Fighting Abuse @Scale 2019 recap

Meta

Fighting abuse presents unique challenges for large-scale organizations working to keep the people on their platforms safe. At Fighting Abuse @Scale 2019, engineers, data scientists, product manage…

Overview

The article recaps the Fighting Abuse @Scale 2019 event, where experts from major tech companies gathered to discuss advanced technologies and strategies for combating online abuse, fraud, and spam. Key topics included human review systems, machine learning frameworks, and collaborative efforts between platforms to enhance safety.

What You'll Learn

1

How to detect targeted spyware abuse using server fingerprinting and DNS probing

2

Why human labeling is crucial for training machine learning models in anti-abuse systems

3

How to implement deep entity classification for detecting abusive accounts

4

How to apply unsupervised learning techniques for detecting fake accounts

5

How to leverage collaboration between platforms to enhance content moderation

Key Questions Answered

What are the main challenges in detecting spyware abuse?

The main challenges include the sophistication of attackers using zero-click exploits and the difficulty of noticing surveillance. Researchers can use methods like server fingerprinting and Internet scanning to identify victims and deter future misuse.

How does Facebook's deep entity classification framework work?

Facebook's deep entity classification framework uses machine learning to identify abusive accounts by analyzing their embeddings in the social graph. This approach helps mitigate the risk of adversaries reverse-engineering features used for classification.

What role does human labeling play in anti-abuse systems?

Human labeling is essential for measuring abuse, generating training data for machine learning models, and determining enforcement actions. High-quality data from human reviewers is crucial for effective content moderation at scale.

What collaborative efforts are being made to combat online abuse?

Facebook and Google are collaborating to identify and remove terrorist content from their platforms. By sharing insights and data, they aim to enhance their ability to detect and mitigate abusive content more effectively.

Key Statistics & Figures

Fake accounts removed

Hundreds of millions

This statistic highlights the effectiveness of Facebook's deep entity classification system in combating abusive accounts.

Technologies & Tools

Backend

Machine Learning

Used for detecting abusive accounts and improving content moderation systems.

Key Actionable Insights

1
Implementing a robust human review system can significantly improve the quality of content moderation.
As organizations scale, maintaining high-quality data from human reviewers becomes challenging. By focusing on effective representation of information for reviewers, companies can enhance their anti-abuse efforts.

2
Utilizing unsupervised learning techniques can help detect fake accounts more effectively.
Given the adversarial nature of online abuse, traditional supervised learning may fall short. Unsupervised methods can adapt to evolving patterns of abuse and improve detection rates.

3
Collaboration between platforms is essential for combating shared threats like extremist content.
By understanding how malicious actors utilize shared platforms, companies can develop strategies to identify and remove abusive content more efficiently.

Common Pitfalls

1

Relying solely on supervised learning techniques can lead to ineffective detection of abusive accounts.

As adversaries evolve their tactics, supervised methods may not capture the nuances of new abusive behaviors. Adopting unsupervised learning can provide a more flexible approach.

Related Concepts

Online Abuse Detection

Machine Learning In Security

Content Moderation Strategies

Collaboration In Tech Industry