Tal
Azaria

How We Used Sequence Modeling to Represent Our Merchant’s Behavior?

PayPal

Tal Azaria

Tal
Azaria

How We Used Sequence Modeling to Represent Our Merchant’s Behavior?

PayPal

Tal Azaria

Bio

Tal is a Data Scientist at PayPal Risk, a global team of Data Scientists and Decision Scientists, responsible for balancing between loss and customer experience by providing innovative and cohesive models and solutions throughout the seller lifecycle. Prior to PayPal Tal graduated the ITC fellows’ program where she gained hands-on experience in machine learning, deep learning and statistical analysis.

 

Tal is a Biomedical engineer with a proven track record in the field of cyber-security and medical devices with over 7 years of experience designing embedded system and managing products throughout their full life cycle.

Bio

Tal is a Data Scientist at PayPal Risk, a global team of Data Scientists and Decision Scientists, responsible for balancing between loss and customer experience by providing innovative and cohesive models and solutions throughout the seller lifecycle. Prior to PayPal Tal graduated the ITC fellows’ program where she gained hands-on experience in machine learning, deep learning and statistical analysis.

 

Tal is a Biomedical engineer with a proven track record in the field of cyber-security and medical devices with over 7 years of experience designing embedded system and managing products throughout their full life cycle.

Abstract

Sequence models are vastly used in our daily routine, enabling numerous exciting applications in speech recognition, music synthesis, chatbots, machine translation, natural language understanding, and many more. But can we utilize sequence information to represent behavior?

 

In this poster, I will describe how we represented models’ predictions as sequential information and leveraged sequence modeling to detect and automatically mitigate merchants with risky modus operandi. I’ll present the various models we have considered, starting with TFIDF, Doc2Vec, followed by LSTM and 1D CNN and we’ll discuss the pros and cons in each approach. 

Abstract

Sequence models are vastly used in our daily routine, enabling numerous exciting applications in speech recognition, music synthesis, chatbots, machine translation, natural language understanding, and many more. But can we utilize sequence information to represent behavior?

 

In this poster, I will describe how we represented models’ predictions as sequential information and leveraged sequence modeling to detect and automatically mitigate merchants with risky modus operandi. I’ll present the various models we have considered, starting with TFIDF, Doc2Vec, followed by LSTM and 1D CNN and we’ll discuss the pros and cons in each approach.

Planned Agenda

Planned Agenda