A presentation at Devoxx Ukraine by Alexander Reelsen
Finding relevancy Improving your application search bar! Alexander Reelsen alex@elastic.co | @spinscale
Today’s goal Learn about search, relevancy and the user perspective!
The ultimate search experience
The ultimate search experience: Query intent stock crash 2008 tomorrow’s temperature barack obama age twitter cheap nike hoodie red xl
The ultimate search experience: Query intent Extremely generic Rarely needed for most web applications Many search use-cases are specialized E-Commerce News Social network
E-Commerce Start with a full text search
E-Commerce Start with a full text search Drill down via facetted navigation
E-Commerce Start with a full text search Drill down via facetted navigation Select product
The ultimate search experience: Commercial Intent Scoring is never neutral, influencing factors (E-Commerce) Popularity Profit Availability Previous searches Previous purchases Cross/Upselling products
LinkedIn Search Search for first name Search for last name Prefix search 1st degree > 2nd degree 2nd degree > 3rd degree
LinkedIn Search Rare surname more important than first name
Let’s build such a person search using Elasticsearch!
Product Overview
Elastic Stack
Elasticsearch in one minute Search Engine (FTS, Analytics, Geo), near real-time Distributed, scalable, highly available, resilient Interface: HTTP & JSON
Sample app: Social network Entity: Person Fields: Matching: First & last name Scoring: Employer & direct contacts Meta: Title & image
Sample app: Social network { } “name” : { “last” : “Reelsen”, “first” : “Alexander” }, “employer” : “Elastic”, “title” : “Community Advocate”, “image” : “1”, “contacts” : [ “philippk”, “philipk”, “philippl” ]
Demo
Run node application git clone https://github.com/spinscale/social-network-search-sample cd social-network-search-sample npm install # create static build npm run build npm run start # or for development mode w/ livereload npm run build
Install Elasticsearch # https://www.elastic.co/downloads/elasticsearch wget https://artifacts.elastic.co/downloads/elasticsearch/elasticsearch-7.9.3-darwin-x86_64.tar.gz # wget https://artifacts.elastic.co/downloads/elasticsearch/elasticsearch-7.9.3-linux-x86_64.tar.gz # wget https://artifacts.elastic.co/downloads/elasticsearch/elasticsearch-7.9.3-windows-x86_64.zip tar zxf elasticsearch-7.9.3-darwin-x86_64.tar.gz cd elasticsearch-7.9.3 ./bin/elasticsearch
Install Kibana # https://www.elastic.co/downloads/kibana wget https://artifacts.elastic.co/downloads/kibana/kibana-7.9.3-darwin-x86_64.tar.gz # wget https://artifacts.elastic.co/downloads/kibana/kibana-7.9.3-linux-x86_64.tar.gz # wget https://artifacts.elastic.co/downloads/kibana/kibana-7.9.3-windows-x86_64.zip tar zxf kibana-7.9.3-darwin-x86_64.tar.gz cd kibana-7.9.3 ./bin/kibana Point your browser to http://localhost:5601/
Summary Coming up with a simple query is easy Figuring out user intents is hard Coming up with a proper query will take time and tries Analytics & metrics are vital
Next steps rank_feature : Boost score based on numeric field value vector : Similarity search based on a feature vector
Can you quantify search experience
How to measure search improvements
Continuous feedback loop
Thanks for listening Q&A Alexander Reelsen Community Advocate alex@elastic.co | @spinscale
Resources GitHub Sample Repo rank_feature field type documentation vector field type documentation search_as_you_type field type documentation
Community & Meetups https://community.elastic.co
Discuss Forum https://discuss.elastic.co
Thanks for listening Q&A Alexander Reelsen Community Advocate alex@elastic.co | @spinscale
You may have heard, that you ‘just’ need to add a search bar to your application and people will use that input field the moment it is added. Well, not so fast! Coming up with a google like search experience is not an easy task! Finding the right balance between ease-of-use, relevancy for your users, relevancy for your platform, usability, precision, recall, scoring, keeping the quality and the ability for your users to make sense of those search results will quick evolve into a ongoing task.
In this talk we will built a small search application backed by Elasticsearch, which could be solved by a single query. But is this really enough? No, to improve the search experience we will improve our queries step by step to make sure our users are happy and delve deeper into full text search features while doing so.
You can find the GitHub repo with a sample Svelte application at https://github.com/spinscale/social-network-search-sample
for free. You
can too.