim Lee, Nuri Park, Jaegul Choo, Jong-Hyun Kim, and Chang Hun Kim. Unlike previous research on Bitcoin forums, which did not give enough attention to noteworthy comments, this new study used an approach that extracted keywords from user comments in an attempt to predict the price and extent of transaction fluctuation based on Bitcoin online forum data ranging over a period of almost three years from December 2013 to September 2016. The researchers then developed a model based on deep learning to predict the bitcoin transaction count and price.
The first step in creating the price predicting model was gathering the data; Bitcoin-related posts on the online forum (bitcointalk.org), daily bitcoin transaction counts, and bitcoin price made up the data that was gathered. Keywords were extracted from the bitcointalk forum data, and these were judged and measured based on forum score ratings. This forum data was collected by data crawlers. The ‘Bitcoin Discussion’ subsection under the ‘Bitcoin’ section was used in this study because this is where comments are posted most actively, but no personally identifiable user data was collected.
Then a document-term matrix was constructed from the 17,381 forum articles and 627,122 user comments collected from the Bitcoin forum. Each article contained five attributes: ‘content’, ‘topic’, ‘comments’, ‘date’, and ‘views’, and each comment contained the ‘content’ and ‘date’ features. Using the ‘date’ field, the researchers split up the document-term matrix for daily analysis, then applied topic modeling to each day to extract different topic sets and their representative keywords across different dates.
In addition to data from Bitcointalk.org, Coindesk price data was used, and Google Trends data along with Wikipedia usage data was examined to reinforce the learning model. The results were remarkable. The most accurate prediction model for the bitcoin price yielded an accuracy rate of 80.39 percent, while the most accurate prediction model for bitcoin transaction had an accuracy rate of 81.37 percent.
“That said, the proposed method has a limitation in terms of its broader applicability due to the fact that the concepts were constructed for a long period of time. Moreover, the present findings warrant further studies on the analysis of user comments relative to the characteristics of Bitcoin forums.”
It is noteworthy to mention that the same methods and model used for predicting the bitcoin price and transactions in this particular study could also potentially be used to predict price and transactions for other cryptocurrencies. This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education, Science, ICT and Future Planning and Institute for Information and Communications Technology Promotion (IITP) grant funded by the Korea government.
Although we are still a long way off from being able to predict with 100 percent accuracy the ups and downs of any traditional currency or cryptocurrency, especially when taking into account unpredictable black swan events, this research introduces a functional foundation for using social media and trends to determine currency fluctuations. The research can still be built upon and fine-tuned to create an even more accurate model in the future.