Community Optimization

In recent years a number of web-technology supported communities of humans have been developed. Such a web community is able to let emerge a collective intelligence with a higher performance in solving problems than the single members of the community. Thus, collective intelligence systems are explicitly designed to take advantage of these increased capabilities. A well-known collective intelligence system is

Wikipedia

, the web encyclopedia. It uses a collaborative web community of authors, which improves and completes the content of articles. The quality of a certain number of these articles comes close to some degree to that of a famous printed encyclopedia. Based on such successes of collective intelligence systems, the question arises, whether such a collaborative web community could also be capable of function optimization.

This paper introduces an optimization algorithm called Community Optimization (CO), which optimizes a function by simulating a collaborative web community, which edits or improves an article-base, or, more general, a knowledge-base. The knowledge-base represents the problem to be solved and is realized as a real valued vector. The different vector components (decision variables) represent different topics contained in this knowledge-base. Thus, the dimension of the problem is the number of topics to be improved by the simulated community, whereby the dimension remains static. In order to realize this, CO implements a behavioral model of collaborative human communities derived from the human behavior that can be observed within certain web communities (e.g.,

Wikipedia

or

open source

communities). The introduced CO method is applied to eight well-known benchmark problems for lower as well as higher dimensions. CO turns out to be the best choice in 9 cases and the Fully Informed Particle Swarm Optimization (FIPS) as well as Differential Evolution (DE) approaches in 4 cases. Concerning the high dimensional problems, CO significantly outperformed FIPS as well as DE in 6 of 8 cases and seems to be a suitable approach for high dimensional problems.