Anti sybil package for BrightID

# BrightID Anti-Sybil

This package provides a framework to evaluate the quality of different anti-sybil algorithms, by simulating different attacks to BrightID's social graph.

Comparing performance of different algorithms in detecting sybils in different attacks

## Algorithms

• SybilRank is a well-known sybil detection algorithm that is based on the assumption that sybils have limited social connections to real users. It relies on the observation that an early-terminated random walk starting from a non-Sybil node in a social network has a higher degree-normalized (divided by the degree) landing probability to land at a non-sybil node than a sybil node.

• GroupSybilRank is an enhanced version of the SybilRank algorithm. In this algorithm, a graph is created in which the BrightID groups are nodes and edges are weighted based on affinity between groups. Then original SybilRank algorithm will be applied to this graph of groups and users get scores from the best group they belong to. This algorithm achieved best results so far in identifying sybils based on modeled attacks and is being used as official BrightID anti-sybil algorithm. up edges.

• WeightedSybilRank is an enhanced version of the SybilRank algorithm that uses the number of common neighbors of the tow connected nodes as weight (trustworthy factor) of the edge.

## Attacks

### Lone Attacks

One attacker attempting to propagate score to the Sybils to verify them by connecting to other nodes and creating groups. We assumed that an attacker will have one account with a normal or above-average number of direct connections to honest users which they can use for interconnections to sybil accounts.

• One attacker attempts to connect to some of the seed nodes and create some sybil nodes. implementation - graph

• One attacker attempts to connect to some of the top-ranked honests and create some sybil nodes. implementation - graph

• One attacker attempts to connect to some of the honests and create some sybil nodes. implementation - graph

• One attacker attempts to connect to one of the top-ranked honests and create multiple (duplicate) groups of the sybils. implementation - graph

• A seed node attempts to create some sybils nodes. implementation - graph

• An honest node attempts to create some sybils nodes. implementation - graph

### Collaborative Attacks

Multiple attackers attempting to propagate score to the Sybils to verify them by connecting to other nodes and creating groups. Attackers are able to connect to each other and each others’ sybil accounts in any way. We assumed that each attacker will have one account with a normal or above-average number of direct connections to honest users which they can use for interconnections to sybil accounts. All these attacks can be performed by one or more groups of attackers who collaborate together.

### Manual attack

This is a way to manually add new nodes/edges/groups to the BrightID graph and see how different algorithms rank those nodes. You can use MANUAL_ATTACK_OPTIONS variable in the config.py file to define the manual attack. This example adds 3 sybil nodes, connect them to xGUyVQLYV80pajm8QP-9cfHC7xri49V58k02kqTAiUI as attacker and add them to a new group. graph for this manual attack

MANUAL_ATTACK_OPTIONS = {
'top': True,
'connections': [
['xGUyVQLYV80pajm8QP-9cfHC7xri49V58k02kqTAiUI', 'sybil1'],
['xGUyVQLYV80pajm8QP-9cfHC7xri49V58k02kqTAiUI', 'sybil2'],
['xGUyVQLYV80pajm8QP-9cfHC7xri49V58k02kqTAiUI', 'sybil3'],
['sybil1', 'sybil2'],
['sybil1', 'sybil3'],
],
'groups': {
'new_group_1': [
'sybil1',
'sybil2',
'sybil3'
]
}
}


## Install

$git clone https://github.com/BrightID/BrightID-AntiSybil.git$ cd BrightID-AntiSybil
$pip3 install .  ## Running Tests You can configure the algorithms and attacks you want to test by editing the config.py file in anti_sybil/tests/attacks/config.py and then run the tests by $ python3 anti_sybil/tests/attacks/run.py


The result will contain:

• An interactive graph (example) per algorithm/attack that visualize the graph and scores each sybil/attacker/honest node achieved

• A CSV file (example) that has a column per algorithm/attack and provide following information for each algorithm/attack
Results GroupSybilRank one group group target attack
No. Successful Honests 416
Successful Honests Percent 78.1954887218045
Sybils scored >= % 0.080091533180778
Avg Honest - Avg Sybil 17.4819290581162
Max Seed 100
Avg Seed 59.6248484848485
Min Seed 31.88
Max Honest 100
Avg Honest 29.3145290581162
Min Honest 0
Max Attacker 13.96
Avg Attacker 13.96
Min Attacker 13.96
Max Sybil 13.96
Avg Sybil 11.8326
Min Sybil 5.87
Border 14
• A chart to compare effectiveness of different anti-sybil algorithms to detect sybils in different attacks

## Old Version

The old version of BrightID Anti-Sybil algorithms, tests and documents can be found here.

## Release history Release notifications | RSS feed

Uploaded source
Uploaded py3