Efficient Parallel Mining of User Movement Patterns

• Main Authors: Huei-Cin Hsiao, 蕭暉勤
• Other Authors: Yao-Te Wang
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/48463270692289623449

碩士 === 靜宜大學 === 資訊管理學系研究所 === 99 === With the rapid development of the mobile communication technology, mobile phones have become the indispensable electronic equipment. Increasing competitiveness effectively and expanding market share rapidly are very important topics in the telecommunications industry. Through the use of popular data mining techniques are to find frequent movement patterns from the tracks of mobile clients, the mining results can be used to predict the motion trends and behavior of the mobile users and hence to provide mobile users with better services and reduce the transaction costs significantly. In addition, according to the mining results, the telecommunications operators can evaluate where to add base stations in the GSM network to improve the quality of services. Parallel mining of frequent movement patterns can improve the efficiency of mining tasks significantly. In this research we conducted the mining experiments in a four-node cluster system. This research targeted the movement paths of mobile users, and proposed two algorithms for sequential pattern mining and parallel mining, namely UMPM (User Movement Pattern Mining) and PUMPM (Parallel User Movement Pattern Mining) respectively. UMPM consists of four mining stages: 1. Retrieve the sequences from the database; 2. Build the tree structure; 3. Traverse the tree structure to find frequent patterns; 4. Extract the maximal patterns. PUMPM consists of five stages: 1. Retrieve the sequences from the database; 2. Determine the prefixes of the sequences dispatched to the computing nodes; 3. All computing nodes build their own tree structures; 4. Each computing node traverses the tree structure to find frequent patterns; 5. Prune the sub-patterns. A series of experiments are conducted in this study to evaluate the performance of both UMPM and PUMPM algorithms. The experimental results show that the UMPM algorithm is highly efficient and the PUMPM algorithm improves the mining performance more significantly.