SearchLove San Diego 2017 | Will Critchlow | Knowing Ranking Factors Won't Be Enough: How To Avoid...
Transcript of SearchLove San Diego 2017 | Will Critchlow | Knowing Ranking Factors Won't Be Enough: How To Avoid...
Knowing ranking factors won’t be enough
How to avoid losing your job to a robot
@willcritchlow
I’m going to tell you about a robot that understands ranking factors
better than any of you
...but before I get to that, let’s look at a bit of history...
The other day I searched:
Unsurprisingly, I got an answer
But it got me thinking about how, in 2009, the results would have looked more like this.
In 2009, it would have looked more like this.
With every title containing the keyphrase.
In 2009, it would have looked more like this.
With every title containing the keyphrase.
Most at the beginning.
OK. Maybe wikipedia would have been #1.
We used to have a pretty good understanding of ranking factors
My mental model for c. 2009 ranking factors had three different modes:
My mental model for ~2009 ranking factors had three different modes:
One in the hyper-competitive
head
One in the
competitive
mid-tail
...and o
ne in th
e
long-t
ail
One in the hyper-competitive
head
Tons of perfectly on-topic pages to choose from
One in the hyper-competitive
head
So pick only perfectly-on-topic pages
One in the hyper-competitive
head
...and rank by authority (*)
(*) Page authority, but the domain inevitably factors into that calculation. This is why
so many homepages ranked
One in the hyper-competitive
head
This resulted in a mix of homepages of mid-size sites, and inner pages on huge sites
One in the hyper-competitive
head
But the general way to move up was through increased authority
One in the hyper-competitive
head
Kind of search result
Pages ranking To move up...
Head Homepages of mid-size sites and inner pages of massive sites. All perfectly-targeted.
Improve authority.
Mid-tail
Long-tail
One in the hyper-competitive
head
One in the
competitive
mid-tail
Wealth of ROUGHLY on-topic pages to choose from
One in the
competitive
mid-tail
PERFECTLY on-topic could do well even on a relatively weak site
One in the
competitive
mid-tail
Rank the roughly on-topic pages by
authority x “on-topicness”
One in the
competitive
mid-tail
Move up with better targeting or more authority
One in the
competitive
mid-tail
Kind of search result
Pages ranking To move up...
Head Homepages of mid-size sites and inner pages of massive sites. All perfectly-targeted.
Improve authority.
Mid-tail Perfectly on-topic pages on relatively weak sites plus roughly on-topic on bigger sites.
Improve targeting or authority.
Long-tail
One in the
competitive
mid-tail
One in the hyper-competitive
head
...and o
ne in th
e
long-t
ail
In the long-tail, a site of arbitrary weakness could rank if it was the most relevant
...and o
ne in th
e
long-t
ail
Otherwise, massive sites rank with off-topic pages that mention something similar
...and o
ne in th
e
long-t
ail
Generally, move up with better targeting
...and o
ne in th
e
long-t
ail
Kind of search result
Pages ranking To move up...
Head Homepages of mid-size sites and inner pages of massive sites. All perfectly-targeted.
Improve authority.
Mid-tail Perfectly on-topic pages on relatively weak sites plus roughly on-topic on bigger sites.
Improve targeting or authority.
Long-tail Arbitrarily-weak on-topic pages and roughly-targeted deep pages on massive sites.
Improve targeting.
Kind of search result
Pages ranking To move up...
Head Homepages of mid-size sites and inner pages of massive sites. All perfectly-targeted.
Improve authority.
Mid-tail Perfectly on-topic pages on relatively weak sites plus roughly on-topic on bigger sites.
Improve targeting or authority.
Long-tail Arbitrarily-weak on-topic pages and roughly-targeted deep pages on massive sites.
Improve targeting.
So that was
~2009
It’s not so simple any more.Google is harder to understand these days.
PageRank(the first algorithm to use the link structure
of the web)
We know how we got to ~2009...
Information retrieval
PageRank
Information retrieval
PageRankOriginal research
Information retrieval
PageRankOriginal research
TWEAKS
...with growing complexity in subsequent years
When Amit left Google, there was a fascinating thread on Hacker News in discussion of this article
Particularly this comment from a user called Kevin Lacker (@lacker):
I was thinking about it like it was a math puzzle and if I just thought
really hard it would all make sense.
-- Kevin Lacker (@lacker)
Hey why don't you take the square root?
-- Amit Singhal according to Kevin Lacker (@lacker)
oh... am I allowed to write code that doesn't make any sense?
-- Kevin Lacker (@lacker)
Multiply by 2 if it helps, add 5, whatever, just make things work and we can make it make sense
later.
-- Amit Singhal according to Kevin Lacker (@lacker)
Why does this make the algorithm so hard to understand?
High-dimension
Non-linear
Discontinuous
3 big reasons:
High-dimension
Non-linear
Discontinuous
High-dimension
Non-linear
Discontinuous
High-dimension
Non-linear
Discontinuous
You might know what any one of the levers does, but they can
interact with each other in complex ways
This is what a high-dimensional function looks like
High-dimension
Non-linear
Discontinuous
We sell custom cigar humidors. Our custom cigar humidors are handmade. If you’re thinking of buying a custom cigar
humidor, please contact our custom cigar humidor specialists at
What this needs is another mention of [cigar humidors]
With no mentions of [cigar] or [humidor] this page would be unlikely to rank
And yet you can clearly go too far, and have the effect turn negative.
This is called nonlinearity.
The cigar example is taken directly from Google’s quality guidelines.
High-dimension
Non-linear
Discontinuous
Discontinuities are steps in the function
Think about so-called “over-optimization” tipping points
Let’s put all this togetherinto a practical example:
Think about category pages:Do you recommend removing “SEO text”?
We’ve tested it, so we know the answer.
If you said “yes”, congratulations(+3.1% organic sessions in a split-test)
Unless you’re responsible for this siteNo effect / possible negative effect
No, but I’m still pretty good at this
You’re thinking this to yourself right now.
I promised to tell you about a robot that is better than even
experienced SEOs...
Well. It turns out all we needed was a coin to flip. You’re all fired.
It’s only going to get worse under Sundar Pichai
Who knows who this is?(This is the only CC-licensed photo of him on the internet)
ENHANCEWhat about now?
John Giannandrea - Google’s head of searchSundar’s choice to lead search after Amit. Previously running machine learning.
...and of course Jeff Dean is doing Jeff Dean things(c.f. Chuck Norris)
Jeff Dean puts his pants on one leg at a time, but if he had more legs,
you would see that his approach is O(log n).
Source: Jeff Dean facts
Once, in early 2002, when the search back-ends went down, Jeff
Dean answered user queries manually for two hours.
Result quality improved markedly during this time
When Jeff Dean goes on vacation, production services across Google mysteriously stop working within a
few days.
This was reportedly actually true
The original Google Translate was the result of the work of hundreds of engineers over 10 years.
Director of Translate, Macduff Hughes said that it sounded to him as if maybe they could pull off a neural-network-based replacement in three years.
Jeff Dean said “we can do it by the end of the year, if we put our minds to it”.
Hughes: “I’m not going to be the one to say Jeff Dean can’t deliver speed.”
A month later, the work of a team of 3 engineers was tested against the existing system. The improvement was roughly equivalent to the improvement of the old system over the previous 10 years.
Hughes sent his team an email. All projects on the old system were to be suspended immediately.
[Read the whole story ]
Background reading: (backchannel, bloomberg)
How to avoid losing your job to a robot
This is what you promised, Will.
Let’s start by understanding
some robot weaknesses
What’s this?
Ooh. Ooh.
I know this one.
-- robot
“It’s a leopard. I’m like 99% sure.”
Computers are better than humans at classification, but struggle with adversaries
Read more about this here -- Cheetah, Leopard, Jaguar
Lesson:
We expect adversarial abilities to take a step backwards
They will remain good at classifying bad links but will be likely to fall
prey to weird outcomes in adversarial situations
Example:
Remember Tay, the Microsoft chatbot that Twitter taught to be
racist and sexist in less than a day?
Read more here
We’re going to see new kinds of bugs
Rules of ML [PDF] outlines engineering lessons from getting ML into production at Google
Example lesson: There will be silent failures
“This is a problem that occurs more for machine learning systems than for other
kinds of systems. Suppose that a particular table that is being joined is no longer
being updated. The machine learning system will adjust, and behavior will
continue to be reasonably good, decaying gradually. Sometimes tables are found
that were months out of date, and a simple refresh improved performance more
than any other launch that quarter! For example, the coverage of a feature may
change due to implementation changes: for example a feature column could be
populated in 90% of the examples, and suddenly drop to 60% of the examples.
Play once had a table that was stale for 6 months, and refreshing the table alone
gave a boost of 2% in install rate. If you track statistics of the data, as well as
manually inspect the data on occassion, you can reduce these kinds of failures.”
Example lesson: There will be silent failures
“This is a problem that occurs more for machine learning systems than for other
kinds of systems. Suppose that a particular table that is being joined is no longer
being updated. The machine learning system will adjust, and behavior will
continue to be reasonably good, decaying gradually. Sometimes tables are found
that were months out of date, and a simple refresh improved performance more
than any other launch that quarter! For example, the coverage of a feature may
change due to implementation changes: for example a feature column could be
populated in 90% of the examples, and suddenly drop to 60% of the examples.
Play once had a table that was stale for 6 months, and refreshing the table alone
gave a boost of 2% in install rate. If you track statistics of the data, as well as
manually inspect the data on occassion, you can reduce these kinds of failures.”
Example lesson: There will be silent failures
“This is a problem that occurs more for machine learning systems than for other
kinds of systems. Suppose that a particular table that is being joined is no longer
being updated. The machine learning system will adjust, and behavior will
continue to be reasonably good, decaying gradually. Sometimes tables are found
that were months out of date, and a simple refresh improved performance more
than any other launch that quarter! For example, the coverage of a feature may
change due to implementation changes: for example a feature column could be
populated in 90% of the examples, and suddenly drop to 60% of the examples.
Play once had a table that was stale for 6 months, and refreshing the table alone
gave a boost of 2% in install rate. If you track statistics of the data, as well as
manually inspect the data on occassion, you can reduce these kinds of failures.”
That document also has a section on trying to understand what the machines are doing
But human explainability may not even be possible
Not every concept a neural network uses fits neatly into a concept for
which we have a word. It’s not clear this is a weakness per se, but...
...this means that engineers won’t always know more than we do
about why a page does or doesn’t rank
The big knowledge gap of the future is data - clickthrough rates,
bounce rates etc.
As Tom Capper said, engineers’ statements can already be misleading
...and remember the confounding split-testsIt’s already not always as simple as “feature X is good”
Which all means we may need to be more independent-minded and do more of our own research
So how do we fight back?
Michael Lewis’ latest book is about Kahneman and Tversky spelling.
It recounts a story about a piece of medical software that existed in the 1960s.
It was designed to encapsulate how a range of doctors diagnosed stomach cancer from x-rays.
It proceeded to outperform those same doctors despite only containing their expertise.
Real people have biases, and fool themselves.
Encapsulate your own expert knowledge.
At Distilled, we use a methodology we call the balanced digital scorecard.
This encapsulates our beliefs about how to build a high-performing business.
Applying it helps avoid our own biases.
Also, while we are talking about books, The Checklist Manifesto is an important part of avoiding the same cognitive biases.
Focus on consulting skills
I’ve written a few things about this (DistilledU module, writing better business documents, using split-tests to consult better).
Use case studies and creativity. Computers are better at diagnosis than cure.
This means: getting things done, convincing organizations, applying general knowledge, learning new things.
We are going to need to be better than ever at debugging things.
I wrote about debugging skills for non-developers here.
A lot of the story of enterprise consulting is going to be about figuring out why things have gone wrong in the face of sparse or incorrect information from Google.
Disregard expert surveys
Firstly, there are all the problems outlined in the search result pairs study - both in the ability of experts to understand factors, and in your ability to use the information even if they do.
Secondly, they are broken with another bias called the “law of small numbers” from Lewis’ book.
PS - I say this as a participant in many of them
Me
Equally, building your digital strategy on what Google tells you to do will become an even worse idea than it already is.
This is why we have been investing so much in split-testing
Check out www.distilledodn.com if you haven’t already.
The team will be happy to demo for you.
We’re now serving ~1.5 billion requests / month, and recently published information covering everything from response times to our +£100k / month split test.
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
4. We remain relevant by:
a. Using methodologies and checklists to capture human capabilities and
avoid our biases
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
4. We remain relevant by:
a. Using methodologies and checklists to capture human capabilities and
avoid our biases
b. Becoming great consultants and change agents
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
4. We remain relevant by:
a. Using methodologies and checklists to capture human capabilities and
avoid our biases
b. Becoming great consultants and change agents
c. Debugging the heck out of everything
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
4. We remain relevant by:
a. Using methodologies and checklists to capture human capabilities and
avoid our biases
b. Becoming great consultants and change agents
c. Debugging the heck out of everything
d. Avoiding being misled by experts or Google
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
4. We remain relevant by:
a. Using methodologies and checklists to capture human capabilities and
avoid our biases
b. Becoming great consultants and change agents
c. Debugging the heck out of everything
d. Avoiding being misled by experts or Google
e. Testing!
Oh, and one more thing
What about that robot I promised you?
The coin flip wasn’t really it
keras.io
The specifics of DeepRank
Gather and process
training data
We started with a broad range of unbranded keywords from our STAT rank tracking.
For each of the URLs ranking in the top 10, we gathered key metrics about the domain and page - both from direct crawling and various APIs.
We turned this into a set of pairs of URLs {A,B} with their associated keyword, metrics, and their rank ordering.
The specifics of DeepRank
Gather and process
training data
We started with a broad range of unbranded keywords from our STAT rank tracking.
For each of the URLs ranking in the top 10, we gathered key metrics about the domain and page - both from direct crawling and various APIs.
We turned this into a set of pairs of URLs {A,B} with their associated keyword, metrics, and their rank ordering.
The specifics of DeepRank
We have so far trained on just 10 metrics for a relatively small sample (hundreds) of keywords.
Our current version is only a few layers deep with only 10 hidden dimensions.
The current training samples 30 pairs at a time and trains against them for 500 epochs.
Train the model
Gather and process
training data
The specifics of DeepRank
The next task is to get way more metrics for thousands of keywords.
This will enable us to train a much deeper model for much longer without overfitting.
We also have some more hyperparameter tuning to do,
Model
Train the model
Gather and process
training data
To run the model, we input a pair of pages with their associated metrics.
New input
Model
New input
We get back a probability of page A outranking page B.
Model
Probability-weighted
predictions
New input
The goal is a winning combination of human and machine
Human + computer beats computer (for now)
Let’s recap
1. Even in a world of 200+ “classical” ranking factors, humans were bad at
understanding the algorithm
2. Machine learning will make this worse, and is accelerating under Sundar
3. There are things computers remain bad at, and rankings will become more
opaque even to Google engineers
4. We remain relevant by:
a. Using methodologies and checklists to capture human capabilities and
avoid our biases
b. Becoming great consultants and change agents
c. Debugging the heck out of everything
d. Avoiding being misled by experts or Google
e. Testing!
5. Human + robot is the only thing that has a chance of beating the robots
Questions: @willcritchlow
Image credits
● Mobius strip
● Confusion
● Signal box
● Cigar
● Discontinuity
● Confidence
● Burt Totaro
● Sundar Pichai
● John Giannandrea
● Chuck Norris
● Jeff Dean
● Fencing
● Keyboard
● Go
● Robot
● Leopard print sofa
● Leopard
● Bug
● Lego robots
● Iron Man
● San Diego