融合潜在兴趣和多类型情景信息的兴趣点推荐模型  

Point of Interest Recommendation Model Combining Potential Check-ins and Multi-type Contextual Information

作　　者：陈江美 张岐山 张文德[2] 何珑[3] CHEN Jiang-mei;ZHANG Qi-shan;ZHANG Wen-de;HE Long(School of Economics&Management,Fuzhou University,Fuzhou 35QW8,China;Institute of Information Management,Fuzhou University,Fuzhou 350108,China;Network and Information Construction Office,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学经济与管理学院,福建福州350108 [2]福州大学信息管理研究所,福建福州350108 [3]福州大学网络与信息化建设办公室,福建福州350108

出　　处：《情报科学》2021年第3期143-149,160,共8页Information Science

基　　金：国家自然科学基金青年项目“基于灰色方法的社交网络群体识别问题研究”(61300104);福建省自然科学基金项目“基于灰色方法的个性化隐私保护数据挖掘研究”(2018J01791);赛尔网络下一代互联网创新项目“融合评论标签的个性化学习资源推荐关键技术研究”(NG20170522),“面向慕课课程评论的情感标签信息抽取关键技术研究”(NGII20180509)。

摘　　要：【目的/意义】在现有的兴趣点推荐工作普遍存在数据稀疏和低精确率问题的基础上,提出了一种融合潜在兴趣和多类型情景信息的兴趣点推荐模型。【方法/过程】该模型分为矩阵填充和矩阵分解两个阶段。首先利用社交和地理信息建立矩阵填充模型,为每个用户学习一组待填充兴趣点。其次,将分类信息嵌入加权矩阵分解模型来学习用户偏好。最后,采用自适应核密度估计对地理影响建模,结合矩阵分解的结果得到GSC-WMF模型。【结果/结论】实验结果表明,该模型在推荐的准确率和召回率上相较其他主流模型取得了更好的结果。【创新/局限】提出了矩阵填充模型来发掘用户的潜在兴趣,并有效地将融合分类信息来解决用户隐式信息反馈问题。在未来的研究工作中,将考虑利用深度学习技术来改善推荐性能。【Purpose/significance】Since the current works of point of interest recommendation exist sparsity and low accuracy, this paper proposed a point of interest recommendation model combining potential check-ins and multi-type contextual information.【Method/process】The model is divided into two stages of matrix filling and matrix decomposition. Firstly, established a matrix filling model by using social and geographic information to learn a set of points of interest for each user. Secondly, the classification information is embedded into the weighted matrix decomposition model to learn user preferences. Finally, the adaptive kernel density estimation is used to model the geographical influence, and the GSC-WMF model is obtained by combining the results of matrix decomposition.【Result/conclusion】The experimental results show the proposed model outperforms state-of-the-art recommendation models in terms of precision and recall.【Innovation/limitation】Matrix filling model is proposed to explore the potential interest, and the category information is effectively fused to address the problem of implicit information feedback. In the future research work, we will introduce the deep learning to improve recommendation performance.

关 键 词：兴趣点推荐 社交关系 分类信息 加权矩阵分解 核密度估计 地理信息 

分 类 号：G254[文化科学—图书馆学]