基于多影响嵌入的个性化POI推荐方法

发布时间：2018-01-28 04:27

本文关键词： 基于位置服务 POI推荐嵌入学习图嵌入序列嵌入　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着智能移动设备的快速普及以及基于位置社交网络服务(Location-based Social Networking Services,LBSNs)的快速发展,基于 check-in 数据挖掘的 POI(Point of Interest)推荐成为帮助用户发现新场所和探索不熟悉区域的重要方式。然而,POI推荐面临严重的数据稀疏性问题,用户旅行局部性现象更是恶化了这一问题。最近许多相关工作试图通过考虑社交、时间、地理、序列、语义等方面影响来解决上述数据稀疏性问题,但是他们仅利用了部分方面影响,没有一个并能准确整合多方面影响的方法。为了解决上述挑战,我们提出了一个基于图和序列联合嵌入的POI推荐方法。我们通过对7张二分图(用户-用户图、用户-时间段图、POI-时间段图、POI-区域层次图、POI-类别层次图、用户-性别图以及用户-POI图)和check-in序列进行联合嵌入学习,整合了社交、时间、地理、语义、用户性别、用户偏好以及序列方面影响。为了捕获check-in序列中的语义信息,我们方法利用了序列嵌入方法(word2vec),而其它方面影响则利用图嵌入方法,然后通过联合训练算法对上述多方面影响进行联合嵌入学习。需要注意的是我们方法具有一定的扩展性,可以很方便地整合其它方面影响,从而更好地解决数据稀疏性问题,为用户提供高质量的POI推荐。为了验证我们方法的效果,我们在来自Foursquare的大规模真实数据集上进行了充分的实验。实验结果表明,本文提出方法明显超过了其它对比方法。此外,我们还通过实验研究了本文考虑的各方面影响对推荐效果提升的作用大小,结果发现时间和语义影响相对其它方面影响在推荐效果的提升上作用更明显。
[Abstract]:With the rapid spread of smart mobile devices and location-based Social Networking Services. The rapid development of LBSNs. POI(Point of Interest-based check-in data mining is an important way to help users discover new places and explore unfamiliar areas. POI recommends serious data sparsity, which is exacerbated by the phenomenon of user travel locality. Recently, a lot of related work has attempted to consider social, time, geography, and sequence. Semantic impact to solve the problem of data sparsity, but they only take advantage of some aspects of the impact, there is no way to accurately integrate the various aspects of the impact. In order to solve the above challenges. We propose a POI recommendation method based on graph and sequence embedding. We use seven bipartite graphs (user-user graph, user-time graph) and POI-time graph. POI- regional hierarchy map (POI- category hierarchy map, user-gender map and user-POI map) and check-in sequence are jointly embedded learning, integrating social, time, geography. In order to capture semantic information in check-in sequences, our method utilizes sequence embedding method (Word2vec.). Other aspects of the influence is based on graph embedding method, and then the joint training algorithm is used to study the above effects. It is important to note that our method has a certain expansibility. Can easily integrate other aspects of the impact, so as to better solve the problem of data sparsity, provide users with high-quality POI recommendations, in order to verify the effectiveness of our method. We have carried out sufficient experiments on the large scale real data set from Foursquare. The experimental results show that the proposed method is obviously superior to other comparison methods. We also study the effect of each aspect on the improvement of recommendation effect through experiments. The results show that the effect of time and semantics on the promotion of recommendation effect is more obvious than that of other aspects.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】