Adding azuresearch
Download as PPTX, PDF0 likes266 views
We provide search as a service with 99.9% SLA across 11 regions. Our service indexes documents in over 50 languages using Microsoft's NLP stack. Collections contain indexes, indexers, and data sources. Documents are indexed via the REST API or SDK and searches return relevant results with hit highlighting and filtering options. The free tier hosts up to 10k documents in three indexes.
Adding azuresearch
- 3.
- 9. We wake up at 3am so you don’t have to 99.9% SLA Available in 11 regions and growing (latest additions are in Japan, Brazil, and Australia) Search engines are complicated and we’ve tried to make them easier to understand Tune and rank your results in a way that makes sense Full power of Microsoft Office NLP stack so your users find what they want in over 50 languages Easy integration with your favorite data sources Think in terms of documents rather than VMs & clusters Pay for the capacity need as your app grows Prototype on our free tier
- 11. Type Ahead
- 13. FacetsFacets
- 14. Hit Highlighting
- 17. Paging
- 19. Unit of deployment, isolation, and billing Contains indexes, indexers, and data sources Comes in two SKUs: Free (shared resources), and Standard (dedicated resources) Created through the Azure Preview Portal, or programmatically via the Management REST API Collection of documents that all share a common schema (like a table in a SQL database) Can include custom scoring profiles to affect ranking behavior of queries Created through the Azure Preview Portal, or programmatically via the Service REST API or .NET SDK
- 20. • • • • •
- 22. Use the REST API or .NET client to push JSON documents Write a script to do this once, or on a schedule using a web job Create an indexer for one of our supported data sources Set up a schedule to keep things in sync Created through the Azure Preview Portal, or programmatically via the Service REST API or .NET SDK
- 23. Host up to 10,000 documents in up to three indexes https://portal.azure.com One hour free No Azure Subscription, no credit card required https://tryappservice.azure.com/
- 34. What is an analyzer? Input: "Well!" thought Alice to herself "After such a fall as this, I shall think nothing of tumbling down-stairs!“ Standard analyzer output: well thought alice to herself after such a fall as this i shall think nothing of tumbling down stairs
- 35. Lucene English analyzer Input: "Well!" thought Alice to herself "After such a fall as this, I shall think nothing of tumbling down-stairs!“ Output: well thought alic to herself after such a fall as this i shall think noth of tumbl down stair
- 36. Microsoft Natural Language Processing (NLP) technology Differentiating features:Developed for: - Support for 50 languages - Morphological analysis - Spelling variations - Entity recognition
- 37. Microsoft English analyzer Input: "Well!" thought Alice to herself "After such a fall as this, I shall think nothing of tumbling down-stairs!“ Output: well think alice to herself after such a fall as this i shall think nothing of tumble down stair
- 38. Stemming vs. Lemmatization Stemmed Lemmatized think think think thinks think think thinking think think thought though think tumble tumbl tumble tumbles tumbl tumble tumbling tumbl tumble tumbled tumbl tumble nothing noth nothing
- 40. Configuring language analyzers REST API.NET SDK
Editor's Notes
- #5: Search is a primary interaction point for users in content oriented applications Search is becoming the default navigation in many applications In e-commerce settings, up to 83% of navigation occurs through search Good search can make or break: Engagement Return visits Conversions
- #6: User expectations are extremely high Search engine users perfrom ~5 searches per day Years of high quality search engine use mean that users expect results are fast and relevant Spelling correction, instantaneous high counts
- #7: Good search is expensive Search engineers are expensive and hard to find Obviously we aren’t talking about building a search engine here We’re talking about private application search Search engines have very different requirements – much larger scale But the point is still holds that good search is expensive
- #8: Elastic search and solr are great Difference between this and bing Very powerful tools with a great community and loads of documentation But there is a very steep learning curve And although you don’t have to be a search engine expert You will be by the time you’re done Many chose to go this route
- #9: Open source solutions usually mean running a cluster in the cloud somewhere You must now deal with OS image updates Security patches Queries that put your cluster on the floor at 3am Search engine guts like shard management The solution here is usually to hire a guy to babysit: there goes 100k
- #11: Sample online bicycle store – Note search box List 3 main scenarios – eCommerce, user generated content, LOB apps
- #12: Users expect features like type-ahead Google & Bing set the bar
- #13: Results page – A lot going on Matching products
- #14: Facets -- Ways to group and filter results
- #17: Based on user’s location Really important for mobile apps
- #19: Sorting – Sort by fixed criteria Scoring – Tune results to meet business objectives (e.g. – margin)
- #24: Server=tcp:azs-playground.database.windows.net,1433;Database=usgs;User ID=reader;Password=EdrERBt3j6mZDP;Trusted_Connection=False;Encrypt=True;Connection Timeout=30; GeoNamesRI
- #25: Is your data public? Do you want to control how you expose it? Are you using filters for security? Can you send query keys to client directly?
- #40: English is easy Japanese has no spaces German is tough for