Research Of Multi-objective QPSO Algorithm And Its Applications In Text Clustering

Clustering analysis is an important technique of data mining. According to the similarity between data, the data set is divided into different clusters. It is widely used because of clustering flexibility. However, the traditional clustering analysis often require pre-determined clustering number, such as k-means algorithm which is one of the most classical clustering algorithms. So the algorithm has limitations in practical problems. Therefore, it is very meaningful in solving practical problems by the clustering algorithm which is dynamically determining the number of clusters. With the development of the network, the growth trend of data has being explosive. In the face of the complex data set, the traditional clustering analysis is easy to converge to the local optimum. With the rapid development of information technology, excellent algorithms are constantly emerging. In recent years, swarm intelligence and artificial intelligence algorithm is well developed, the swarm intelligence algorithm in solving the problem of mass data, in the face of global problems, has its unique advantages. In practical problems, making full use of their advantages has become possible.Quantum particle swarm optimization algorithm is a global searching algorithm with stochastic adaptive characteristics, good global convergence, strong robustness, potential ability of parallel processing and faster convergence speed.In consideration of the defects of traditional clustering algorithms and the characteristics of the quantum particle swarm optimization (QPSO) algorithm, this paper presents a multi-objective QPSO clustering algorithm which is combining quantum particle swarm and k-means by improving the particle coding way, dynamically changing clustering number M during the iterative process. Finally, the algorithm is applied to the field of text clustering. The main work includes:1. The multi-objective optimization algorithm in the paper is used as a global optimization strategy. The algorithm can find the optimal clustering number.2. The number of clusters parameterization will reduce the influence of subjective factors, make the results more objective with the facts.3. Improved quantum particle encoding, by the threshold matrix controlling the actual numbers of the cluster.4. Introduced a new way of clustering evaluation. Strengths and weaknesses of the division of cluster number is evaluated by using the DB function. In view of the features of text set, F measure evaluation is increased.5. K-means algorithm with fast convergence speed improves articles produced by the quantum particle swarm optimization algorithm in every new generation and converge the algorithm quickly.The realization algorithm in the paper is tested by using Iris, Wine and Glass data set of UCI, compared with PSO and QPSO clustering algorithm. The experimental results show the effectiveness of the algorithm. Finally the algorithm is applied to text clustering. The experiment results show that this algorithm can automatically set text divided into better grouping, and relatively close to artificial grouping result. The conclusions show that the algorithm has high practicability and strong intelligence.