A novel recommendation system in location-based social networks using distributed ELM
A novel recommendation system in location-based social networks using distributed ELM
复制标题
使用分布式 ELM 的基于位置的社交网络中的新型推荐系统
DOI：
10.1007/s12293-017-0227-4
复制
发表时间：
2018
影响因子：
4.7
通讯作者：
Zhang Zhen
中科院分区：
文献类型：
--
作者：
Zhao Xiangguo;Ma Zhongyu;Zhang Zhen
基于位置的社交网络（LBSNs）已经成为人们相互交流的热门平台。随着越来越多的用户使用LBSNs分享他们的经验和感受，推荐问题在学术界和工业界都引起了相当大的关注。机器学习已经广泛应用于许多推荐系统中，用于向LBSNs的用户推荐新的朋友或感兴趣的地方（poi）。然而，现有的推荐系统大多功能单一，只使用小规模的数据集来提供推荐服务。在大数据时代，推荐系统应该能够充分利用有限的计算资源，从大规模的LBSN数据中挖掘潜在的关系。本文利用分布式极限学习机GR-DELM，在大规模数据集中同时考虑好友推荐和POI推荐，提出了一种新的通用推荐系统。对于POI推荐，提取了三个特征：(1)地理影响特征，(2)人气影响特征，(3)社会影响特征。对于朋友推荐，提取了两个特征：(1)基于邻居的特征和(2)基于路径的特征。这些特点进一步提高了大规模推荐的效率和准确性。最后，一系列的实验表明，GR-DELM系统优于现有的推荐系统。
Location-based social networks (LBSNs) have become a popular platform for people to communicate with each other. The recommendation problem has attracted considerable attention in both academia and industry as increasingly more users share their experiences and feelings using LBSNs. Machine learning has been widely used in many recommendation systems for recommending new friends or places of interest (POIs) to users in LBSNs. However, the majority of the existing recommendation systems were single function and only used small-scale datasets to provide recommendation services. In the era of big data, recommendation systems should have the ability to fully utilize limited computing resources for mining potential relationships from large-scale LBSN data. In this paper, a novel generic recommendation system is proposed by utilizing a distributed extreme learning machine called GR-DELM, which considers both friend recommendation and POI recommendation in large-scale datasets. For POI recommendation, three features are extracted: (1) geography-influenced feature, (2) popularity-influenced feature, and (3) social-influenced feature. For friend recommendation, two features are extracted: (1) neighborhood-based feature and (2) path-based feature. These features further improve the efficiency and accuracy of large-scale recommendation. Finally, a series of experiments demonstrate that the GR-DELM system outperforms the existing recommendation systems.