基于位置社会网络的双重细粒度兴趣点推荐

发布时间：2018-07-26 11:20

【摘要】：兴趣点推荐是在基于位置社会网络(location-based social network,LBSN)中流行起来的一种全新形式的推荐.利用LBSN所包含的丰富信息进行个性化推荐能有效增强用户体验和提高用户对LBSN的依赖度.针对无显示用户偏好、兴趣非一致性和数据稀疏性等挑战性问题,研究一种针对LBSN的双重细粒度POI推荐策略,即一方面将用户的全部历史签到信息以小时为单位细分为24个时间段,另一方面将每个POI细分为多个潜在主题及其分布,同时利用用户的历史签到信息和评论信息挖掘出用户在不同时间段的主题偏好,以实现POI的Top-N推荐.为实现该推荐思路,首先,根据用户的评论信息,运用LDA模型提取出每个POI的主题分布;然后,对于每个用户,将其签到信息划分到24个时间段中,通过连接相应的POI主题分布映射出用户在不同时间段对每个主题的兴趣偏好.为解决数据稀疏问题,运用高阶奇异值分解算法对用户-主题-时间三阶张量进行分解,获取用户在每个时间段对每个主题更为准确的兴趣评分.在真实数据集上进行了性能测试,结果表明所提出的推荐策略具有较好的推荐效果.
[Abstract]:Point of interest recommendation is a new form of recommendation popular in location-based social Network LBSN. Making use of the rich information contained in LBSN for personalized recommendation can effectively enhance the user experience and enhance the dependence of users on LBSN. A dual fine-grained POI recommendation strategy for LBSN is proposed to solve the challenging problems of non-display user preference, interest inconsistency and data sparsity. That is, on the one hand, the total historical check-in information of the user is subdivided into 24 hours, on the other hand, each POI is subdivided into multiple potential topics and their distribution. At the same time, the user's topic preference in different time periods is mined by using the user's history check-in information and comment information, in order to realize the Top-N recommendation of POI. In order to realize this recommendation, firstly, the topic distribution of each POI is extracted by using LDA model according to the user's comment information, and then, for each user, the check-in information is divided into 24 time periods. By connecting the corresponding POI topic distribution, the user's interest preference for each topic in different time periods is mapped. In order to solve the problem of data sparsity, a higher order singular value decomposition algorithm is used to decompose the user-theme-time third-order Zhang Liang to obtain a more accurate interest score for each topic in each time period. The performance tests on real data sets show that the proposed recommendation strategy has a good recommendation effect.
【作者单位】：江西财经大学信息管理学院;江西财经大学江西省高校数据与知识工程重点实验室;
【基金】：国家自然科学基金项目(61772245,61262009) 江西省自然科学基金项目(20151122040083) 江西省优势科技创新团队建设计划项目(20113BCB24008) 江西省教育厅重点科技项目(GJJ160419)~~
【分类号】：TP391.3

【相似文献】