Local Method For Wechat Friendship Flow Advertisement Recommendation Algorithms

With the rapid development of information technology, a variety of social network platforms continue to emerge, and social media has becomed an important channel for modern corporate marketing. Using community detection algorithms for large and complex networks can effectively identify the target community in the social network, this build a bridge for the communication between corporate and consumers. Wechat is the most popular social media in China, a growing number of company release ads through Wachat friendship ads systems. Since it’s a new advertising system, consumers generally feel that the receipt of product ads inconsistent with their demand. Based on this, the paper attempts to find a new user selection algorithm to improve the accuracy of the ads system.First, the paper introduces several community detection algorithms with great significance, though comparing their strengths and weaknesses, we discover that global detection methods need to master the information of the whole network, have high computational complexity and time-consuming, local algorithm is the opposite. Based on this, the weaknesses of Wechat flow ads recommendation algorithm has been analysised, combining the characteristics of the Wechat network, we propose a local community detection method for Wechat ads systems. The algorithm begins with the target Wechat Official Accounts, taking all the neighbours of the target Wechat Official Accounts as the first-tier nodes and added them to the community. Then search for the neighbours of the first-tier nodes as the second-tier nodes and added them to the community. Using modularity M as optimization index and select the second-tier nodes. Finally, the target community is consisted of the target Wechat Official Accounts, the first-tier nodes and the selected second-tier nodes. The paper improves L-shell algorithm on three aspects:First, on initial node selection, taking target Wechat official account as initial node; secondly, on stop standard setting, stopping the algorithm after the second-tier nodes have been selected; finally, on nodes optimization, taking objective indicator M as nodes selection standard.Studies using Igraph for analysis and visualization the network, our algorithm is tested on classic network, such as Zachary’s Karate Club, Renren network, Dolphins’Social Network and Books about US Politics. Then use community quality as advertisement effect and take modularity Q that is proposed by Newman and running time of the algorithm as evaluation index, comparing our algorithm with eight typical community detection algorithms. The results show that the algorithm can correctly identify the belonging of the nodes, the target community has a good structure, and use modularity M as vertex remove standard can effectively identify the boundary nodes. However, the running time of our algorithm is slower than the typical method in most of the cases. What’s more, our findings support for Zhang’s conclusions:although random networks do not have community structure in theory, a large degree of segmentation may still occur; modularity proposed by Newman is not suitable for measuring networks which have significant difference. Finally, the research work is summarized, some suggestions for company and user are put forward, and perspectives on the future are presented.