Search model shall be capable of “self-calibration”. In other words, it shall be able to assess its algorithms and their relative share and compare the synthetic data with generally accessible search engine in order to figure out the most precise search mechanism that allows modeling of any environment.
However, computing the analysis of thousands of parameters in an attempt to find the best combination thereof is astronomically expensive and very complex.
Considering the above, is there a way to create a self-calibrating search model? It appears that the last resort that can help us is… birds. Yes, you haven’t misread it, we’re talking about the feathered type!
Optimization using the particle swarm optimization (PSO) chart
It’s quite common that the most enormous problems can have the most surprising solutions. For example, one shall take a look at particle swarm optimization, which is an artificial intelligence method first described in 1995 based on social-psychological behavioral pattern of a crowd. Originally the technique was modeled based on the concept of behavior of birds in a flock.
Actually, none of the existing rule-based algorithms can be used to find at least an approximate solution of the most complex numerical maximization or minimization problems. However, using such a relatively simple model as a flock of birds, you can immediately see the answer. We all heard horrifying forecasts that the artificial intelligence is going to take over the world some day. However, in this particular case it is our most precious ally.
Scientists have been working on design and implementation of many projects dedicated to the swarm intellect. The “Millibot” project, previously known as “Cyberscout”, launched in February 1998 is a software sued by the U.S. Navy. Cyberscout was inherently a swarm of tiny robots that could invade a building, occupying its entire space. Ability of these hi-tech tiny tots to communicate and share information allowed the “swarm” of robots to act as a unified whole, which turned the labor intensive task of building investigation into an amble walk along the hallway (most robots were only capable to travel a couple of meters).
Why does it work?
What’s really great about PSO is that the method makes no assumptions regarding the problem you’re trying to solve. It is actually something in between a rule-based algorithm that tries to work out a solution and neural networks of the artificial intelligence that aim at investigating the problematics. Therefore this algorithm is a compromise between research and exploitative behavior.
While not being research-oriented in its nature, this algorithm of an approach to optimization would certainly turn into what’s called the “local maximum” in statistics (solution that may seem optimal, while actually it isn’t).
You shall begin with a series of “swarms” or guesses. For example, in search model these might be various weighting factors of scoring algorithms. If, for instance, you have 7 various entry points, you shall begin with no less than 7 assumptions regarding such weights.
The Idea of PSO is that each of these assumptions shall be as far away from others as possible. There are several techniques you can use to ensure that your starting points are optimal without going into 7-dimensional calculations.
After that you will start developing your guesses. You will do so in a way of imitating the behavior of birds in a flock, when there is food nearby. One of the random guesses (flocks) will be closer than others, and each consecutive guess will be modified based on the shared data.
Visualization below is a clear demonstration of this process.
Luckily there are numerous possibilities to implement this method in various programming languages. And the best thing about the particle swarm optimization is that it can easily be applied in practice! The technique has minimum setup parameters (which is an attribute of a strong algorithm) and very short list of limitations.
Depending on your problem, the idea implementation may be found in the local minimum (not an optimal solution). You can easily fix it by introducing the neighborhood topology, which will quickly limit the feedback cycle by only the best adjacent assumptions.
Your main work will be to develop the “adaptability function”, or a ranking algorithm that you will use to determine the level of aboutness to the target correlation. In our case related to SEO, we will have to correlate data with some predefined object, such as search results from Google or any other search engine.
When you have a working scoring system, your PSO algorithm will try to maximize the results through trillions of potential combinations. Soring system can be as simple as performing the Pearson correlation between your search model and internet users’ actual search results. Or it may be as complicated as simultaneous activation of these correlations and assessing a score to each particular scenario.
Correlation in relation to the “Black Box”
Many SEO-optimizers nowadays try to perform a correlation in relation to the Google’s “Black Box”. Naturally, these conditions have some reasoning to them, however they’re mostly useless. And here is why.
First of all, correlation doesn’t always implicate the cause-and-effect relationship. Especially when entry points to your black box are located not too close to exit points. Consider the following example, where entry points are very close to relative exit points – the ice cream transportation business. People buy more ice cream when it’s hot outside. It’s easy to see that the entry point (air temperature) is closely related o the exit point (ice cream).
Unfortunately most SEO-optimizers don’t use static proximity between their optimizations (entries) and related search results (exits).
Moreover, their entries or optimizations are often placed before bypass components in the search engine. Actually, typical optimization shall include 4 levels: content bypass, indexing, scoring and, finally, real-time request. Trying to correlate this way can result in nothing but wasted expectations.
In reality Google provides significant noise ratio, just as the U.S. government creates a noise around its GPS system, so that civic people are unable to receive same precision data as military. This is called the real-time request level. And this stage is becoming a significant constraining factor for the SEO-correlation tactics.
As an example, one can take a look at a garden hose. On the scoring level of the search engine you can see the company outlook on what is going on. Water that comes out of a garden hose is organized and predictable, i.e. you can change the hose position and predict the resulting change in the water flow movement (search results).
In our case, the request level disperses this water (search results) into millions of drops (variations of search results) depending on a user. Most changing algorithms today appear on the request level in order to provide larger number of variations of search results to the same number of users. For example, take a look at the Google Hummingbird algorithm. Shifts on the request level allow search engines to generate more trading platforms for their PPC ads.
Requests level is the outlook of users on what’s happening rather than that of a company. Therefore, correlations created in such a way will rarely have cause-and-effect relationships.At that, this is subject to a condition that you only have one tool to find and simulate data. In reality SEO-optimizers use a set of entry data, which increases noises and decreases the probability of finding cause-and-effect relationships.
Search for cause-and-effect relationships in SEO
To achieve correlation when working with the search engine model, one shall tighten entries and exits as much as possible. Input or variable data shall be placed on the scoring level of the search engine model or above. How can one do it? We need to separate the search engine’s black box into key components and then build a search model from scratch.
Optimization of exits is even more complicated due to enormous noises that appear from the real-time requests level, which creates millions of variations for each user. We would have to at least create such entries for our search engine model that would be located before the normal layer with variations of requests. This will guarantee the stability of at least one of the compared sides.
After building the search engine model from scratch we’ll be able to display search results that come not from the requests level, but rather directly from the scoring level. This gives us more stable and reliable relationship between entries and exits that we’re trying to correlate. In such case these stable and illustrative relationships between exits and entries will allow correlation to reflect on the cause-and-effect relationship. By concentrating on one entry we can receive direct feedback with the results that we see. Later we can perform a classic SEO-analysis in order to determine the most beneficial optimization option for the existing search engine model.
One can’t help but be amazed, how something simple that happens in wild nature leads to scientific discoveries or technological breakthroughs. Having a search engine model that allows us to openly connect scoring exits with non-customized search results, we can connect correlation with the cause-and-effect relationship.
Add particle swarm optimization into the mix, and you can witness a technological breakthrough – a self-calibrating search model.