Centralized and MapReduce Algorithms for Dynamic Group Formation on Hadoop

• Main Authors: An-Te Huang, 黃安德
• Other Authors: Shiow-Yang Wu
• Format: Others
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/41031715184313343937

碩士 === 國立東華大學 === 資訊工程學系 === 101 === Social network services become mature these years. The main purpose is to build an online community and provide communication service for people who have the same preference, job or personality. There is a request that people who has the same proximity will be formed into a group temporarily. We call the kind of request as Dynamic Group Formation Problem. This goal of the research studies how to solve dynamic group formation problem, like groupon, car-sharing, and temporarily form a tour group. We will define these problems to a graph theory problem. We propose two approaches to solve dynamic group formation problem, centralized algorithm and distributed MapReduce algorithm. Centralized approach decides how to assign a user into a group which depends on the timing and the locations of groups and the preference of users. With the characteristic of MapReduce, distributed approach assigns users into groups according to their preference simultaneously. We expect to feature the advantage of distributed system in the situation of a large sum of groups and users. In this research, we implemented MapReduce algorithm by using Hadoop and HBase and compared to centralized algorithm in different amount of groups and users. For the largest CPU execution time in our experiment, distributed approach has 75% execution time less than centralized approach. Although the group complete rate and group satisfied rate for distributed approach are not as well as centralize approach, both rates can be 100% when the data become large.