Scaling ML/AI has been a major challenge for medium to large organisations. Most organisations can productionize only 30% of their ML efforts.
Join us for a deep dive into how we are confronting these challenges at Rakuten. With this session, you can learn about Rakuten’s Data Science Platform, Architecture and Modernize Feature Store capabilities that help in productionizing ML predictions faster at scale and enable collaboration across organizational boundaries to build value.
Day 2
August 20, 2021
Rakuten Data Platform provides Data and Analytics services to various services across Rakuten Group. It today collects, manages and serves data sets from 70+ services with 43+ integrated services, 20K+ managed datasets and 6500+ weekly Active Users.
In this session we will briefly touch overall data landscape, Journey, the evolution of this platform over time, high-level architecture, the common challenges we are trying to address and the approach we have taken to address those challenges. We will then drill down into the Data Discovery & Utilization part of the platform which is the key to the democratization of Data within Rakuten Group.
You will also have a sneak peek into New products & toolsets which we are building some of which are in alpha/beta stages today.
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 Geodata and Geo Science applications are key to optimization and overall success in the fields of Logistics, Site Planning, Area Marketing and improving Customer Experience and many more. In this talk, we will cover some of the Geo initiatives at Rakuten and discuss some interesting questions around the viability for online data to fully represent a user and how to potentially rectify that via offline data and data science.
Product Matching In E-Commerce, Product Matching is one of the fundamental problems for various use cases like (1) Competitive pricing of products, (2) Deduplication of products in the catalog, (3) Grouping items from various merchants (4) Recommending products. The requirement is to match a product accurately against a catalog spread across tens of thousands of taxonomy nodes and millions of items. Product matching results must be accurate, and the margin for error is minimal to use Product Matching across use cases. In this talk, we touch upon the “Six honest serving-men” by Rudyard Kipling –What, Why, How, Where, When, and Who of Product Matching. We share the overall approach, experience in designing and orchestrating Deep Learning models with a scalable architecture to achieve the required results. Here we have approached the problem at the grass-root level consisting of five stages (1) Identifying attributes per taxonomy node (2) Classification of Products (3) Attribute Enrichment from NER (Text) and Image feature extraction (4) Search against multiple indices and filter results for mandatory attributes (5) Re-rank to improve the relevancy of shortlisted results. We also address Product data quality, measured at every stage to improve the overall performance of Product Matching, towards meeting business needs.
Learn how retailers use AI-based product enrichment techniques and image recognition models to make products more searchable. Understand how using AI smart search and complementary product recommendations like “shop the look” can help eCommerce companies engage consumers and drive increased revenue.
Most chatbot projects fail. In this session, we’ll first look at the standard approach that leads to the ‘typical’ chatbot you’ll encounter on many websites or mobile apps, and why it never works. Then, we’ll present the approach we use at Rakuten to ensure our chatbots provide real value to our businesses through natural, conversational experiences that empower customers to self-serve successfully. Finally, we’ll walk through a specific example of our Rakuten Mobile AI Assistant, show the steps we are taking to keep its performance high, improve its coverage over time, and measure its value to communicate back to the business.
This talk will touch upon the need for digital automation for behaviour analysis. Human behaviour analysis and activity analysis is an extremely complex subject. AI has been attempting to play a bigger role in the domain of human activity analysis. The need for human analysis allows better solutions to be built for 1.Media, 2. Sports, 3. Consumer Packaged Goods, 4. Manufacturing. Video is known to be the source of truth for capturing information. The challenge with video is: it’s expensive to analyse the unstructured data inside videos and generate useful insights on what is happening across a sequence of frames( read activities). With this background, we would like to cover how the need for an end to end activity recognition platform has evolved and has been well received for use cases specific to the Consumer Packaged Goods industry. This platform is capable of enabling DIY Video AI for activity recognition.
1. A data annotation component with the ability to do automated annotation
2. A component to build models that can be deployed on-premise or in the cloud.
3. A component for inferencing, correction of inferences and feedback to model building and training.
4. An insights component that is capable of analysis of detections and generation of intuitive analytics insights.
5. How an integration between AI in the cloud and AI on edge has been implemented and pushed to production.
We will also cover how all of the above are helping large multinational majors in the CPG segment apply their domain knowledge to generate new behaviour insights and apply activity recognition to analyse product use and measure the effectiveness of product redesign initiatives.
Most of the entities in the world are connected via any kind of relationship such as social networks, mobile networks, financial networks, user-product relationships. A graph (or network) is a natural way to represent such representations. Recently graph neural networks are getting paid a lot of attention from industries and academia to perform representation learning on graphs with neural networks so that we can use learned embedding for downstream tasks such as community detection, recommender systems, anomaly detection, and so forth. This talk covers an introduction to graph neural networks and their applications to various areas.
12
Page 1 of 2
