Genetic-Algorithm Based Mechanism Optimization In Innovation Contest

In recent years, there has been rapidly increasing activities conducted on integrating internal and external input into the development of new products. Along with the increase of competitiveness the complexity of product design systems is correspondingly increased. Thus the traditional internal R&D innovation can no longer satisfy the intensive competition in the market place. To cope with the increasingly competitiveness, the in-house R&D strategy shifts to an open R&D strategy by which the innovation processes that are carried out by the external world. Additionally, the emergence of the internet has facilitated the development of open innovation through Computer Aided Innovation (CAI). Under this circumstance, the fact that activities of innovation process are conducted through internet instead of in the physical world is considered to be more efficient. Therefore, IT-based innovation contests are the very approach to solve the challenge. This IT-based innovation contests can further benefit various areas, including academic, business and so on.In an innovation contest, an organization (the seeker) aims to solve an innovation-related problem with external help which refers to a number of individuals (the solvers). Seeker sets up reward system to motivate solvers with the purpose of getting the most satisfied solution from a diversified solution pool. In turn, the solver who generates the most valuable solution gets the reward by devoting to the game. In this paper, the focus is on how to make an innovation contest work more efficiently (maximizing seeker’s profit). Facing a large population of contestants, solvers have the concern that their effort might not be financially rewarded. Thus they tend to be less productive which leads the innovation contest to be less effective. Although scholars mentioned that in order to eliminate this underinvestment affect solvers should be reduced to two, a latest literature argues that "the benefits of diversity can outweigh or at least mitigate the negative effect of underinvestment" insisted by researchers. Furthermore, they also mention that changing award system, from existing fix-price award system to performance-contingent award system, can reduce the inefficiency of innovation contests as well. Besides changing award system, some scholars have discovered that increasing innovation contests round can significantly motivate solvers. Only qualified solvers can participate in the following rounds. In the later rounds, each solver makes more effort in the contest, because he/she has higher probability to win. However, new challenge has risen that is running more rounds massively shrinks seeker’s profit. This thesis aims to solve this new challenge by optimizing mechanism of innovation contests and then improve seeker’s profit.As far as discussed here, this paper proposes an approach, which is inspired by survival of the fittest concept of Darwin, to solve this problem. Based on genetic algorithm theory, a computational model is constructed to simulate processes of innovation contests which are learning, variation, and selection. This paper interprets learning and variation processes as random crossover and recombination. For selection process, fitness function is constructed. Based on fitness value of each solution, probability of being selected is assigned and then selection process is performed. This article uncovers a possible affection of information sharing and learning behavior on the discovery of the first viable solution in innovation contests. By conducting simulation, this paper demonstrates three main contributive findings: First, learning behavior of solvers in innovation contests can accelerate the discovery the first viable solution. This finding is proved by conducting a computational simulation and further confirmed by comparing with a stochastic model; second, once the viable solution comes out, it spreads rapidly to the rest of solvers. Then, the entire solution level increases correspondingly; third, multi-reward strategy works even better on accelerating the discovery of the viable solution. Under this condition, learning behavior no longer guides the environmental selection. Later on, sensitivity analysis is carried out which figures out the influences of main elements on the innovation contest. Later, this paper constructs a numerical example which elaborates the application of the results in designing real-world innovation contest. The last part of this paper summarizes all the results and applications, and figures out the limitation of our research and provides the outlooks for further research.