基于异构网络面向多标签系统的推荐模型研究

发布时间：2018-02-01 04:58

本文关键词： 异构网络网络嵌入标签推荐标签系统图模型　出处：《软件学报》2017年10期 　论文类型：期刊论文

【摘要】：标签成为信息组织的重要方式之一,随着推荐系统的蓬勃发展,标签推荐成为学者们研究的重要问题之一.目前存在各种各样的标签系统,其功能千差万别,标签数据信息越来越复杂.目前研究往往针对特定类型标签数据,缺乏既综合考虑标签数据中不同类型对象的复杂信息又能适用于多种标签系统数据的标签推荐模型.构建了标签推荐模型Hn MTR,该模型首先针对标签数据中不同类型对象构建异构网络模型,其次对异构网络模型中不同类型顶点进行同空间映射,使不同类型的顶点和边可在同一空间进行量化比较;最后基于同空间映射后网络,引入多参数马尔可夫模型进行标签评分和推荐.通过基于豆瓣、Delicious和Meetup这3个标签系统数据实验,其结果表明,Hn MTR模型平均准确率比目前主流算法提高10%以上,取得了较好的推荐结果.
[Abstract]:Tag has become one of the important ways of information organization. With the rapid development of recommendation system, label recommendation has become one of the most important issues studied by scholars. At present, there are a variety of tag systems, whose functions vary widely. The information of tag data is becoming more and more complex. There is a lack of tag recommendation model which considers the complex information of different types of objects in tag data and can be applied to various tag system data. A tag recommendation model Hn MTR is constructed. The model firstly constructs a heterogeneous network model for different types of objects in the label data, and then maps different types of vertices in the heterogeneous network model in the same space. Different types of vertices and edges can be quantified and compared in the same space; Finally, based on the same spatial mapping network, a multi-parameter Markov model is introduced to score and recommend the label. The experiment is based on the three tagging systems of delicious and Meetup. The results show that the average accuracy of the Hn MTR model is more than 10% higher than that of the current mainstream algorithm, and a good recommended result is obtained.
【作者单位】：中国科学院软件研究所;中国科学院大学;
【基金】：中国科学院先导专项(XDA06010600)~~
【分类号】：TP391.3
【正文快照】： 1引言随着信息技术的不断发展,人类走入信息过载时代,越来越多的系统引入标签作为其信息表示方式,标签推荐技术取得快速发展.《连线》杂志创始人Kevin Kelly[1]指出,信息组织方式进化经历了3个阶段,从文件通过文件夹的组织形式存储在台式机中,到网页以链接的方式形成网络,再

【相似文献】