%0 Journal Article
%T 融合隐性和显性社交信息的连续兴趣点推荐方法研究<br>Research on Successive Point-of-Interest Recommendation Model with Explicit and Implicit Social Information
%A 谷敏
%A 郑建国
%J Computer Science and Application
%P 2212-2226
%@ 2161-881X
%D 2020
%I Hans Publishing
%R 10.12677/CSA.2020.1012233
%X <div style=\"text-align:justify;\">
	<span style=\"font-size:14px;\">利用位置社交网络中的签到数据和用户社交关系开展对连续兴趣点推荐问题的研究。本文利用非负矩阵分解技术，构建用户的社交信息模型，考虑用户隐性和显性社交关系，将社交网络图转化为低维特征向量；改进LSTM结构，提出融合社交信息的连续兴趣点推荐模型SLSTM，该模型共享非负矩阵分解技术训练的图顶点向量，实现了图结构数据和签到序列数据的有效融合。在Gowalla和BrightKite签到数据集上进行实验，结果表明SLSTM模型优于目前主流的连续兴趣点推荐算法。在Gowalla数据集上，SLSTM模型在Recall@10指标的性能较SERM模型提高了17%，在BrightKite数据集上，Recall@10指标提高了15.2%，说明SLSTM模型在连续兴趣点推荐中的有效性。基于位置的社交网络包含丰富的上下文信息，本文只着重考虑了社交信息对推荐结果的影响。用户隐性社交关系对用户的行为偏好有重要影响；融合隐性和显性社交信息的连续兴趣点推荐方法具有较好的推荐结果。<br />
This paper uses the check-in data and the user’s social relationship in the location based social networks (LBSN) to carry out the research on the successive point-of-interests recommendation. Firstly, this paper uses the non-negative matrix decomposition technique to model users’ implicit and explicit social information and transform the social network graph into a low-dimensional fea-ture vector. Secondly, by improving the LSTM model, this paper proposes a successive point-of-interest recommendation SLSTM that merges social information, which shares the graph vertex vector trained by the non-negative matrix decomposition technique and realizes the effective fusion of graph structure data and check-in sequence data. The experimental results show that the SLSTM model improves the performance of the Recall@10 metric by 17% and 15.2% over the SERM model on the Gowalla and BrightKite dataset respectively. Location-based social networks contain rich contextual information. This paper only focuses on the impact of social information on recommendation results. The implicit social relationships have an important influence on their behavioral preferences in location-based social networks. Successive point-of-interest recommenda-tion model with explicit and implicit social information performs better than other recommendation algorithms.</span> 
</div>
%K 位置社交网络，连续兴趣点推荐，社交信息，非负矩阵分解，长短时记忆网络<br>Location-Based Social Network
%K Successive Point-of-Interest Recommendation
%K Social Information
%K Non-Negative Matrix Decomposition
%K Long Short-Term Memory
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=39241