DSP ML Team, LINE Machine Learning Dept

Slide 1

Slide 1 text

DSP ML Team Machine Learning Dept. Data Science Center Takahiro Himura 2021/07/01

Slide 2

Slide 2 text

DSP ML Team

Slide 3

Slide 3 text

OUR MISSION 機械学習を通じて LINE広告の競争力・利益に貢献する

Slide 4

Slide 4 text

LINE広告のご紹介 ※キャプチャ画像はイメージです

Slide 5

Slide 5 text

LINE広告のご紹介 ⚫ LINEに広告配信するためのプラットフォーム ⚫ 国内月間アクティブユーザ 8,800 万人突破 ⚫ 国内最大規模の運用型広告プラットフォーム ⚫ LINE広告ネットワークを通じて LINE外の配信面にも広告配信が可能 ※キャプチャ画像はイメージです

Slide 6

Slide 6 text

Real Time Bidding RTT < 50ms LINE SSP Supply Side Platform LINE DSP Demand Side Platform BidRequest BidResponse

Slide 7

Slide 7 text

Machine Learning for ADs LINE DSP Demand Side Platform ⚫ CTR (Click-Through Rate) Optimization ⚫ CVR (Conversion Rate) Optimization ⚫ VR (View Rate) Optimization ⚫ Budget Optimization ⚫ Auto Bidding ⚫ etc. . Bid Price = Bid Amount × pCTR × pCVR Rank Ads CTR Models CVR Models FwFM, etc.. DFM FM, etc..

Slide 8

Slide 8 text

Project 例: User x Ad Similarity embedding user id embedding ad item id ∫ FM user features (age, gender, etc.) ad features site features others

Slide 9

Slide 9 text

技術スタック ⚫Python / Java / Kotlin / Go ⚫Spring Framework ⚫gRPC ⚫Redis ⚫Kafka ⚫Spark / Hive ⚫Kubernetes ⚫Argo Workflow ⚫Ansible / Drone.io / Jenkins ⚫Prometheus / Grafana We are Hiring Let’s learn, build and WOW users together

Slide 10

Slide 10 text

System Overview ap-worker Kafka Receiver imp, click, .. Consumer Kafka Redis Joiner Kafka joined event Kafka base event Worker Worker Worker Feature Redis join click & imp Batch Batch Argo Workflow DSP-ML Kubernetes

Slide 11

Slide 11 text

No content