基于超图排序和组稀疏最优化的推荐系统

Recommendation system based on hypergraph ranking and group sparsity optimization

针对同时性图像标注和地理位置预测,基于超图排序和组稀疏最优化提出一种方法。根据超图概念构建起包含不同类型对象的超图G(V,E,w),构建包含排序向量f和查询向量y的一般推荐系统问题;通过执行组稀疏最优化,把超图顶点分割成S个非重叠对象组,对每个对象组分配不同的权值,使它能够充分利用各种类型的信息,提高图像标注和地理位置预测的准确性。基于从中国旅评网抽取出的数据集的实验结果表明,提出方法相比其它方法,能够获得更高的召回-精确率和F1测量值,能够对排名前3位的地理位置获得更高的正确预测率和累计评分。

A method for simultaneous image tagging and geographic location prediction based on hypergraph ranking and group sparsity optimization was proposed.A hypergraph G（V,E,w）including objects of different type was constructed according to the concept of hypergraph and a general recommendation system containing a ranking vector fand a query vector y was constructed.The implementation of group sparsity optimization was made.The hypergraph vertices were split into Snonoverlapping object groups and different weights were assigned to each object group,so that it made full use of various types of information to improve the accuracy of image tagging and geographic location prediction.Experimental results implemented basing on the data sets extracted from the network for Chinese brigade assessment show that the propose method,compared with the other methods,can not only get higher recall-precision and F1 measurement,but also higher correct prediction rate and cumulative geotagging prediction rate for the top 3 ranked geographical position.