The analysis of Bitcoin Transaction Patterns

Combining existing research results with graph models and graph DBs, including thoughts on applying machine learning algorithms for pattern matching.

Transcript

1. 7+( $1$/<6,6 2) %,7&2,1 75$16$&7,21 3$77(516
2. :+$7 $5( :( *2,1* 72 +($5" 0DNLQJ VHQVH RI ZKDW LV JRLQJ RQ :+$7 +$6 %((1 $&+,(9(' 62 )$5" STATUS QUO OF CURRENT RESEARCH RESULTS AND FINDINGS :+$7 :( :$17 72 $&+,(9(" APPLYING NEW ANALYSIS TECHNIQUES (GRAPH DB, ML) AND COMBINE EXISTING FINDINGS FOR DEEPER INSIGHTS :+(5( :( $5( *2,1*" FUTURE RESEARCH OBJECTIVES AND CHALLENGES
3. :+$7 +$6 %((1 $&+,(9(' 62 )$5" 3DUVH WKH WUDQVDFWLRQ XVHU JUDSK P. Tasca, S. Liu, A. S. Hayes, The Evolution of the Bitcoin Economy: Extracting and Analyzing the Network of Payment Relationships, 2016 30 Mio addr. to 2.850 (1k BTC) super-clusters with input addr. heuristic to 209 “known groups”
4. :+2 :$176 3,==$ )25  0,2" M. Spagnuolo, F. Maggi, S. Zanero, Bitiodine: Extracting intelligence from the bitcoin network, 2014 3URRIRIFRQFHSW  ²  *PEO %0  ²  0DWXUDWLRQ  ²  %LWFRLQ ,QFRPH E\ FDWHJRU\ ´/DV]ORµ FOXVWHU WUDQVDFWLRQ Source P. Tasca, S. Liu, A. S. Hayes The Evolution of the Bitcoin Economy: Extracting and Analyzing the Network of Payment Relationships, 2016 P. Tasca, S. Liu, A. S. Hayes, The Evolution of the Bitcoin Economy: Extracting and Analyzing the Network of Payment Relationships, 2016
5. :+$7 :( :$17 72 $&+,(9(" Greg Walker http://learnmeabitcoin.com/about 1.7 bn nodes 2.7 bn relationships Ź474k blocks Ź240m tx Ź280m addresses Ź650m output 3DUVH WKH WUDQVDFWLRQ JUDSK WR JUDSK GDWDEDVH a *%
6. :+$7 :( :$17 72 $&+,(9(" 0HWD LQIRUPDWLRQ WR HQULFK VFKHPD known categories web scrapers mega clusters heuristics multi-input addresses … shadow addresses Combining work on heuristics, known addresses etc. from Paola Tasca et al., Michele Spagnuolo et al., Greg Walker $SSO\ 0/ DOJRV ZKHUH IHDVLEOH (QULFK WKH GDWDVHW DQG DSSO\ 0/ ZKHUH IHDVLEOH unsupervised supervised semi-sup. Neo4j APOC D. McGinn, D. Birch, D. Akroyd, M. Molina- Solana, Y. Guo, J. Knottenbelt, Visualizing Dynamic Bitcoin Transaction Patterns, 2016
7. :+$7 :( :$17 72 $&+,(9(" ,QWHUHVWLQJ SKHQRPHQD IRU WUDLQLQJ VHWV HJ 0HPSRRO Z N YHUWLFHV D. McGinn, D. Birch, D. Akroyd, M. Molina-Solana, Y. Guo, J. Knottenbelt, Visualizing Dynamic Bitcoin Transaction Patterns, 2016 3DUDVLWLF :RUP • Blocks#364133, 364618 • Sudden increase in transaction rate • high frequency of Bitcoins into tiny amounts • occupying network resources • increased mempool pending verification • to force block increase above 1 MB! -> SegWit discussion 7XPRU DWWDFN • Blocks#367409, 368580 from29th July to 6th August 2015 • 2 phases, focused on data density vs. transaction rate • attack had limited impact on the TX backlog • but negative impact on unspent TX output backlog • origin is unknown / to be investigated
8. :+$7 :( :$17 72 $&+,(9(" &HQWUDOLW\ ! 5HOHYDQF\ • PageRank • Betweenness Centrality • Closeness Centrality 3DUWLWLRQLQJ ! 6WUXFWXUH • Label Propagation • (Weakly) Connected Components • Strongly Connected Components 3DWK)LQGLQJ ! ([SORULQJ • Minimum Weight Spanning Tree • All Pairs – and Single Source – Shortest Path • Multi-Source, Breadth-First Search 144 CPUs 260 GB Ram (util.) 1DWLYH 1HRM $32& JUDSK DOJRULWKPV IRU SHUIRUPDQW JUDSK DQDO\VLV %LWFRLQ 1HRM 0HHWXS Src: https://www.youtube.com/watch?v=55uB_t0RKTE Neo4j APOC native procedures (first R on 8th May 16) Neo4j native graph algorithms (first R on 24th July 17!) https://neo4j-contrib.github.io/neo4j-graph-algorithms/
9. :+$7 :( :$17 72 $&+,(9(" 1DWLYH 1HRM $32& JUDSK DOJRULWKPV ² UDZ SURFHVVLQJ Degree distribution of ‘locked’ relationships for ‘addresses’ (how often have addresses participated in transactions) Primitive clustering (address -> money -> address) without considering graph specifics People obviously did change addresses in transaction – but wait, Heuristics! Easy to identify clusters with algo – but wait, Mega clusters! M. Hunger, Bitcoin Neo4j Meetup, 2017 https://www.youtube.com/watch?v=55uB_t0RKTE
10. :+(5( :( $5( *2,1*" Challenge to combine graph structure, meta information and machine learning How will Bitcoin cash impact the ecosystem? What is going on with other alt-coins ETH, LTC etc.? 7KH VN\ LV WKH OLPLW MLlib Classification: logistic regression, naive Bayes,... Regression: generalized linear regression, survival regression,... Decision trees, random forests, and gradient-boosted trees Recommendation: alternating least squares (ALS) Clustering: K-means, Gaussian mixtures (GMMs),... Topic modeling: latent Dirichlet allocation (LDA) Frequent itemsets, association rules, and sequential pattern mining “Move from dots on a chart and rows in a table to connected-dots as pre-processing before moving to decision trees etc., helping the statistical correctness”
11. WKH HQG