This is the official repository for the release of the CrypTop12 dataset. If you would like to use our research, please cite the following:
A. Garg, T. Shah, V. K. Jain and R. Sharma, "CrypTop12: A Dataset For Cryptocurrency Price Movement Prediction From Tweets And Historical Prices," 2021 20th IEEE International Conference on Machine Learning and Applications (ICMLA), 2021, pp. 379-384, doi: 10.1109/ICMLA52953.2021.00065.
Feel free to reach us out at [email protected] if you have any queries and suggestions for us!
The dataset consists of two components: the historical prices and the tweets.
Number of Coins | Duration | Number of Days | Average Tweets |
---|---|---|---|
12 | 21/09/2017 to 27/03/2021 | 1255 | 38.28 |
There are two main components:
The price data is taken from CoinMarketCap.
The raw and preprocessed data for prices is organized as follows:
- ./price/raw
- ./price/preprocessed
The tweet data is taken from Twitter.
The raw and preprocessed data for tweets is organized as follows:
- ./tweet/raw
- ./tweet/preprocessed
- Date: Date of observation (YYYY-MM-DD)
- High: Highest price on the given date
- Low: Lowest price on the given date
- Open: Opening price on the given date
- Close: Closing price on the given date
- Volume: Volume of transactions on the given date
- Marketcap: Market capitalization in USD on the given date
For pre-processed version:
-
We normalize Open, Close, High, Low as:
- Normalized Price(t) = (Price(t) / Close Price(t - 1)) - 1
-
We add an additional feature for movement percentage in the price dataset.
- Movement%(t) = (Close Price(t) - Close Price(t - 1)) / Close Price(t - 1)
-
We also remove the Marketcap attribute.
- id: Tweet Id
- username: Username of account posting the given tweet
- created_at: Date-Time of posting of the given tweet
- tweet: Content of tweet in raw format
- replies_count: Count of replies to the given tweet
- retweets_count: Count of retweets to the given tweet
- likes_count: Count of likes to the given tweet
- Follower_count: Count of followers of the username
For pre-processed version:
- We use TweetTokenizer by NLTK for primary tokenization of each tweet.
- "AT_USER" prefix for the usernames mentioned in the tweet.
- "URL" for all the hyperlinks.
- "#" prefix for the hashtags.
- Amish Garg
- Tanav Shah
- Vinay Kumar Jain
- Prof. Raksha Sharma