Algolia introduction - DEMO and Ranking Formula

Slide 1

Slide 1 text

Algolia introduction DEMO and Ranking Formula Eiji Shinohara Senior Manager, Solutions Engineer - Japan [email protected] @shinodogg

Slide 2

Slide 2 text

Who am I? ● Eiji Shinohara ○ Twitter: @shinodogg ○ Blog: https://shinodogg.com ○ 1st Japan-based employee at

Slide 3

Slide 3 text

Who am I? ● Eiji Shinohara ○ Worked at Rakuten ■ Learned English - Eigo Chotto Dekiru ■ My Search engineering SENSEIs: Minoru san & Kazu san (now at Mercari) ○ Worked at Amazon Web Services ■ In charge of Japanese Startups ■ Used to visit Mercari office (1st floor is MIZUHO Bank) years ago

Slide 4

Slide 4 text

Our Investors $74M Funding Founded 2012 Employees 350+ Our Business Algolia in a nutshell 100 Countries Customers 7,000+ Search queries/mo 60B+

Slide 5

Slide 5 text

17 Regions 70+ Data centers Offices Infrastructure Regions 60B+ Searches/mo 150B+ API calls/mo Global Availability across 17 regions worldwide

Slide 6

Slide 6 text

Our 7,000+ customers across 100 countries trust Us Retail Media Other industries

Slide 7

Slide 7 text

In Japan

Slide 8

Slide 8 text

https://www.globenewswire.com/news-release/2019/05/22/1840823/0/en/Algolia-Announces-Global-Exp ansion-Into-Japan.html

Slide 9

Slide 9 text

DEMO

Slide 10

Slide 10 text

Ranking Formula

Slide 11

Slide 11 text

TF-IDF Term Frequency

Slide 12

Slide 12 text

TF-IDF Term Frequency

Slide 13

Slide 13 text

TF-IDF Term Frequency Inv. Doc. Frequency

Slide 14

Slide 14 text

TF-IDF Term Frequency Inv. Doc. Frequency 0.92 0.44

Slide 15

Slide 15 text

Useful when We don’t know the type of data in advance Content is completely different from one document to the other No popularity taken into account TF-IDF Term Frequency Inv. Doc. Frequency

Slide 16

Slide 16 text

Relevance in Algolia Structured Data Data that already exists in your Database You know this data (what is useful in it for search, what is not)

Slide 17

Slide 17 text

Title or description? First or last word? Exact? How many follower/likes?

Slide 18

Slide 18 text

Ranking Formula criteria Not based on a TF-IDF variation Uses 8 rules to evaluate/measure relevance

Slide 19

Slide 19 text

Ranking Formula criteria 1. Typo 2. Geo (if applicable) 3. Words (if applicable) 4. Filters 5. Proximity 6. Attribute 7. Exact 8. Custom

Slide 20

Slide 20 text

Tie-breaking algorithm

Slide 21

Slide 21 text

Ranking formula The Tie Breaking Algorithm Lego bricks example 3 attributes or qualities : ● Shape ● Color ● Studs

Slide 22

Slide 22 text

Ranking formula The Tie Breaking Algorithm ● Rank the set by applying a series of criteria ● Eliminate ties, criterion per criterion ● The order we apply criterion has a big impact on the ﬁnal ranking of the whole set Goal

Slide 23

Slide 23 text

Tie-breaking Algorithm ● Color (Blue > Red > White) Ranking Formula A ● Studs (less is best) ● Shape ( > )

Slide 24

Slide 24 text

Tie-breaking Algorithm Color Shape Studs (Blue > Red > White) Start Rect > Round Less is best

Slide 25

Slide 25 text

Tie-breaking Algorithm ● Shape ( > ) Ranking Formula B ● Color (Blue > Red > White) ● Studs (less is best)

Slide 26

Slide 26 text

Tie-breaking Algorithm Color (Blue > Red > White) Shape Rect > Round Studs Less is best Start

Slide 27

Slide 27 text

Tie-breaking Algorithm ● Color (Blue > Red > White) ● Shape ( > ) Ranking Formula A ● Studs (less is best) ● Shape ( > ) ● Color (Blue > Red > White) Ranking Formula B ● Studs (less is best)

Slide 28

Slide 28 text

ANY QUESTIONS? Thank you. Eiji Shinohara Senior Manager, Solutions Engineer - Japan [email protected] @shinodogg