I’ve recently been using Genetic algorithms in Actionscript 3 and have since been contemplating the benefits of using them to enhance webpages by implementing some features server side.
Genetic algorithms are a search technique that utilise the speed of computing to ‘brutishly’ locate or find a match/optimum solution.
The one idea that really resonated with me whilst working on this topic was the idea of a self optimising website that would continually modify itself within the boundaries I define.
To clarify, the optimisation would be in the form of design and layout. There are a number of design/UX testing methodologies (e.g. Test multiple designs on site users by randomly sending them to site design ‘A’ or site design ‘B’) to monitor which design best fulfils your objectives by checking over the resulting conversions/visit duration etc.
The algorithm I’m working on uses the principles of genetic algorithms / evolutionary programming combined with the aforementioned testing methodologies (A/B Testing / ‘Bucket Testing’) to accomplish and process the same result.
So, from the top;
The site when built has a number of defined regions very similar to what a basic page layout wireframe looks like:

Example layout containing five regions.
Using these defined regions instead of following standard web principles and statically placing content we would define rules that allow the site’s advert for example to appear in region 1, 3, 4 or 5 and that the searchbox remain above the fold (regions 3, 4 or 5).
We would also define some styles for the site, maybe a yellow oriented colour palette and a black oriented palette.
Finally we must define some targets for the algorithm e.g. ‘User visit duration’, ‘Advert clicks’, ‘Average pages per session’.
Using the resulting statistics, each time the algorithm is called to render the page it can take into account aggregated results from previous generations and decide which combination of placement and styles gains the best results.
I expect to find some patterns occur over time e.g. User visit duration increasing when the dark colour palette is used at night time and the lighter palette used in the daytime (grouping of visitors from similar timezones would better this).
I also expect to see the ads move down the page to region 5 if they were just standard Adsense style ads and maybe up the page (region 1 or 3) if they contained a fantastic offer on a relevant and well liked product etc.
The algorithm I’m working on will be tested on a separate site over the coming months after which I will write a conclusion post with the results and code.
Great post! Just wanted to let you know you have a new subscriber- me!
Man, I have been reading quite a bit on genetic algos, and neural networks lately, trying to find a useful way to set something up in AS3 to internalize the process a bit more.
The mona lisa image generator was an interesting example of the concept. It would be cool to see more of this type of experimentation. looking forward to more
Hi, very nice post. I have been wonder’n bout this issue,so thanks for posting
Nice work TG
Your site is worth beeing in the top cause it contains really amazing information.
Hello, can you please post some more information on this topic? I would like to read more.
Thanks guys. Your support really makes writing these worthwhile! More to come on genetic algorithms soon. I have a few ‘in progress’ articles to be released first then i’ll re-visit this with some code examples and few more interesting tidbits!
Very interesting! Can’t wait to see how this works in reality – and what kind of site it generates. Top work!