AI-powered classifications vs Keywords. Part 2/2: Evolution over time.

For content selection: AI-powered classifications can sense Editorial Orientations AND Evolution over time. Keywords cannot.

For years, access to knowledge was all about the presence or absence of keywords to trigger the selection of content: A 1-dimensional access, keywords based, to knowledge. Linear. Limited to 0 (absent) or 1 (present).

Last week, we covered the first advantage of AI-Powered Classifications vs keywords based selection, Editorial Orientations, and showed how the same event, on 3 different publications can have different Editorial Orientations.

This is an additional dimension to access knowledge.

Read postLet’s now have a look at a 3rd dimension: Sensitivity over time.

Perception of an event evolves with time, so do our AI-Powered classifications.

France has been through a lot of social movements with the pension reform the French government is pushing for.

From the beginning of the protests until now, the perception has evolved.

Let’s look at the same article and how AI classifies it at two different times.

This article was published on Dec 10th 2019:

Pension reform: “It would be a misdiagnosis to talk about minced runs.”  [google Translation] (Réforme des retraites: “Ne parler que de parcours hachés serait une erreur de diagnostic”)

On Dec 10th, top classification was:

We are at the beginning of the movement, Employment and Unemployment is the top classification.

On Dec 31st, top classifications are now:

3 weeks later, the very same article with the very same content is classified as Senior first, then Social Assistance and, now in 3rd, Employment and Unemployment

Clearly, after 3 weeks of protests, Aging and Social are topping the Employment dimension.

How can AI-Powered Classification do this?

In a previous post, we explained how our AI worked:

How our AI-powered classification works.

Every new article is classified as follow:

Which means the day the article is published, we use Classifications Datasets (aka bags of words) on that very day.

Classification Datasets are also updated to sync with every single classification and sense the depth of expertise over time. This means some words can be in and out and with a different weight over time. This means classifications are set, by default, for the day an article is published but can be re-run on a different day and produce a different classification. Like in real life, your perception of something evolves with time.

Why it matters.

Simply because time is a vital dimension of perception.

Simply relying on the presence of keywords to select content for analytics, expose your brand via advertising etc… is dangerous.

What’s true at publication time might not be at analytics time, or advertising time…

In the example above, you may or may not want articles about “Seniors”. At publication time, the article was under the radar, 3 weeks later it is classified as “Seniors”. Is it still where your brand wants to be exposed? are those content the one you want to analyze today? do those articles matter for the education of your teams?

Relying on keywords that are present in content forever, not only does not give you the orientation of the content but is not sensitive to the evolution of perception. And as we know, in Marketing:

Perception is reality.

Questions? Ask!

 

AI-powered classifications vs Keywords. Part 1/2: Editorial Orientations detection.[updated]

Going beyond a 1-dimensional access to knowledge.

For years, access to knowledge has been ruled by keywords presence. Search engines, corpus selection for business intelligence, DSPs for online advertising, Brand Safety, Watch alerts…

All is about the presence or absence of keywords to trigger the selection of content: A 1-dimensional access, keywords based, to knowledge. Linear. Limited to 0 (absent) or 1 (present).

Keywords presence does not sense angles, subtlety and orientations taken by the author (nor the sensitivity over time. Today’s meanings are the same any other day).

For example, the presence of “Christmas gift” might be “ok” but is it in a context of “Military defense” and “Weapon”? Can you maintain queries excluding all related, always evolving dictionaries of synonyms and be sure your brand won’t be exposed?

After all, a word can have several meanings depending on its context and the time it is read. AI-Powered Classifications are the solution:

AI-Powered Classifications are adding 2 more dimensions: Editorial orientations and timing context.

Today, we will focus on editorial orientations detection.

Next week, we will explain the sensitivity to the time of publication.

[update] The second part is now published:

AI-powered classifications vs Keywords. Part 2/2: Evolution over time.

Both, AI-Powered Classifications and keywords based selections are unbiased, universal and up-to-date. Because TrustedOut is AI-Powered, our machine learning guarantees the same non-humain, machine powered benefits.

Editorial Orientations Example:
1 event, 2 countries, 3 articles, 10 classifications.

The event: let’s take North Korea announcing a special Gift to the US.

The 2 countries: We then selected 3 articles from a Google Search on “North Korea Gift” for the US and “Coree du Nord Cadeau” for France.

The 3 articles: we randomly picked USAToday, CBSNews and Le Figaro.

Here are the top 10 classifications TrustedOut came up with. For each we’ve added how the media is spotted for its Political Orientations (beta)

USAToday

Vase or missiles? US awaits Christmas ‘gift’ from North Korea’s Kim

1 General › Politics › Diplomacy
2 General › Politics › International
3 Industries › Aerospace And Defense › Weapon
4 General › Politics › Military Defense
5 General › Politics › Civil Defense
6 Industries › Energy › Nuclear Power
7 Industries › Aerospace And Defense › Naval System
8 General › Politics › Administration
9 Industries › Aerospace And Defense › Aerospace Systems
10 General › Politics › Government

CBSNews

No sign of “Christmas gift” from North Korea yet, but deadline looms

1 General › Politics › Military Defense
2 Industries › Aerospace And Defense › Weapon
3 General › Politics › Diplomacy
4 General › Politics › International
5 Industries › Aerospace And Defense › Naval System
6 Industries › Aerospace And Defense › Aerospace Systems
7 Industries › Aerospace And Defense › Missiles And Rockets
8 Industries › Energy › Nuclear Power
9 Industries › Aerospace And Defense › Satellite
10 Industries › Transportation › Ship

Le Figaro

Trump is hoping for a “nice vase” instead of a North Korean missile for Christmas. (Trump espère un «beau vase» au lieu d’un missile nord-coréen pour Noël)

1 General › Politics › Diplomacy
2 Industries › Aerospace And Defense › Weapon
3 Industries › Aerospace And Defense › Aerospace Systems
4 Industries › Aerospace And Defense › Missiles And Rockets
5 General › Politics › International
6 General › Politics › Military Defense
7 People › Society › Opinion And Idea
8 Industries › Aerospace And Defense › Satellite
9 General › Law › International
10 Industries › Aerospace And Defense › Aircraft

Editorial Angles

Here’s a summary of the classifications for the 3 articles:

A few remarks:

  • USAToday and Le Figaro top classification is Diplomacy. CBSNews is Military Defense

  • The 2 US articles have the same top 4. (in a different order)

  • Le Figaro does not have Nuclear Power in its Top 10

  • All have Military Defense. Only USAToday has Civil Defense

  • All have Aerospace and Defense > Weapon in their top 3

  • Only Le Figaro has Society > Opinion and Idea and Law > International in its top 10

  • For Industry > Aerospace and Defense, USAToday has 3, CBSNews has 4, Le Figaro has 5 out of their Top 10.

Here’s how TrustedOut saw the Aerospace and Defense Industry, back in October:

Corpus Intelligence for an Industry: Aerospace & Defense – October 2019

Next: Evolution over time.

How AI-Powered Classifications are sensitive to the time of publication: Meaning, Classifications evolve with the time as our “bag of words” are permanently updated and why it matters… Continue to part 2/2

Questions? Ask!

Lessons for growing publisher revenue by removing 3rd party tracking*

*Source: Brave https://brave.com/publisher-3rd-party-tracking/

The end of 3rd party cookies

On January 14th of this year, Google wrote: Building a more private web: A path towards making third party cookies obsolete

How personal data are “broadcast”.

We highly recommend this document, “Behavioural advertising and personal data”, from Dr Johnny Ryan, where we can read:

“…every time a person loads a page on a website that uses real-time bidding advertising, personal data about them are broadcast to tens – or hundreds – of companies. Here is a sample of the personal data broadcast.

●  What you are reading or watching
●  Your location (OpenRTB also includes full IP address)
●  Description of your device
●  Unique tracking ID or a “cookie match” to allow advertising technology companies to try to identify you the next time you are seen, so that a long-term profile can be built or consolidated with offline data about you
●  Your IP address (depending on the version of “RTB” system)
●  Data broker segment ID, if available. This could denote things like your income bracket, age and gender, habits, social media influence, ethnicity, sexual orientation, religion, political leaning, etc. (depending on the version of “RTB” system)”

“We used to read the newspaper, now the news reads us.”

This quote from the Global Editors Network. We strongly encourage you to read the article using the quote as a title and try the section “What happens when you read an article online”. Below is a screenshot for Spiegel.de

1 out of 5 happy for their data to be shared (UK, 2017)

In 2017, GFK was commissioned by IAB Europe (the AdTech industry’s own trade body) to survey 11,000 people across the EU about their attitudes to online media and advertising. GFK reported that only “20% would be happy for their data to be shared with third parties for advertising purposes”. [source]

Finding#1: Removing 3rd party tracking/AdTech and investing in Context increases revenue!

The first chart and the chart below are from the article from Brave.com, “lessons for growing publisher revenue by removing 3rd party tracking” both demoing the revenue increase is attributable to removing 3rd party tracking and adtech.

NPO and its sales house, Ster, invested in contextual targeting and testing, and produced vast sales increases even with sites that do not appear to dominate their categories.

The Covid-19 market shock shifted the market from video to display

Finding#2: “legitimate publishers of all size can increase revenue”. The New York Times example…

On their site, Open.nytimes.com, they wrote: “As of April 2019, we [The New York Times] removed all third-party data controllers from our homepage, section fronts and articles. … This reduced the amount of data we shared with third-party data controllers by over 90 percent. We are working on ways to improve this number…”

Finding#3. “Context is powerful.”

“NPO properties now provide no geotagging, no frequency capping, and no cross device measurement. Despite the absence of these features, extensive testing with advertisers has proven that the ads are effective, and advertisers are spending more with NPO than before.”

Next read:

TrustedOut partners with Xandr to bring new intelligence in targeting capabilities

Introducing the Brand Safety Report

The game is rigged: A former marketer shows you how Big Tech’s advertising practices harm us all

You have questions? Let us know!

contact@trustedout.com

or

TrustedOut’s Ad Campaign Curation: Simple, Safe and Permanently Updated.

Your Corpus is ready? So is your Curation for your DSP.

Short post today, as the process of connecting your curation made with a TrustedOut Corpus is ridiculously simple!

3 clicks…

Click #1: Get Media and Sources (as shown on the picture above)

Click #2: Connect

Click #3: Select Xandr (formerly AppNexus (Read Partnership announcement))

… and get a Deal ID ready for your DSP!

Insert this Deal ID in your DSP and let TrustedOut feed your campaign with Content you’ve selected.

Simple, Safe and Permanently Updated.

Simple as above reviewed the 3 click process.

Safe and Premium as demonstrated in this post: TrustedOut for Xandr: First results

Permanently Updated for 2 reasons:

1/ You can at all time go back to your Corpus, make changes and a click on “Save” will update the curation

2/ TrustedOut permanently updates its profiling, so if media are added or removed from your Corpus, they are also updated in your campaign.

Want to run a test?

contact@trustedout.com

or

 

Contextual relevance is key to customers. Profiling is key to Contextual relevance.

Credits IAS https://insider.integralads.com/contextual-advertising-research

Ads must be relevant to content.

In a report we encourage to read, IAS explores the power of context on consumer perception, we can read:

“Contextual relevance is preferred across all verticals

When shown articles representing different verticals, consumers were consistent: they always preferred contextual relevance. Across the board, consumers paired the advertisements they prefer with articles categorized in the same content vertical.”

The picture above shows the majority of consumers prefers to have ads relevant to the content where they are inserted.

It does make sense to avoid any opposition or distraction from the content.

Profiling makes content relevant.

TrustedOut’s Holistic Profiling works like this:

Which means, not only the content where the ads will be inserted is classified and gauged in expertise but the Perception and the Orientation of the Media of insertion are also gauged.

Ex: How it applies to Entertainment:

No more unmanageable, biased, irrelevant over time keywords

With TrustedOut, Classifications in our taxonomy define a Vertical.

For Entertainment, for example, brand classification “Entertainment & Leisure” comes to mind. But then, why not Information and Communication with its Motion pictures, Online Media, etc… and then why not Culture and Arts with its Arts, Comics, Dance… and then what about content about Eating and Drinking?…

Geo: USA, As of 2020/08/28

Why it matters?

No dependance of unmanageable, irrelevant over time lists of keywords.

TrustedOut qualifies every piece of content at the moment of use. Expressions and their weight are permanently updated.

An amazing opportunity for greater context relevancy.

In our example above, adjust ad messages to the type of Entertainment. Greater context relevancy, greater approval from the customer!

Relevant… and safe!

Now that context is relevant, but…
… is the publisher of this content spotted as Fake News, Junk Science, Conspiracy Theory, Revisionism or Hate News?
… is the publisher politically oriented? Religiously oriented? Humorous/Satirical?

Say you are looking for Entertainment in the largest sense as shown above but you don’t want publishers spotted with toxic content, not far right or far left and not humorous/satirical. No filter on Religion.

Your ad campaign will run within this corpus:

Related read:

TrustedOut partners with Xandr to bring new intelligence in targeting capabilities

TrustedOut for Xandr: First results

Questions? Let us know!

contact@trustedout.com

or

 

Corpus Reports: Brand Safety Surveillance and Analytics Perimeters Watch.

New: Corpus Reports.
Easy to read, Easy to share.

We are delighted to introduce within our Customer User Interface, a new feature coming from the feedback we’ve collected:

“How to get a daily report on the Corpus I’m using for an ad campaign or analytics on a product launch?”, “I want to share this report with my management, so no learning curve, must be straight forward”

Well, here it is. In the Customer UI, there is a new “Report” button. At any time you can get a PDF of what your Corpus is made of and share it amongst your team or/and client for review or approval.

Brand Safety Surveillance.

Let’s take an example: You are running an ad campaign to get traffic to a page. TrustedOut analyzes the destination page and build the following Corpus for this campaign:

  • Content must be French for France
  • Media must have covered “Society” AND talk about “Digital Life” over the past quarter to get stable classifications

Corpus looks like this:

A click on the “Report” button will give you this PDF:

Click on the button to get the report

Fine tuning your Corpus to get the desired Report

At all time you can tweak your Corpus to correct things you don’t like in the Report.

For example, page 29 shows:

… and you don’t want:

  • Toxic content
  • You can tolerate Politics but don’t want Far Right, nor Far Left
  • You’re ok with Religions and Humorous/Satirical

Then, change your Corpus definition to:

Next: Connect your Corpus to your DSP. (Spoiler alert: Blog post coming soon :))

Analytics Perimeters Watch

Controlling and sharing the Corpus you use for your analytics is critical.

After all, trusting decisions you are going to make impose to share the content you use to make your analytics and thus the decisions from those.

(reminder: the name TrustedOut comes from “If it’s not Trusted In, it cannot be Trusted Out”)

To pursue with our example above, the Report shows on page 9 the trends of your Corpus over time:

Now, say you don’t want to use, for any reason, media talking about Preschool and Primary Schools:

Simply change your Corpus definition with the addition line:

And now, Trends look like this:

Get management and clients involved by sharing Corpuses!

Questions? Let us know:

contact@trustedout.com

or

 

 

3 curation methods to ensure your brand is safe and visible.

3 curation methods for totally safe and broad reach whitelists.

1/ Brand defined context. Got Targets, Get Context.

This first method is straightforward. Define your primary target and get the corresponding media/sources list for your campaign.

Example: You are looking for Music media.

Want to broaden the scope to classifications surrounding Music? Like Shows? And Motion Pictures?

Let’s add:

  • Industries › Information And Communication › Motion Picture And Sound Recording
  • People › Entertainment And Leisure › Show And Performance

You now get 118 Media covering one or another (or 665 if you go lower in Editorial focus)

Instead you want to go more specific? Like Media covering Music AND Shows OR Motion Pictures?

You now get 48 Media covering one AND the two another (or 399 if you go lower in Editorial focus.

2/ Destination profiling for related context. Let your Landing Page make your whitelist.

Looking for a frictionless, no learning curve experience? This method is for you.

TrustedOut will profile your landing page and from it, create the corresponding corpus. Simple as that!

3/ Connex context to primary context. Expand your reach with editorial related Media/Sources.

This method was explained in this post: Discover new media/sources with Related Profiles.

Principle is simple:

1/ Get the primary target. In our example, Music

2/ Find the Top 3 growing classifications over a period of time, say the Past Week

3/ Create a Corpus with those 3 classifications over the same period of time. Voila.

Now you get 2.5x more Media.

In our post, you get 85 new media + the initial 56 = 141 Media. +2.5x more Media

Ready? Get your Deal ID to Connect your Corpus to your Campaign. Voila!

As we wrote here: TrustedOut’s Ad Campaign Curation: Simple, Safe and Permanently Updated.

TrustedOut gives you the deal ID for your Corpus. Once entered your Corpus,  updates will automatically be sync’ed with your DSP.

Questions? Shoot!

contactus@trustedout.com

or

Keywords vs classifications blacklists for your ad campaigns.

How to avoid some context?
Keywords vs Classifications bans.

Both methods are compatible and serve different purposes.

Keywords bans avoid specific words presence at the page level.

Pros: Very targeted and at the page level. Avoid a competitor brand, a named reference like a city or someone…

Cons: Got to be very specific on those keywords as they do not handle nicknames or synonyms, just to name those two. This generates very long lists to ensure safety and those long lists may be biased, outdated and prevent you from context you might indeed desire.

Classifications bans avoid specific associations to your brand at the Media/Source level.

Pros: No need to worry about new terms as bag of words for a classification are permanently updated. Words/expressions get in and out. Automatically. This is critical if you don’t want some risky associations to your brand.

Cons: This will not work to avoid specific words or brands where keywords ban is a better option.

An example? Want to be associated with Wellness, but not with Disease?

Your brand wants to be exposed in Wellness publication but not those associated with Diseases (like Covid).

For demonstration purposes here, we’ll focus on a small list of publications to easily identify differences.

Corpus will be made of:

  • Media in America and in English
  • Media Taxonomy is Wellness > Specialized > over the past week

This gives us 11 Media.

Now, you want to avoid Media with those 11 that do Cover Diseases. Your corpus becomes:

10 Media left. One is gone which you can look at by changing IS NOT to IS in the Disease condition and find:

Let’s double check with TrustedOut’s profile over the past week:

Want to double check with Google?

Search for site:issaquahreporter.com and then site:issaquahreporter.com covid to see 5 out of the 7 links over the past 7 days are mentioning the word “Covid” (11/7-14/20).

Want to give it a shot? Contact us!

contactus@trustedout.com

or

Introducing URL Curation. Feeding BI and AdTech with context they need.

Click on the chart to zoom in

Lowering granularity in content intelligence.

We demoed you don’t have to choose between Brand Safety and Performance in our recent post: TrustedOut for Xandr: Branding AND Business Performance.

From Sources Whitelisting for BI and Domains Whitelisting for AdTech, we’re now offering curation at the URL level.

Delivering URLs your analytics and ad campaigns need. Now.

This lowest level of granularity is now available to Analytics tools for the BI sector and your DSP for the AdTech sector.

Process remains the same, as simple as possible: As shown in the chart above:

  1. Create your Corpus with your definition of content you trust: taxonomy with level of expertise, content from publishers perceived toxic and political, religious or satirical orientations.
  2. Get corresponding URLs delivered to your BI system or your DSP in a timely manner.
  3. Voila.

AdTech test drive: A/B testing with vs without URL Whitelisting.

You now know we love A/B testing 🙂

So did a new one to compare the very same campaign, very same creatives, very same budget. One campaign with TrustedOut URL Curation and one withtout TrustedOut.

Here’s what we noticed:

Budget: Much better Click-Thru-Rate.

Measured CTR is the ratio clicks vs measured impressions. CTR with TrustedOut is significantly better.

Branding: Much better Viewability.

Viewability Rate is the percentage of ads which are actually seen by a user.

Prospects: High profile from huge percentage browsing on mobile.

  • Upper Socio-professionals. (Harris)
  • Greater attention span. (PEW)

Want to give it a shot? Contact us!

contactus@trustedout.com

or

Watch the content your BI is based on and where your brand is exposed in your ad campaigns.

Is the content you base your decision on the one you trust?
Is your brand appearing within sites you trust safe?

To make sure you answer yes to those questions (and more like where is focused your PR effort, are you tracking the right market…), TrustedOut is proud to offer, a posteriori profiling report.

It’s never too late. To know and fix.

As shown in the graphic above:

1/ Get the list of sources/media from your analytics tools and/or your DSP

Simply export the list of sources and/or media used for your analytics or get the list of domains from your DSP or ask your ad agency. This list can be in CSV, XLS or else in text format.

2/ Simply use the new “is in file” condition

New in TrustedOut is the “Files” menu. Upload your CSV or XLS made of your sources/medias as seen in previous step

Add the new condition for “Website”: “is in file” and select your CSV/XLS

3/ Click Report. Voila.

A real example?

Of course, we will keep this example anonymous. The privacy of people who trust us is top priority for us.

From an ad campaign, not using TrustedOut for its curation, we got the CSV file of all domains where a bid was won and thus, the client’s brand was exposed.

The Corpus producing the Report is dead simple but you can use this new filter line to dig into taxonomy, Spotted as, Countries, Languages etc…

Let’s go super simple. 1 line. That is it.:

Again, to keep this example all private, we won’t show the real and full report but here are few pages

Is your brand exposed in editorial focuses you want?

TrustedOut Report shows the editorial evolution of the domains where your brand was exposed.

Of course, the same applies to a list of sources used for your strategic decisions out of your analytics tools.

Is your brand exposed in sites spotted as toxic, political, religious or satirical?

Do you truly knew your brand was exposed there?

In our example, the client didn’t know and didn’t want to be seen on sites perceived as Far left or Far Right.

Well, just to mention one, its brand was on Breitbart. And more…

In the country and language you picked?

Client aimed for France and French.
TrustedOut found only 58.9% were in French and less than half in France.

The Solution: Use TrustedOut to ensure your curation is made of the content you trust.

In the example above, we offered a different whitelist made of media in France and in French, the desired taxonomy, both in expertise level, over a chosen period of time and blocked all toxic and extreme political orientation. We could do more.

Straight forward:

We encourage you to read this post: TrustedOut for Xandr: First results

Want to test drive?

Contact us to see if you qualify for a free report (large ad agency, enterprise, in the US or in France).

or email us at contactus@trustedout.com