Why Taxonomies Are the Secret Sauce

Humans want structure and order. Use taxonomies to categorize and knowledge graphs show relationships.

Presented by

Lief Erickson

Intuitive Stack

www.surveymonkey.com/r/ym9z6rh

Terms Term lists

Taxonomy Ontology

Knowledge graph

Taxonomy is described sometimes as a science and sometimes as an art, but really it’s a battleground.

Bill Bryson, A Short History of Nearly Everything

Taxonomy

Findability and searchability through intelligently tagged content

Tagged content can be surfaced in search

Facets to filter search results

Intelligent tagging can be human or machine. Those who saw Jason Fox’s Bootstrap 5 demo yesterday saw an example of keywords.

Customers of Amazon, Netflix, Spotify, govt reports of invasive species, Puzzle Warehouse are familiar with categories.

App on phone to record bird sounds, send it to the cloud, identify the bird by song. This can be useful for tracking migration, especially for birds that are heard and not seen. The auditory principles and pattern matching connects to a Linnean taxonomy.

I'll show you one example of taxonomy where the results were...unexpected.

Employees need search too

Alignment in understanding, use, approach

Expose extra info for support agents

Feed the search engines

Get your content in shape for AI/NLP

Automatically tag content

Bridge between content silos

Don't. Use an existing one

Define your goal

Analyze what you currently have

Extract existing data

Normalize (Fight. Fight. Fight.)

Governance

RDF, schema.org, Dublin Core, industry-specific

As with any project, knowing what you want to create give you a direction and lets you know when you've arrived.

Some goals you'll hear about today are wanting to find a recipe, starting a smart playlist, and choosing a board game. These will make more sense in a bit.

Gathering term lists, existing taxonomies from other departments, and so on

Like nearly every content project, there's always something to do. Ensure you have a plan for dealing with changes.

Vale and Schematron can do basic (and even some complex) terminology checking

Congree and Acrolinx can do more advanced rule enforcement

PoolParty or OntoText are looking beyond relationships

Do: Focus on concepts rather than terms (preferred terms: auto instead of car).

Don't: Fail to use subject matter experts.

Do: Connect the theoretical with the practical.

Don't: Forget about goals. What is the person looking to achieve?

Do: Use automated tagging. It can expose new concepts for you.

Don't: Think you know it all. Other departments have plenty to offer.

FYI: File systems/folders only have a single dimension. You need metadata to enhance this content to make it more "intelligent."

Words matter

OK Google.

Which movies were these actors in?

Nameless friend

This is a real question a friend asked Google.

What do you think Google responded?

The other end of the spectrum is equally fascinating.

Long, convoluted requests can also be troublesome.

AI technology has to convert speech to text, which is a feat in itself for some dialects, parse for keyword entities, determine intent, search the network, and return results. And it usually has to go to space and back. And we have results in mere moments. That's amazing.

OK Google.

What movies was Harry Styles in?

A fan

[T]he process was conducted iteratively and tested repeatedly until the researcher was satisfied that a consistent information need taxonomy was achieved…

Frummet, A., et al.

Recent research published in Assoc Computing Machinery. Voice-based search, by Frummet, et al, Understanding Cooking-Related Information Needs in Conversational Search

Do I think like a human or a machine? Story about Dan and his immensely convoluted questions!

Understand the environment.

Different than asking a friend.

Multi-tasking person who needs practical information: lists, tasks, etc.

Real-world practical problem that leads to a (technological) solution

Source: Understanding Cooking-Related Information Needs in Conversational Search

Fact or competence?

Amount, time, knowledge, equipment, temperature, or ingredient

Personalization

www.surveymonkey.com/r/ym9z6rh

Finding the secret sauce

Dozens of issues
Each issue has 15-20 recipes
Organized by issue number
Dividers every 10 issues

Tabletop board games

Using classification systems I can look up all sorts of information about a game and its related details.

These classification systems allow me to traverse the graph exploring new content.

What games are similar to GameX?

What game can I play in about 45 minutes with 4 players about pirates? No dice please.

Look at the header. What do we see? What about the Classification section?

https://boardgamegeek.com/geeklist/262047/bgg-game-classification-framework-revised-edition

www.surveymonkey.com/r/ym9z6rh

Video streaming services

Netflix
- Category – not just genre. Nearly 4,000 categories.
- Why is it important that the category be a unique ID rather than a name?
- Fan scraped DVD category codes and posted them online. That kinda locked Netflix into those codes.
- Others like you
- Thumbs up?
- Double thumbs up?
- Star rating 1-5?
- Separate discussion about UX and findability
Amazon Prime

Dramas
From the 1970s
Action from the 1970s
Sci-Fi from the 1970s
From the 1980s
Action
Features
Comedy
Fantasy
Dark Fantasy

Music gets the best of me

Why am I telling you about my music ecosystem and playlists? It's an example of metadata and taxonomies in real, practical use, albeit not a commercial one.

Spotify is a commercial company leveraging the exact same type of data.

My spouse will tell you I spend an inordinate amount of time playing around with the metadata, trying to "clean" it.

Alignment: singer songwriter vs singer-songwriter

On this screen we have...

I prefer browsing by album artist rather than artist.

Why would I choose to have an album artists filter?

Artist vs. ablum artist

Track "featuring" ArtistX

What do I do about Prince, Prince and the Revolution, Prince and the New Power Generation, The Artist Formerly Known as Prince, and his other projects?

By using album artist I create a relationship between Prince many of his incarnations.

Speaking of Prince, let's talk about another way to look at the music library: by location.

Let me show you an example: Minneapolis. What goes into this category? Just artists from Minneapolis? (Does St. Paul count? What about from outside the Twin Cities?)

Prince

Hüsker Dü

Janet Jackson

Candy Dulfer

Lizzo

What about albums recorded in the Twin Cities?

What about songs about the Twin Cities?

Now we're in Twin Cities, where the Mississippi rises and then falls. One is Minneapolis, the other though less famous, is St. Paul.

Genre

Last played

File format: lossless vs lossy

My Library

Rating 1-10

Song length

Has cover

Cover dimensions

Original artist

Year

Artist

Album artist

Producer

Songwriter

Album cover image

Track image

Image size

Album title

Track title

Compilation album

Country

Language

Key

BPM

Mood

Tempo

Seen in concert?

Location recorded

1960s & 70s
1960s & 70s (7+)
7+ from year
7+ from year (RLM safe)
7+ Pick an Artist
Never Played Music (Lief's Crap)
Rated 5 or 6 365
Rated 7+ 12 months
Single Song Repeat

Artist: BT
Title: Nuovo Morceau Subrosa
Acoustic fingerprint: (2775 bytes)
Acoustic ID: f38a4b0e-d654-496a-b2dB-11893c59ba0e
BPM: 136
Comment: Robyn;Lief;2012
Key: A#
Genre: Trance
Mood: Extremely bored
Mood Happy: 10
Mood Relaxed: 86

The decision for business can be based on how perfect of a classifier do you want, based on how quickly it trains and performs, how much data you have, and how does this need to work in production.

Kulsoom Abdullah, PhD

www.surveymonkey.com/r/ym9z6rh

Ontologies and knowledge graphs

Taxonomies can only go so far. If your question requires using information about the RELATIONSHIP between concepts, then an ontology or knowledge graph is necessary.

The connections in ontologies and knowledge graphs tell a story.

An ontology shows the concepts and their relationships and interrelationships. But in a generic, abstract way.

A knowledge graph takes the structure of an ontology and injects specific data.

Metadata plays an important and significant role.

Powerful business tools.

In music it's not the sounds, but the space in between where music is heard.

Semantic content graph guild

Notes from Microsoft presentation?

Semantics. Semantics structure of documents, of content, can give us plenty of information to work with.

JSON-LD

Graph vs. Network?

Knowledge graph showing the romantic connections between characters in The L Word.

The idea is that we're all connected, sometimes in ways we're not expecting and closer in distance than we might think.

Everybody in this room is connect by virtue of being here today, although not romantically.

Six degrees of Kevin Bacon / Separation theory. FB: avg. distance: 4.7

Twitter: avg distance: 3.4. Perhaps that is why we think of these networks as echochambers?

thinkingdocs.com

Taxonomies in graphs
Knowledge graphs at Microsoft
DITA as an ontology source and knowledge graph feeder
Reaching across silos

www.surveymonkey.com/r/ym9z6rh

Survey says

intuitivestack.io/presentations/2022/convex-eu

Press spacebar move sequentially

Press s to see citations and speaker scribbles notes

info@intuitivestack.io

linkedin.com/in/lief-erickson

@Lief_Erickson

Why Taxonomies Are the Secret Sauce

What do you call this shoe?

Foreshadowing

Them's fightin' words

Taxonomy

Customers

Your business

Machines

How to create your taxonomy

Cannot tag it all

Things to consider when creating your own

Words matter

Huh?

Google thinks maybe…

Entities for the win

Google says…

What can I cook with these ingredients?

Information needs

Personalization

What do you call this shoe?

Finding the secret sauce

Recipes on a shelf

Finding a recipe

Browsing or searching online for a recipe

Espresso search

Tabletop board games

New tabletop games by year

Tabletop board games

Board game categories

What do you call this shoe?

Video streaming services

What should I watch?

Ryan Murphy Collection

Anime

Same code, different titles

Music gets the best of me

Play songs rated 7 or higher that haven't been played in six months or longer

Play songs rated 7 or higher that haven't been played in six months or longer that are spouse-approved

Spotify taxonomy

Album artists

Minneapolis?

Paisley Park Studios

R.E.M

Everything But The Girl

Underworld

Categories

Music libraries

Smart playlists

Tracks

Track metadata

In action

What do you call this shoe?

Ontologies and knowledge graphs

The Chart

Oxygen blog graph

DITA-OT domain

DITA-OT knowledge graph

Semantic Content Graph Guild

What do you call this shoe?

ConVEx answers

Find this presentation at

Connect with me