Machine Learning for Trust and Safety at Pinterest

Explore machine learning solutions for trust and safety challenges at Pinterest in this 32-minute conference talk by Vishwakarma Singh, a Machine Learning Researcher at the company. Gain insights into the proactive prevention of policy-violating content, including spam, pornography, misinformation, and hate speech. Discover the complexities involved in developing T&S solutions, such as adversarial challenges, scalability issues, cost-effectiveness concerns, and legal considerations. Learn about the successful implementation of batch and near real-time ML solutions using advanced techniques like Deep Neural Networks, Convolutional Neural Networks, and Graph Convolutional Networks. Understand the processes and guidelines for designing, developing, and productionizing these models, focusing on cost-effectiveness, performance elasticity, and adaptability. Examine the comprehensive offline metrics used for model evaluation, combining standard ML metrics with business metrics. Delve into the challenges faced during development and productionization, including dataset limitations and stringent false positive rate requirements. Gain valuable insights into ML model design principles and solution development processes employed at Pinterest.