Introduction

This dataset contains 264,960 transactions between block 270,000 (2013-11-16 22:09:06 UTC) and block 300,000 (2014-05-10 06:32:34 UTC), labeled by Blockstream.info’s method of CoinJoin detection and a method of SharedCoin detection. In this period SharedCoin is active.

In order to provide more relevant information, the dataset includes a large number of transactions related to the above 264,960 transactions, but this transactions are not labeled.

There are a total of 116,537,811 transactions in the dataset, all of which are real Bitcoin transactions, but the original TXIDs have been replaced with consecutive integers starting from 0 to 116,537,810.

We have computed 15 features for each transaction and provided the ids of the precursor and successor transactions for these transactions.

Download Data

Google Drive

Baidu Netdisk (Access Code: wnr5)

Tree Data

File	Size	Lines	Description	Format
positive_ids.txt	1.2 MB	132,480	IDs of positive cases (Coin Mixing Transactions)	One ID per line
negative_ids.txt	1.2 MB	132,480	IDs of negative cases	One ID per line
tx_features.txt	14 GB	116,537,811	Processed features of transactions	One JSON List per line, empty line for Coinbase TX
tree_forward.txt	4.6 GB	116,537,811	Precursor of each node IDs	One JSON List per line, empty line for Coinbase TX
tree_back.txt	2.2 GB	116,537,811	Successor of each node in IDs	One JSON List per line, empty line for Unspent TX

There are 116,537,811 transactions in this dataset. The original TXID is removed and a consecutive integer starting from 0 is used as ID.

In tx_features.txt, tree_forward.txt and tree_back.txt, the first line refers to the Transaction of ID 0, the second line refers to the ID 1 ....

Load Data

1. tree_forward.txt

   import json
   
   forward_tree = []
   with open('tree_forward.txt') as f:
   	for l in f:
   		if l == '\n':
   			n = []
           else:
               n = json.loads(l)
           forward_tree.append(n)

2. tree_back.txt

   import json
   
   backward_tree = []
   with open('tree_back.txt') as f:
   	for l in f:
   		if l == '\n':
   			n = []
           else:
               n = json.loads(l)
           backward_tree.append(n)

3. tx_features.txt

   import json
   
   tx_features = []
   with open('tx_features.txt') as f:
   	for l in f:
   		if l == '\n':
   			n = []
           else:
               n = json.loads(l)
           tx_features.append(n)

Processed Features

File	Size	Lines	Description	Format
dev_set_full.zip	1.43 GB	79,488	Dev set	One JSON per line
train_set_full.zip	3.34 GB	185,472	Train set	One JSON per line

Load Data

See https://github.com/sxwxs/BitcoinCoinMixingDataSetWithRuleBasedLabel/tree/main/example/LSTM-TC/LSTM-TC_Classifier.ipynb

Reference

Our paper published in Applied Intelligence.

@article{sun2021lstm,
  title={LSTM-TC: Bitcoin coin mixing detection method with a high recall},
  author={Sun, Xiaowen and Yang, Tan and Hu, Bo},
  journal={Applied Intelligence},
  pages={1--14},
  year={2021},
  publisher={Springer}
}