Eliminate AI and machine learning in search.

The world of AI and machine learning has many layers and can be quite complex to learn. Many terms are out there and can be quite confusing unless you have a basic understanding of the landscape. In this article, expert Eric Ang will introduce the basic concepts and try to eliminate them all for you. It is also the first in a four-part article series covering many interesting aspects of the AI ​​landscape.

The other three articles in this series will be:

  • Introduction to Natural Language Processing
  • GPT-3.: What is it and how to take advantage of it?
  • Current Google AI algorithms: Rankbrain, BERT, Mother, And Smith.

Basic background on AI

There are so many different terms that it can be difficult to sort out what they all mean. So let’s start with some definitions:

  • Artificial intelligence. – This refers to the intelligence that machines possess / manifest, as opposed to the natural intelligence that we see in humans and other animals.
  • Artificial General Intelligence (AGI) – It is a level of intelligence where machines are able to solve any task that man can do. It doesn’t exist yet, but many people are trying to make it.
  • Machine learning – This is a subset of AI that uses data and iteration testing to learn how to perform specific tasks.
  • Deep education – This is a subset of machine learning that takes advantage of highly complex neural networks to solve more complex machine learning problems.
  • Natural Language Processing (NLP) – This is especially the case with the focus on language and the focus on understanding.
  • Neural networks. – This is one of the most popular types of machine learning algorithms that try to model the way neurons interact in the brain.

These are all closely related and it is useful to see how they all fit together:

In short, artificial intelligence encompasses all of these concepts, deep learning is a subset of machine learning, and the processing of natural language uses a wide range of AI algorithms to better understand language.

An example of how the neural network works

There are many different types of machine learning algorithms. The most popular of these are neural network algorithms and I want to give you a little bit of context which I will cover next.

Consider the issue of determining an employee’s salary. For example, what do we pay someone with 10 years of experience? To answer this question we can collect some data on what others are being paid and their years of experience, and it might look like this:

With such statistics we can easily calculate what the salary of this particular employee should be by making a line graph:

For that particular person, it offers a salary of just over $ 90,000 per year. However, we can all quickly recognize that this is not really a sufficient theory because we also need to consider the nature of the job and the level of employee performance. An introduction to these two variables will lead us to the data chart:

This is a very difficult problem to solve but machine learning is relatively easy. Still, we’re not completely done with adding complexity to the factors that affect salaries, because where you are located also has a huge impact. For example, San Francisco Bay Area jobs in technology are significantly higher than similar jobs in many other parts of the country, largely because of the large difference in the cost of living.

An isolated example of a simple vector map of the United States (United States). Names of borders and states (territories) Gray Silhouettes. White outline

The basic approach that the neural network will use is to estimate the correct equation using variables (job, years of experience, performance level) and calculate the possible salary using this equation and see that How much it corresponds to our real world statistics. This process is how neural networks are formed and is called “gradient descent”. The simplest way to explain this is to call it “successive estimation”.

Real pay data is what a neural network uses as “training data” to determine when it has developed an algorithm that is relevant to real-world experience. Let’s move on to a simple example that starts with our actual dataset, just years of experience and salary data.

To put it simply, let’s assume that the neural network we will use for this assumes that 0 years of experience equals $ 45,000 in salary and that the basic form of the equation should be Should: Salary = Year of Service * X + $ 45,000. We need to work on the value of X to come up with the correct equation to use. As a first step, the neural network can estimate that the X costs ، 1,500. In practice, these algorithms make these initial estimates randomly, but it will for now. When we try the 1500 price, we get:

As we can see from the resultant statistics, the calculated values ​​are very low. Neural networks are designed to compare numbered values ​​with real values ​​and are provided as feedback which can then be used to make a second guess as to what the correct answer is. ۔ For our example, let’s set the next estimate of $ 3,000 as the correct value for X. This is what we get this time:

As we can see our results have improved, which is good! However, we still need to re-evaluate because we are not close enough to the right values. So, let’s try to estimate 6000 this time:

Interestingly, we now see that our error margin has increased slightly, but now we are much higher! Maybe we need to lower our equations a bit. Let’s try 4500:

Now we see that we are very close! We can continue to strive for additional values ​​so that we can see how we can improve results. This adds another key dimension to machine learning that we want to know how accurate our algorithm is and when we can prevent repetition. But here we are very close to the purpose of our example and hopefully you will understand how it all works.

Our example of machine learning is a very simple algorithm to build which we just need to get the equation in this form, however, if we were trying to calculate a real pay algorithm that takes into account all these factors. To affect user pay, we need:

  • Huge data set to use as our training data.
  • To create more complex algorithms.

You can see how machine learning models can become increasingly complex. Imagine the complications when we are dealing with something on the scale of natural language processing!

Other types of basic machine learning algorithms.

An example of machine learning shared above is what we call “supervised machine learning”. We call it supervised because we provided a training dataset that had target output values ​​and the algorithm was able to use it to generate an equation that would produce the same (or close to) output results. There is also a class of machine learning algorithms that do “unsupervised machine learning”.

With this class of algorithms, we still provide an input data set but do not provide examples of output data. Machine learning algorithms need to review data and find meaning within the data on their own. It sounds awful like human intelligence, but no, we’re not there right now. Let’s explain two examples of this type of machine learning in the world.

An example of unsupervised machine learning is Google News. Google has a system for discovering articles that get more traffic than new search queries that appear to be driven by new events. But how does it know that all articles are on the same topic? Although it is as traditional as regular search in Google News, it is done through algorithms that help them determine the similarities between pieces of content.

As illustrated in the example above, Google has successfully grouped a number of articles regarding the passage of the Infrastructure Bill on August 10, 2021. As you might expect, every article that focuses on self-narration of events and bills has a lot in common.

Another interesting class of machine learning is what we call “recommendation systems”. We see it in the real world on e-commerce sites like Amazon, or movie sites like Netflix. On Amazon, we can see “frequently purchased” at the bottom of the listing on a product page. On other sites, it may be labeled as “People who bought it also bought it.”

Movie sites like Netflix use similar systems to give you movie recommendations. These specific preferences can be based on your rated movies, or your movie selection date. One popular way to do this is to compare movies you’ve seen and rated very well with movies that other users have watched and rated.

For example, if you rated 4 action movies too high, and a different user (which we will call John) also rated action movies too high, the system may recommend you other movies that John has seen but you have not. This general approach is what is called “collaborative filtering” and is one of the many ways to create a referral system.

Note: Thanks. Chris Payne To review this article and provide guidance.


The views expressed in this article are those of the guest author and do not necessarily reflect those of Search Engine Land. The authors of the staff are listed. Here.


About the Author

Eric Eng is the General Manager of Percent Digital, a full service, award-winning digital agency. Prior to that, Eric Stone was the founder and CEO of Temple, also an award-winning digital marketing agency, acquired by Percent in July 2018. In Industry 2016 as the “Bible of SEO”, Ange was awarded Search Engine Land’s Landy Award for Search Marketer of the Year, and the US Search Award for Search Personality of the Year. He is a well-known author, researcher, teacher and a keynote speaker and panelist at major industry conferences.

Leave a Reply

Your email address will not be published.