USING AMAZON NEPTUNE TO BUILD FASHION KNOWLEDGE GRAPH

如果无法正常显示,请先停止浏览器的去广告插件。
分享至:
相关话题: #Amazon #zalando
1. USING AMAZON NEPTUNE TO BUILD FASHION KNOWLEDGE GRAPH AWS Finland September Meetup Helsinki September 2019
2. ZALANDO - EUROPE’S LARGEST ONLINE FASHION RETAILER 17 countries (+2 in 2019) 28+ million active customers >300 million visits per month ~10 million orders per month ~5.4 billion € revenue 2018 15 500+ employees in Europe 130+ nationalities Visit us: tech.zalando.com
3. ZALANDO HELSINKI TECH HUB Our office is located in BUILDING OUR ECOMMERCE PLATFORM KAMPPI AWS, Microservices, Scala, Android and iOS 90 employees 27 Nationalities 12 3 Autonomous delivery teams working with modern technologies
4.
5.
6. FASHION KNOWLEDGE GRAPH 6
7. What is a knowledge graph? Collection of elementary sentences 7 Subject Predicate Object Entity (IRI) Named (IRI) Directed Entity (IRI) or Literal
8. Why a Knowledge Graph? Problem: “Beyonce” was one of our most failing search queries. Solution: Record the link between Beyonce and Ivy Park in the Knowledge Graph. Search system can use this information. Beyonce IvyPark 8 isDesignerOf Person Brand
9. 9
10. Technical implementation 10
11. RDF - Resource Description Framework RDF is used to model the Knowledge graph An RDF graph is a set of RDF triples 11
12. SPARQL - SPARQL Protocol And RDF Query Language 12
13. SPARQL - Property path 13
14. SPARQL - Property path 14
15. Why Graph Graph DBs perform well for navigating highly connected data For a social network containing 1,000,000 people, each with approximately 50 friends, the results strongly suggest that graph databases outperform RDBMS for highly connected data At depth two (friends-of-friends), both the relational database and the graph database perform well enough but after that RDBMS joins will become expensive
16. The Graph Database 16
17. Blazegraph ● ● ● ● Blazegraph is an open source (GPLv2) graph database. Its protocol is HTTP-based, no special libraries is needed. One of the most complete in terms of SPARQL / RDF support. Used by the Wikimedia foundation for their wikidata query service since 2015 , but... ● ● No replication or cluster support in the open source version. The commercial version is not available since 2017... … because the company that developed Blazegraph is acquired by Amazon
18. Amazon Neptune Blazegraph appears as the foundation of Amazon Neptune database: ★ = + + ★ ★ ★ Replication and the R/O endpoint to distribute load among replicas; Redundancy via failover; Cloud backups; Technical support.
19. Fashion Knowledge Graph Deployment ● ● ● Neptune endpoint supports no authentication; Let’s introduce a low-level API: ○ Uses Zalando OAuth2 for authentication and authorization; ○ Does basic SPARQL validation. High-level APIs may be deployed in different VPCs.
20. Neptune is not an easy guy ● ● ● ● ● There may be bugs! A concurrency problem that we have reported in mid-2018, has been finally fixed in the version 296 in May 2019; Restoration from backup is VERY SLOW; Backups are confined inside AWS - you can’t download them. We used data dump (using a SPARQL SELECT) and upload instead for migration between AWS regions; Test environment is rather expensive (instance types start from r5.large); Performance may be really bad even with small graphs, if your queries are poorly written.
21. Performance No magic, it requires work even if you use the right DB for your task Issues with complex queries (especially property path) Denormalize the data - - 21 Transactions to ensure consistency Named graph can help
22. Conclusion ● ● ● ● ● 22 Amazon Neptune is a good solution for RDF/SPARQL-based knowledge graph; It can scale horizontally for read queries (by increasing the number of replicas); But it’s not magic, you need to work on your queries; You still can use Blazegraph as a compatible replacement for Neptune in tests and development; Think over your backup&restore (and data migration) strategy. Test it works as you expect.
23. Feel free to reach us: Matthieu Guillermin <matthieu.guillermin@zalando.fi> Uri Savelchev <uri.savelchev@zalando.fi> Find out more about our Culture, People & Jobs: ● ON SOCIAL MEDIA: Linkedin @Zalando SE Facebook @ Inside Zalando Instagram @insidezalando Twitter @ZalandoTech ● ● CAREER WEBSITE: jobs.zalando.com CORPORATE WEBSITE: corporate.zalando.com

Home - Wiki
Copyright © 2011-2025 iteam. Current version is 2.142.0. UTC+08:00, 2025-02-22 04:43
浙ICP备14020137号-1 $Map of visitor$