Evolutionary Clustering With Application Based On Evolutionary Multi-objective Optimization

Recently, some features of the data in many applications evolve over time. At different time steps, analysis and mining of these data become a new issue. For dealing with this issue, the framework of the temporal smoothness in evolutionary clustering was proposed and has caused a wide attention in academia and industrial community. The framework of the temporal smoothness in evolutionary clustering has the follow features:(1) the clusterings reflect the current data characteristics as much as possible;(2) the clusterings are not deviating too much from the most recent historical clusterings. These two festures in evolutionary clustering are conflicting that one increment must lead to the other degradation.Evolutionary muti-objective optimization algorithem (EMOA) is a new method developed recently, which employs eolutionary computation to solve multi-objective optimization problem. Because of its high efficiency and practicability, it is given more and more attention by academia. The algorithm is proposed by employing EMOA to optimize the two conflicting functions in evolutionary clustering.In chapter three, EMOA is introduced into evolutionary clustering. A novel evolutionary clustering, evolutionary muti-objective optimization evolutionary clustering based on MOEA/D, is proposed. In this algorithm, we employ the MOEA/D optimize the two conflicting functions in evolutionary k-means and avoid setting the weight parameter in advance. In addition, due to the introduction of evolutionary computing, the perfotmance of algorithm is improved and can find the better solution than the state-of-the-art algorithms. Synthetic datasets and UCI datasets are addressed to illustrate that our algorithm has the better performance.In chapter four, the framework of temporal smoothness based MOEA/D is adopted in community detection. A novel method to detect community, multi-objective evolutionary algorithm community detection in dynamic networks based on MOEA/D, is proposed. In this algorithm, firstly, appropriate criterion functions are adopted as the community quality function and history cost function, and we employ the MOEA/D to optimize these two functions simultaneously and get a set of optimum solutions to avoid setting the weight parameter in advance. Secondly, we adopt the locus-based adjacency representation with no need setting the number of clusters in advance and adopt the appropriate crossover oprator and mutation operator to improve the search capability. Lastly, experiments on computer-generated networks demonstrate that the proposed algorithm can not only find community structure and capture community evolution more accurately, but also be more stable than compared algorithm.In chapter five, we propose the decomposition-based multi-objective evolutionary algorithm with local search for community detection in dynamic networks, which is an improved algorithm with proposed in chapter four by incorporating a local search strategy. Experiments on computer-generated and real-world networks demonstrate that the proposed algorithm can not only find community structure and capture community evolution more accurately, but also be more stable than compared algorithm.