一种结合时间因子聚类的群组兴趣点推荐模型

Group POI Recommendation Model Based on Time Factor Clustering

推荐系统本质上是一种信息检索工具,它检索出有用信息并推荐给特定的用户.组推荐系统通过不同的融合策略融合群组偏好,支持群组用户访问当前的热门兴趣点.传统组推荐模型没有将时间因子对用户选择兴趣点的影响计算在内,且传统协同过滤推荐算法往往对数据的稀疏性较为敏感.本文提出一个混合推荐模型(AGRT),综合K-均值聚类算法和隐语义模型(LFM)技术,将其应用于群组兴趣点.考虑到用户在不同时间点的不同兴趣偏好,AGRT利用K-means算法对用户数据集合基于时间点聚类,划分为不同的簇,在与当前推荐时间最为接近的用户数据簇上进行兴趣点推荐,采用LFM隐语义模型对用户数据进行矩阵分解,通过将分解矩阵再次相乘获得用户对未评分地点的评分数据,解决用户数据稀疏性的问题.实验结果表明,AGRT模型在低相似度(随机)群组和高相似度群组评测条件下下较文献[3]中提出的HAaB提高了5. 19%和2. 06%,具有有效的改进.

Recommendation system is essentially an information retrieval tool,which explores useful information and present it to specific individuals. By aggregating group preferences through different aggregation strategies,Group recommendation system help group users to access current hot interest points. The traditional group recommendation model does not consider the impact of time factor on user interest point selection,while the recommendation algorithm based on collaborative filtering is often sensitive to data sparsity. This paper proposes a hybrid recommendation model( AGRT),which integrates K-means clustering algorithm and implicit semantic model( LFM) technology,and applies it to group interest points. taking into account the user in different time points of different preferences,The AGRT model uses K-means algorithm to cluster user data sets based on time points,divide them into different clusters,recommend interest points on user data clusters which are closest to the current recommendation time,decompose user data using LFM implicit semantics model,and obtain user’s score data on Unrated sites by multiplying the decomposition matrix again,so as to solve the problem of sparsity of user data. The experimental results show that the AGRT model improves 5. 19% and 2. 06% compared with HAaB under the condition of low similarity( random) group and high similarity group.