The Web Science Trust

The Life and Mating Habits of the Brown Bordered DIV: Emergent Semantic Elements from Genetic Design on the Web

Hockenberry, Matthew and Arroyo, Ernesto (2009) The Life and Mating Habits of the Brown Bordered DIV: Emergent Semantic Elements from Genetic Design on the Web. In: Proceedings of the WebSci'09: Society On-Line, 18-20 March 2009, Athens, Greece. (In Press)

PDF (Poster Description) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
PDF (Poster Artwork) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader


Genetic design techniques for the web are presented as a use of genetic algorithms and evolutionary approaches to the problem of design and development on the web. We establish a scaffolding to model a number of genetic algorithm adjustments, simulate the results and draw implications about the influence of linked data and semantic clustering in distributed systems. The goal of this paper is to present genetic design as a tool that is uniquely suited for the design and evaluation of web systems and establish an understanding of the possibilities and problems with employing these techniques by studying the possible and probable outcomes associated with their use. Genetic design employs the principles of genetic algorithms, the computing technique that incorporates the process of evolutionary growth and selection in order to approach approximate answers to search and operation problems, as a solution to the problem of design - approximating the answer of what is the most ideal or usable or useful design solution. By observing this process, we can make not of emergent properties of the system that have relevance for our understanding of the semantic grouping of elements and the implications of linked data. We begin by modeling the base genetic operators for genetic design on the web. Within the context of genetic design initialization is the design seed we provide originally, mutation is the process of altering the design within the constraints of our design language, selection is the human facing evaluation work, and crossover is the reproduction or propagation of successful designs. Selection is one of the issues that may be most relevant from our perspective. Selection is the fitness evaluation of a design. We have the advantage of having developed a large number of measures and metrics with which we can test the fitness of a design. The important part of these metrics is that most of them require human experiential accounts as a necessary part of computing them. Traditional interactive genetic algorithm (IGA) work has necessarily begun to converge to either a very small number of operations (producing small populations of solutions) or to reduce the number of selection operations made by humans (producing discrete human selection). The justification is simple - human fatigue. As the designers of distributed software systems, either on the web or web-based software deployment, we have an advantage. We can perform a continuous IGA by incorporating it into the software that is being distributed. This increases the size of the possible population and by supplementing this with implicit usability measures, can allow the process of selection to occur at the same time as use. We use this as the base model for our simulations. In order to test different options and operators for genetic design we developed a system that employs the principles of genetic design within the development of web sites and web applications. While we believe the expansion of this to any piece of distributed software is reasonable (with some caveats) the task of designing for the web is very broadly applicable. We model the evolution of a complete web design by beginning with the base unit of any such design, the DIV. The DIV is a generic block element on the page that can be transformed (with minor difficulty) into any other renderable web element with a finite set of styling rules and some additional rules that cover common elements outside of the current framework of Cascading Style Sheets (CSS) (transforming a DIV into a link tag is not possible without the use of some basic Javascript, for example). The system operates with three necessary components: a core listing of all possible design operators we could employ - functioning as our raw set of organism strings or ‘genome’, a client-side javascript listener that measures user interaction and performs in session adjustments, and a backend processor that stores and computes possible mutations for a given Genetic DIV. The raw set of organism strings (or rather possible organism strings) is drawn from the limitations of CSS as a design language. It is a relatively large and robust language that offers a finite set of determinations as possible design rules that we can apply to our DIVs. As a user interacts with the system he or she is providing information to the listener that contributes to the fitness test for that DIVs he or she is interacting with. The measure of fitness we use is derived from mousetracking measures - meaning that the base level of fitness for a DIV is simple, does a human pay attention to it. We believe that this measure is a good base fitness requirement because it is: implicit - the user isn’t necessarily aware that he or she is evaluating the DIV, general - it can be applied to a large number of design problems, simple - it isn’t very hard for us to compute, and nondigital - it gives us varying values for users depending on how long they are on or even near elements (with decreasing determinations of fitness as the user moves away). The fitness rate we used was relatively harsh, in order to ensure that all DIVs eventually die and are replaced. The fitness test we perform at the base of the simulations, testing the attention through mousetracking, is a very general one. Although it requires human selection, it avoids the problems associated with traditional IGAs by conducting these evaluations implicitly as a function of the general use of the web design. This also means that, unlike the other examples of genetic design, we don't need to labor to craft precise fitness requirements. The advantage of not adding them is that we have the freedom to design for unpredictable tasks or tasks and use cases that are difficult to define requirements for. The disadvantage is that it may require time to optimize to these additional tasks. We can now observe the evolutionary process from a high level perspective. Browsers style elements by default, with an intentional emphasis on meaning - this collective evaluation produces an emergent styling that is related to perceived meaning. The emergent properties of the system have favored semantically related elements by rewarding styled elements that are meaninfully related. We propose a mechanism for the identification and propagation of semantically grouped elements that makes use of linked data as a propagation tool. Instead of discussing pre-existing tags or artificial concepts like microformats, we observe unique DIV 'species' in the wild.

Item Type:Conference or Workshop Item (Poster)
Uncontrolled Keywords:Evolutionary Computation, Emergent Identification, Web Design, Semantic Web
Subjects:Web Science Events > Web Science 2009
ID Code:197
Deposited By: W S T Administrator
Deposited On:24 Jan 2009 08:45
Last Modified:25 Oct 2011 16:40

Repository Staff Only: item control page

EPrints Logo
Web Science Repository is powered by EPrints 3 which is developed by the School of Electronics and Computer Science at the University of Southampton. More information and software credits.