Double layered recommendation algorithm based on fast density clustering: Case study on Yelp social networks dataset

摘要

With the wider applications of mobile devices, network-based recommendation has attracted more attentions than ever. At present, most research of recommender system is on basis of either graph theory or algebraic methods. However almost all of these recommendation algorithms are aiming at static history data, which cannot meet the demand of precise online recommendation for dynamic real-time emerging data. Besides, lots of traditional recommendation algorithms are suffered from high time complexity and parameter tuning difficulty. Here, we propose a personalized recommendation algorithm based on density-distance dynamic clustering model (PRA-DCM). A two layered recommendation algorithm based on fast density clustering is first time put forward. Users and items are clustered respectively into user clusters and item clusters by fast density clustering algorithm. For solving parameter self-adaptive tuning, a novel density-distance clustering center automatic determination method (DCC-ADM) is proposed. Cluster centers could be automatically determined in DCC-ADM for users and items clustering respectively. Moreover, double layered bipartite networks are designed to support precise recommendation. In the first layer bipartite network, user cluster and item cluster are treated as projection nodes. Top n items are recommended for each user node to construct the second layered bipartite network, where edges between user node and item node are calculated on basis of review rates. DCC-ACM algorithm is applied to the second layered bipartite network to realize personalized recommendation for each individual user. Finally a case study on Yelp social network dataset is applied. We compare this algorithm with popular ones, such as non-negative matrix factorization (NMF) algorithm and collaborative filtering algorithm (CF), and find that our algorithm has the advantages of high precision and low complexity.