Cryptoeconomics: Data Application for Token Sales Analysis
Abstract
Initial coin offering, which is called ICO, is rapidly raised its volume by $3.5B in 2017 as a bypassed crowdfundings. Its mainly due to five of successful ICOs, Filecoin $257M, Tezos $232M, EOS $185M, Paragon $183M and Bancor $153M. Since theses great successes, tons of startups are jumping into ICOs, but still there is no standard criteria to analyze their marketing strategy. Currently, most of token sales relies on their white papers, SNS channels and Github to attract investors. The goal of this paper is thoughtfully exploring all these data sources using the Big data analysis tools, like Elasticsearch and Kibana, then building the visualized marketing analytics system on ICOs progress. Our main approach is using domain-knowledge to find related criteria then extracting features through unsupervised learning to build the prediction model. There are several criteria which are widely accepted in this market for startups investments. First, we pre-analyzed top 10 ICO success cases to validate these criterias. This paper also adopt unsupervised learning for feature validation that provides clustering method to locate similar features in the near density space. For our classification, we considered top 10 success ICO startups in 2017 as a success label. We collected 500 ICO startups metadata from the icotrackers and build database on founder, whitepapers, SNS likes and followers, ICO results so on. We also collected each exchange rates of cryptocurrencies on USD to see the price effect of ICO investments. Through selected features we modeled the prediction of success using Gradient Boosting method which ensembles weak models to build better models. This method have showed great performance in many other model competitions in Kaggle. Our final result is presented as a visual marketing analytics system which is designed on Elasticsearch, Kibana and Python XGboost. Kibana is web-based visualization system which connected to the full-search Nosql database, Elasticsearch. This system may analyzes current ICOs status in real-time then visualize whole analysis on the Kibana dashboard. The systems uses XGboost for the prediction which is the reliable Gradient Boosting implementation. This visual analysis systems provides summarized criterias analysis and prediction results visualization to monitor ICOs progression.