摘要： 基于用户偏好物品与其在网上浏览的历史记录,推荐系统都能够向用户推荐项目和预测未来的采购意愿,但稀疏性、冷启动等问题影响该方法的推荐效果。为此,提出将深度本体与用户标签结合的Web推荐方法。利用深度本体项目之间的语义关系对数据矩阵降维,根据用户提
供的标签信息,将点击流映射到本体中,结合深度本体中项目之间的关系扩展推荐结果,推荐出top-n信息。实验结果表明,与传统的基于本体方法相比,该方法可解决稀疏性和冷启动等问题,同时推荐的准确性和时效性都有较好的效果。
关键词:
推荐系统,
标签,
深度本体,
降维,
点击流,
推荐扩展
Abstract: Based on users’ likings of items and their browsing history on the world wide Web,recommendation systems are able to predict and recommend items and future purchases to users.However,sparse and cold start problems influence the effect of this approach.This paper proposes a method of Web recommendationsystem based on tags and deep ontology.Through using deep ontology relations and tags marked by the users,a dimensionality reduction method based on the deep ontology is proposed,and click stream is mapped on ontology,and top-n recommendations are provided.Experimental result shows that the method for sparse and cold starting problems has a more obvious improvement,recommendation accuracy and timeliness have better results than traditional methods based on ontology.
Key words:
recommendation system,
tags,
deep ontology,
dimensionality reduction,
clickstream,
recommendation expanding
中图分类号: