Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Content-as-a-Service

 Content-as-a-Service

This talk is centred around building almost real time news search engine with fine details about scalable and fast responsive ( average 10 ms) system architecture.

Avatar for Heeren Sharma

Heeren Sharma

October 09, 2017

Other Decks in Programming

Transcript

  1. Content as a Service ”Building real time news search engine”

    Heeren Sharma Software Engineer Cliqz GmbH
  2. Cliqz • ~3 Million+ downloads across all platforms and products

    • Search Engine : ~2 Billion Indexed Pages (We don’t believe in indexing the web) • Privacy to User : Anti-Tracking & Anti- Phishing • Open Source Contributions • News recommendations
  3. System Architecture - Overview • Streaming Data • Python, Redis

    and Elasticsearch (core technologies) • Full Containerized – Micro-service architecture • Scalable – Horizontal and Vertical • Quick Responsiveness - ~10ms (avg)
  4. Cliqz News Technology - Search • Custom Query Search Language

    • Real time ranking • Topical Clustering • Trending News Events
  5. Cliqz News Technology – Search - QSL ”Tom Petty” site:spiegel.de

    url_path:musik contrib_twt_ids:2834511 lang:de
  6. Cliqz News Technology – Search - Real-time Ranking • Cliqz

    Boost Algorithm • Function Scores (Elasticsearch) • Mix of Signals
  7. Cliqz News Technology – Search - Real-time Ranking: Function Scores

    • Query Part • Script Score Part doc["engagement"].value*36000 / (params.now - doc["firstseen_at"].value)
  8. Cliqz News Technology – Search - Clustering and Trends •

    Cluster articles over same topic – Unsupervised • Representing terms from each cluster • Permutations and combinations – Word(s) transformations
  9. Self Serving System • Semi-Automated Data Source update • Expressive

    search queries • Embed your own ranking algorithms • Easy Data Insights – aggregations, dashboards, etc. • Extensive Experience in newsletters curation and editing
  10. Bumps in Journey #challenges #lessons-learned • Data modelling • Performance

    takes extra mile • Test Early • Think Scalable Architecture • Reliability of sources