Genetic Algorithms for SEO?

Introduction

While looking at a cool program that uses a genetic algorithm to design a car that drives over bumpy terrain I started to think about genetic algorithms and whether it would be possible to create an SEO genetic algorithm.

How Genetic Algorithms Work

For those of you who are not familiar with genetic algorithms they are process by which  a complex task can be found by using a process of selection over many generations. Initially a set of random numbers is chosen to form a gene. Each of the numbers in the gene represents an aspect of  a possible solution.

To take the example of the car program, the problem is to create a design for a car which is very good at going over a particular rough terrain. The gene might specify values for the radius of the front wheel, radius of the back wheel,  distances between parts of the frame, mass of counter-weights and so on. All of these aspects of the design can be specified by a number.

To get things started, the program creates 20 or so different designs and tests them all. In the first generation they are all rubbish. The car might not move at all and crashes before it reaches the first bump. But, because of the variation in the numbers, they will perform differently and their performance can be measured against a measure of success, known as the fitness function, such as the distance travelled. While all the designs were rubbish in the first generation, it is still possible to say which were the least poor in terms according to our measure of success.

These winning designs are combined in a process called cross-over, in which the genes of the two best design are combined, taking the gene of one-design up to a point and the rest of the design from the other gene. This is repeated 20 times to create another generation of car designs which also go on to be tested. The winning designs of this generation are selected and form the basis of the next generation of designs.

Occasionally, to add a bit of spice to the mix, a number in one of the  genes might be altered to a random value. This creates a possibility of some interesting new behaviour that may turn out to improve the design.

If this process is allowed to continue for many generations,  the design is honed and the car becomes better at the task and will  take longer before the car fails. It may not find the best solution and may get stuck in a genetic rut (no pun intended).

SEO Genetic Algorithm?

Might it be possible to design an SEO robot that finds new techniques?  Probably not but it is a fun exercise to think about how it might work.

It is fairly easy to define the measure of success or  fitness function, namely that the site which the robot is working on should rank as highly as possible.  A genome of SEO factors such as headings, content placement, titles length, meta length, could be thought up and numbers given to each.

The problem with genetic algorithms is that they take many thousands of cycles to obtain reasonable results. In SEO, a cycle would be  the time taken from applying an idea to seeing the result. Which might be weeks, so a naive genetic algorithm would be non-starter.

However, a robot that was semi-expert in the sense that the gene contained ideas from SEO theory, such as the title tag is important. Would already have a head start.

The main opposition to a genetic algorithm performing SEO would be that each sibling in a generation would have to operate on the same website. After the first had made changes it would be impossible to go back to the state, before the previous algorithm had made changes.