Slide 1

Slide 1 text

1 Trust And Safety Backend Introduction By Nikinsha Kesharwani Technical Lead

Slide 2

Slide 2 text

2 󰠁 Software Engineer on the Trust and Safety team
 🧘 From India
 🏯 Joined Mercari in Feb. 2019, 3.5 years in Japan
 🍲 Hobbies include cooking and playing badminton 🏸
 Nikinsha Kesharwani ✨

Slide 3

Slide 3 text

3 TnS ¥1000,000 ¥90,000

Slide 4

Slide 4 text

4 Some types violations in market Items that violate the law XXX-XXX Offensive comments Please transfer the money to my bank account XX-XX-XX Payment not provided by Mercari Foods nearing expiration date

Slide 5

Slide 5 text

5 2 types of Moderation ● Listed item moderation ● Account moderation

Slide 6

Slide 6 text

6 Why do we need item moderation? ● 20.4 million MAU browsing and listing items. ● More than 1.5 million items listed per day. ● Proactive detection of violating items using technology is necessary.

Slide 7

Slide 7 text

7 Our approach for Item moderation Fill in the information Complete listing Tap “List” Detection + Moderation Prevention

Slide 8

Slide 8 text

8 Overview of Item moderation Rule based system ML System Automatic delete CS decision 2. After Listing Edit Rule based system 1. Before Listing Rule based + ML model Rule based warning/stop New Listing

Slide 9

Slide 9 text

9 Overview of Item moderation Mod Tool Operator Rule ML Rights Holder Customer Item Detection Internal External Rule and ML based system are monitoring all items listed by 20.4 million active users

Slide 10

Slide 10 text

10 Architecture of Item moderation system Gateway Mod Tool Listing Service Reporting worker ML System Rule based System List item Subscribe Publish events (Detection) Subscribe Publish Publish Subscribe Report Item Seller (Prevention)

Slide 11

Slide 11 text

11 Account moderation offensive comment! Warn and restrict! 警告と禁止! CS

Slide 12

Slide 12 text

12 Overview of account moderation ● Major violation that are being detected ○ Detection with user information and behaviour ■ listing not at hand items

Slide 13

Slide 13 text

13 Tech Stack ● Backend Services (6) ○ Golang, GCP technologies like spanner, cloudSQL, pub/sub etc. ● ML Service (1) ○ Python, sklearn, and kubeflow ● Frontend Service (1) ○ TypeScript and React ● Build and release ○ Docker, Kubernetes, Spinnaker

Slide 14

Slide 14 text

14 What do I enjoy ● Working in ○ New technologies. ○ Cross-functional team, with opportunities to work on front-end and ML. ○ Great work culture, people and work life balance.

Slide 15

Slide 15 text

15 People we would love to work with ● Resonates with our mission to “Create safe marketplace using technology”. ● And work together to continuously enhance safety and security of Mercari marketplace.

Slide 16

Slide 16 text

16 Thank you! ありがとうございます