Connect with us


Optimizing For The Human Mind With Machine Learning



Optimizing For The Human Mind With Machine Learning


We’ve been talking with search industry pros and innovators about persistent challenges, trending opportunities, and the technologies people and companies are using to stay relevant in competitive search results.

One trend driving massive advancements in search technology is the shift from keywords to data that better represents the meaning of the query, and what’s known about it.

Keyword search has been driving content discovery since 1230 AD. That’s when French cardinal and biblical commentator Cardinal Hugh de St Cher completed the first known index in history.

Vector search marks a major shift from this traditional method of information retrieval to a future in which all of the complex data that makes up modern content assets can be put to work.

So what do you need to know about it right now?

We reached out to Edo Liberty, the former head of Amazon’s AI lab and now CEO of Pinecone, for a primer on vector search and why you may want to have the associated technologies on your radar.

We asked Liberty:

  • How will vector search redefine traditional keyword search?
  • How would you explain vector search to a 5-year-old?
  • What are some of the challenges that you faced using ML algorithms for Amazon Web Services (AWS) customers, and how did you overcome them?
  • What is Pinecone and what does it do?
  • What tips or advice do you have for SEO beginners who are just stepping into the world of ML and AI?

Let’s start with this – why is natural language processing (NLP) so important to the future of SEO, and how can marketers prepare for what’s next?

We’ve Burned The Ships Of Keyword Search

Edo Liberty: “Just as SEOs mastered the PageRank algorithm, they now need to know about NLP in order to succeed and beat the competition.

Unlike PageRank, however, the field of NLP is growing fast and has thousands of contributors.

It’s going to take more effort than following Matt Cutts (from Google) on Twitter and tracking SERP changes.

Thankfully, although NLP is a more complicated topic, it is not shrouded in mystery like PageRank is.

A lot of the work on NLP is being done in the open, with free and abundant research papers, open-source software, and no-cost online courses on NLP.

One thing is clear about NLP: It’s here to stay.

It’s far from perfect, but it’s improving fast, and the big tech companies have burned the ships of keyword search and there is no going back.”

Vector Search Enables Us To Search The Way We Speak

How will vector search redefine traditional keyword search?

Edo Liberty:Vector search doesn’t redefine keyword search; it replaces it whole-cloth.

Instead of working with keywords – and their synonyms and misspellings – vector search works with vector embeddings.

That’s a piece of data that represents the meaning of the search phrase along with other information known about the query or the user.

(To a human, the vector embedding is unrecognizable and just looks like a long array of numbers.)

This representation of the search phrase and the user is then used to sort through massive collections of embeddings that represent other content and user preferences to find the most relevant result.

From the user’s perspective, this means they can search the way they speak.

They no longer need to learn the quirks and syntaxes of search engines.

From the SEO’s perspective, this means they can truly focus on themes and topics without worrying about precise keywords.”

How Would You Explain Vector Search To A 5-year-old?

Edo Liberty: “Our article explaining vector search basics comes close.

The ELI5 version, as I’ve practiced on my own family, is this: If I say ‘Italian food,’ you might think of pizza or pasta.

You’ve learned that those things are related because you remember eating pizza at an Italian restaurant or learning that pasta is popular in Italy.

But a computer never learned that. So the phrase ‘Italian food’ means exactly that and doesn’t contain information to say it’s related to pasta or pizza.

So, when I ask a computer to search for an ‘Italian restaurant,’ it might leave out the pizza places.

Machine learning is a way of helping computers understand the meaning of what we say or type.

And vector search is a way for those computers to search through everything they know, based on meaning and not exact words.

So now, if I ask the computer to recommend an Italian place, it might suggest your favorite pizza place just like you would.

Organizations can finally focus on creating and organizing content for humans.

There are many thousands of scientists and engineers working tirelessly to make ML and NLP resemble the human mind.

Do you really want to go against that? The winning strategy for SEO is to optimize for the human mind.”

Overcoming Challenges In Machine Learning

What are some of the challenges that you faced using ML algorithms for Amazon Web Services (AWS) customers, and how did you overcome them?

Edo Liberty: “I can’t speak about specific projects or challenges from AWS. I can say more broadly, from my experience, I saw that ML algorithms are no longer the bottlenecks.

To be sure, they are far from perfect, and there’s a lot of work to be done, but that work is happening at breakneck speed.

The next challenge is in running those algorithms at the scale needed to support consumer products and enterprise applications.

Those representations I mentioned earlier, vector embeddings, are computationally costly to search through.

An index of just 1M items (vector embeddings) already requires specialized software along with careful tuning; an index of 100M items requires specialized software and infrastructure; an index of 1B or more items requires you to be Google or Amazon.

(As an aside, this is why I started Pinecone: To make it easy for engineering teams to add vector search to their applications.)”

What Is Pinecone?

What is Pinecone and what does it do?

Edo Liberty: Today, Pinecone makes it easy for engineers to build fast, fresh, and filtered vector search into their applications.

It gives engineering teams the search infrastructure needed to run vector search at scale, all packaged in a managed service with an easy API.

(We’ve dropped the version numbers because the releases come fast, and because as a managed service, users always get the latest version and don’t need to worry about updates.)

Working with algorithms is extremely fun and absolutely worth the challenges.

With vector search, we’re at the intersection of cutting-edge algorithms, database architectures, and serverless applications.

And, we get to see our customers apply this technology to products that are revolutionizing both consumer and enterprise applications like semantic search, recommendation systems, IT security, wearables, computer vision, and more.

Getting Started In ML & AI

What tips or advice do you have for SEO beginners who are just stepping into the worlds of ML and AI?

Edo Liberty: “Don’t feel intimidated. Even the brightest researchers in this field are ‘figuring things out.’

Learning about AI/ML beyond the surface-level articles will make you a better SEO professional, and there are plenty of free resources that help you do that.

For those interested in careers in this field, we are currently hiring across all teams: engineering, research, customer success, sales, marketing, and operations.

More Resources:

Featured Image: Courtesy of Pinecone


Source link


A Comprehensive Guide To Marketing Attribution Models



A Comprehensive Guide To Marketing Attribution Models

We all know that customers interact with a brand through multiple channels and campaigns (online and offline) along their path to conversion.

Surprisingly, within the B2B sector, the average customer is exposed to a brand 36 times before converting into a customer.

With so many touchpoints, it is difficult to really pin down just how much a marketing channel or campaign influenced the decision to buy.

This is where marketing attribution comes in.

Marketing attribution provides insights into the most effective touchpoints along the buyer journey.

In this comprehensive guide, we simplify everything you need to know to get started with marketing attribution models, including an overview of your options and how to use them.

What Is Marketing Attribution?

Marketing attribution is the rule (or set of rules) that says how the credit for a conversion is distributed across a buyer’s journey.

How much credit each touchpoint should get is one of the more complicated marketing topics, which is why so many different types of attribution models are used today.

6 Common Attribution Models

There are six common attribution models, and each distributes conversion value across the buyer’s journey differently.

Don’t worry. We will help you understand all of the models below so you can decide which is best for your needs.

Note: The examples in this guide use Google Analytics 4 cross-channel rules-based models.

Cross-channel rules-based means that it ignores direct traffic. This may not be the case if you use alternative analytics software.

1. Last Click

The last click attribution model gives all the credit to the marketing touchpoint that happens directly before conversion.

Last Click helps you understand which marketing efforts close sales.

For example, a user initially discovers your brand by watching a YouTube Ad for 30 seconds (engaged view).

Later that day, the same user Googles your brand and clicks through an organic search result.

The following week this user is shown a retargeting ad on Facebook, clicks through, and signs up for your email newsletter.

The next day, they click through the email and convert to a customer.

Under a last-click attribution model, 100% of the credit for that conversion is given to email, the touchpoint that closed the sale.

2. First Click

The first click is the opposite of the last click attribution model.

All of the credit for any conversion that may happen is awarded to the first interaction.

The first click helps you to understand which channels create brand awareness.

It doesn’t matter if the customer clicked through a retargeting ad and later converted through an email visit.

If the customer initially interacted with your brand through an engaged YouTube view, Paid Video gets full credit for that conversion because it started the journey.

3. Linear

Linear attribution provides a look at your marketing strategy as a whole.

This model is especially useful if you need to maintain awareness throughout the entire buyer journey.

Credit for conversion is split evenly among all the channels a customer interacts with.

Let’s look at our example: Each of the four touchpoints (Paid Video, Organic, Paid Social, and Email) all get 25% of the conversion value because they’re all given equal credit.

4. Time Decay

Time Decay is useful for short sales cycles like a promotion because it considers when each touchpoint occurred.

The first touch gets the least amount of credit, while the last click gets the most.

Using our example:

  • Paid Video (YouTube engaged view) would get 10% of the credit.
  • Organic search would get 20%.
  • Paid Social (Facebook ad) gets 30%.
  • Email, which occurred the day of the conversion, gets 40%.

Note: Google Analytics 4 distributes this credit using a seven-day half-life.

5. Position-Based

The position-based (U-shaped) approach divides credit for a sale between the two most critical interactions: how a client discovered your brand and the interaction that generated a conversion.

With position-based attribution modeling, Paid Video (YouTube engaged view) and Email would each get 40% of the credit because they were the first and last interaction within our example.

Organic search and the Facebook Ad would each get 10%.

6. Data-Driven (Cross-Channel Linear)

Google Analytics 4 has a unique data-driven attribution model that uses machine learning algorithms.

Credit is assigned based on how each touchpoint changes the estimated conversion probability.

It uses each advertiser’s data to calculate the actual contribution an interaction had for every conversion event.

Best Marketing Attribution Model

There isn’t necessarily a “best” marketing attribution model, and there’s no reason to limit yourself to just one.

Comparing performance under different attribution models will help you to understand the importance of multiple touchpoints along your buyer journey.

Model Comparison In Google Analytics 4 (GA4)

If you want to see how performance changes by attribution model, you can do that easily with GA4.

To access model comparison in Google Analytics 4, click “Advertising” in the left-hand menu and then click “Model comparison” under “Attribution.”

Screenshot from GA4, July 2022

By default, the conversion events will be all, the date range will be the last 28 days, and the dimension will be the default channel grouping.

Start by selecting the date range and conversion event you want to analyze.

GA4 model comparison_choose event and date rangeScreenshot from GA4, July 2022

You can add a filter to view a specific campaign, geographic location, or device using the edit comparison option in the top right of the report.

GA4 Model comparison filterScreenshot from GA4, July 2022

Select the dimension to report on and then use the drown-down menus to select the attribution models to compare.

GA4 model comparison_select dimensionScreenshot from GA4, July 2022

GA4 Model Comparison Example

Let’s say you’re asked to increase new customers to the website.

You could open Google Analytics 4 and compare the “last-click” model to the “first-click” model to discover which marketing efforts start customers down the path to conversion.

GA4 model comparison_increase new customersScreenshot from GA4, July 2022

In the example above, we may choose to look further into the email and paid search further because they appear to be more effective at starting customers down the path to conversion than closing the sale.

How To Change Google Analytics 4 Attribution Model

If you choose a different attribution model for your company, you can edit your attribution settings by clicking the gear icon in the bottom left-hand corner.

Open Attribution Settings under the property column and click the Reporting attribution model drop-down menu.

Here you can choose from the six cross-channel attribution models discussed above or the “ads-preferred last click model.”

Ads-preferred gives full credit to the last Google Ads click along the conversion path.

edit GA4 attribution settingsScreenshot from GA4, July 2022

Please note that attribution model changes will apply to historical and future data.

Final Thoughts

Determining where and when a lead or purchase occurred is easy. The hard part is defining the reason behind a lead or purchase.

Comparing attribution modeling reports help us to understand how the entire buyer journey supported the conversion.

Looking at this information in greater depth enables marketers to maximize ROI.

Got questions? Let us know on Twitter or Linkedin.

More Resources:

Featured Image: Andrii Yalanskyi/Shutterstock

Source link

Continue Reading

Subscribe To our Newsletter
We promise not to spam you. Unsubscribe at any time.
Invalid email address