evo-web
posted by evo-web @ 1:33 PM
13 comments
Monday, October 02, 2006
Evo-Web is a case study website and a research prototype for Master's thesis entitled, "Webpage Optimization Using Genetic Algorithm Driven CSS."
13 Comments:
Seed is a bit big?
(Posted by mgsloan on October 11, 2006 at 02:10:33)
I quite like Genetic Algorithms, implemented a few, read about lots.
This is why I'm currently number one on ranking pages (currently done 18).
However, judging by the design numbering, your 1st generation seed is huge. Especially for a comparative selection system.
From what I can see not many people have participated yet, however you're not going to get a million people to evaluate pages. If you've got 20,000 in the first generation then that's the kind of participation you'll need. Much more practical would be a few hundred individuals, allowing a generation to be processed with only a few times more evaluations.
On second thought I suppose you could be doing a non-generational system where each selection kills two and spawns two from your choice (perhaps with crossover with another chosen).
Ohwell, hope it works out.
Re:
Hi #1 supporter,
The design numbering on sample pages does not reflect the actual population size. We have rather small population in each generation. If you vote enough you will start to see the pattern in design samples.
I agree with you, since we don't have another year or two to wholly devoted to this project and not to mention the low traffic, we're playing it fairly conservatively starting with samller samples
and expecting to see more generations.
Thanks for your participation and input!
everytime the same design
(Posted by anonymous evaluator on October 11, 2006 at 07:40:08)
I can imagine it is not easy to apply GE to some not very defined thing as "design". However, what you do here are (as it seems) variations on the essentially same (standard-3-column-) design, with different colours, font properties and position of the rows. you won't produce a better design by evolutionary means, because you won't produce any different design.
Re:
This project is merely the starting point of applying GA to optimize CSS-genrated webpage. And yes, it's not easy. :(
As you know CSS can play magic with all the design components and the possibilities are endless. For the time and other constraints we have for this project, we came up with rather controllable "pre-designed" layout and color choices and limited their variations by certain ranges.
As a result, there are more of subtle changes in fonts and colors than the whole layout or general look of a page. I'm interest in seeing if certain set of the components and their relationship accross the page is more favorable to others and if this can be driven efficiently by evolutionary computation.
When we will see the results of the test?
(Posted by Porta on October 11, 2006 at 09:23:05)
Re:
Evo-Web is still in its early generation. We don't have a set date to release the final design. In fact, it all depends on how many participants we get and how many tests they go through each time.
As we get more votes you'll start to see "parent" and "mutated" links for some designs.
We are working on the "family tree" feature so possibly you can see later how some designs have evolved or died out, etc.
you did good job~
good luck on your study~!
Interesting concept. I use CSS in all of my sites and think this could prove to be an intriguing way to allow customization and keep the design fresh.
I'm not a web designer or programmer, just a user with very definite opinion of what I would like to see on a web page. I followed this as long as I could but didn't get the sense that it was evolving toward my preferences.??? I quit finally when I had a choice that was somewhat close in design (not color). There were at least two sets of choices where I equally disliked all my choices. Of course, "none of these" was not an option. Just a layperson reaction to the exercise. Good luck with the project.
Re:
Since the voting process is designed to be random (i.e., not every visitor is looking at the same set of design choices that you're looking at any given time), it's hard to steer the design toward certain direction though single person's voting. It's almost like trying to feed a dog with different foods and hoping that one day the dog will grow wings.
OK, that might be a stretch, but seriously it means that the evolution process is a global fitness adaption for all the pages and any single person's preference only contribute to the global fitness value for each design; designs you dislike may somehow attract the votes from other visitors.
It must be noted that the voting process is strictly comparing just the three designs presented, even if they are all ugly, to find the best one among them. This still helps the evolution process even the user may feel that they don't like any one of them. Some of the "good genes" in those ugly designs may later be combined with other good genes from other designs, and transforming an ugly larva into a beautiful butterfly.
Evolution process is not just by design, but by chance as well. What you were doing is more of an "imperative" drive, but not quite the same as "evolvable" touring of the genetic landscape of the optimal webpage designs.
Thanks for your extensive participation and comment!
My decisions ended up based on legibility of font sizes and colors--not on layout differences or other factors. Since I found two of each set of designs difficult to read, I simply voted for the legible design regardless of all other factors.
Even as the page evolved, it was still dreadful. Text was difficult to read at times, colors--ugh, and non-functional placement of features. Hope your research is helpful in evolving to useful webpages, not just pretty ones.
Re:
I agree with both of you (above) on that some type settings and colors still have much room to iterate and improve. Colors are especially tricky since we have less control over choosing “good” colors and ruling out “bad” ones –not to mention the endless combinations of all the colors and some, mutated.
Type, color and layout preferences can all be very personal and so does the preference in hierarchy of those components on a page. One might notice the colors first when others pay more attention to the legibility. It’s a challenge to nail down on a design that will accommodate and best fit all the user needs and preferences. However, if you’ve been a constant visitor you can definitely see the overall improvement of design samples over the last month.
At a minimum, this evolution process is a mechanism of continuous refinement, not so much about generation of new ideas. It’s more effective, like in real design process, to come up with many quick and dirty sketches than mingling over a few. Users often don’t know what they want (or don’t want) until they see it. Evolutionary method will help presenting these low fidelity prototypes to users in quicker and more effective way.
Nature isn’t clean or neat and sometimes it makes mistakes in the short run by rewarding the “ugly” organisms. So does the web design. Even the “best” designs don’t always make to the next generation. The process is merciless and risky in some ways. Nevertheless, in the long run, we hope this to project an efficient method in shifting balance over the long haul of web design process.
Suggestion: to get more reality out of this evolution process, shouldn't the text be different each time the visitor is reading a set of design pages? That way, since the text is meaningful to read on, they will not simply vote on the color or layout. Otherwise, just like one of the comments said, people tend to focus on just color but not the readability of the page font if they have participated in this test for a while.
Re:
Thank you for your suggestion.
Please visit evo-web and see the change!
Using spams as the variable page content is a great idea! To look at these spams is really amusing, as you can tell the spammers are almost trying to create novels, haha.
One more suggestion: shouldn't the spam intro paragraphs be moved down, below each spam excerpt? Right now for the upper part of each page it is still the same text each time; for my small screen I didn't get to see the spams until I scroll down.
Post a Comment
<< Home