CUSTOMER PROGRAM

Deep customer understanding to improve Rakuten membership value

Customer program consists of a group of research scientists specialized in customer research and analytics. The mission of this group is to know more about the customer using available data and predict customer behavior in their entire customer life cycle so that marketers can take the required actions to acquire, nurture, and retain customers. This group has been a key contributor to enrich Rakuten’s CustomerDNA by envisaging inferred attributes, attributes like lifestyle, marital status, income level, etc. which are not known from customer data, but can be inferred based on their usage pattern, buying behavior, browsing history, etc. Other notable initiatives include devising an AI based credit scoring model that will let lenders like Rakuten Bank to give loans to a larger audience with reduced stress of non-payment, building models that can be used for customer prospecting (using AIris, which is a lookalike model for target prospecting) and banner optimization using PITARI (which is solution for customer targeting based on profile, A/B testing, and delivering personalized ad content).

Customer Pre-trained Model

Research scientists try to build universal customer models and applications, motivated by the recent success of pre-trained models in the language and vision fields. Such models can be adapted to improve a diverse set of service solutions including recommendation systems, customer targeting solutions, and credit scoring functions, to name a few. For initial and simpler applications, we predict customer attributes with CustomerDNA.

Credit Scoring

Credit Scoring is a crucial research topic across the Rakuten group, specifically in our fintech businesses including credit card, payment, securities, and banking businesses. Research scientists apply machine learning, data mining, and optimization techniques to solve challenging problems, and build models and solutions regarding credit scoring-related projects. Our solutions include scoring for both business clients and individual users.

Policy Optimization

Rakuten group is contacting our customers with various kinds of communication channels, such as showing banners on web pages, sending email, pushing notification through apps, to promote our products and services. In order to deliver the right content at the right time to the right customers, research scientists apply reinforcement learning algoirthms to build models to content delivery platforms.

CDNA-IA

Customer DNA (CDNA) is a platform to manage Rakuten customer profiles by utilizing data from different Rakuten businesses. CDNA-IAs, a part of CDNA, are predicted customer profiles using machine learning profiles. The team builds and operates many models for CDNA-IAs. The team does apply recent research progress to develop efficient data pipeline and model creation.

Social Graph

Identifying social relationships between Rakuten members, such as family and friends, enables us to improve performance of the marketing measures. Under this assumption, research scientists build a social graph structure among Rakuten members and their relatives by collecting data from multiple Rakuten services, and apply graph algorithms including graph neural networks to conduct marketing measures, such as a family introduction campaign.

Geo Science

With more than 70 different services in the Rakuten Group globally, our data science teams cover diverse initiatives in both the online and offline worlds. Offline Geo data and Geo Science applications are key to the optimization and overall success in fields including Logistics, Site Planning, Area Marketing and improving Customer Experience. The Geo Science Team creates algorithms and builds new Geo Intelligence layers to better understand offline activity to support many Rakuten Group services.

PUBLICATIONS

VIEW ALL

Neural Representations in Hybrid Recommender Systems: Prediction versus Regularization

Conference: SIGIR 2021

Author: Ramin Raziperchikolae, Tianyu Li, Young Joo Chung

Publication Year: 2021

Improving Cold-start Item Advertisement for Small Businesses

Conference: SIGIReCOM

Author: Yang Shi, Young Joo Chung

Publication Year: 2021

Steady-State Analysis of Episodic Reinforcement Learning

Conference: NuerIPS2020

Author: Bojun Huang

Publication Year: 2020

Learning to Profile: User Meta-Profile Network for Few-Shot Learning

Conference: CIKM2020

Author: Hao Gong

Publication Year: 2020

Item 1 of 4

CAREERS

We are always looking for great talent and researchers to lead us and work with us.

LEARN MORE