基于标签权重评分的推荐模型及算法研究  

Research on the Modeling and Related Algorithms of Label-Weight Rating Based Recommendation System

作　　者：孔欣欣[1] 苏本昌[1] 王宏志[1] 高宏[1] 李建中[1] KONG Xin-Xin SU Ben-Chang WANG Hong-Zhi GAO Hong LI Jian-Zhong(School of Computer Science and Technology, Harbin Institute of Technology, Harbin 150001)

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《计算机学报》2017年第6期1440-1452,共13页Chinese Journal of Computers

基　　金：国家自然科学基金(61472099;61003046);国家"九七三"重点基础研究发展规划项目基金(2012CB316200);国家科技支撑计划(2015BAH10F00)资助~~

摘　　要：推荐系统已经被越来越频繁地应用到电子商务网站与一些社交网站,在提高用户满意度的同时也带来了巨大的商业利益.然而,当前的推荐算法由于原始数据的不完整性以及算法本身处理数据的特殊性,导致推荐效果不理想.例如,某些推荐系统会产生冷启动、复杂兴趣推荐困难、解释性差等问题.为此,该文提出一种基于标签权重评分的推荐系统模型(Label-Weight Rating based Recommendation,LWR),旨在使用一种较为简洁的方式——标签权重评分来获取用户最准确的评价和需求,并通过改进当前的一些推荐算法来处理标签权重评分数据,从而生成对用户的推荐,最后以标签权重评分的形式向用户展示推荐结果并作出合理的解释.扩展实验中,通过电影推荐实验,证明了该文技术的有效性和可行性.Recommendation System has been frequently applied into various e-commerce websites and social networking sites. With improving users＇ satisfaction, recommendation system has also brought huge commercial interests. However, as the original data is incomplete and some recommendation algorithms have their own special way of processing data, current recommendation system sometimes cannot work very well. For example, some recommendation systems are bothered with cold-start problem, difficult for complex interest recommendation problem, poor interpretability and so on. Consequently, in the paper, we propose a recommendation system modeling based on label-weight rating. In this system, first we will get the most accurate evaluation and demanding information of users in a more concise way--label-weight rating method. Then we will generate recommendations using improved existing recommendation algorithm. Finally, we will show the recommendations to the users in the form of label-weight rating and make reasonable explanation to users. In the extended experiments we design a series of movie recommendations experiments to prove the effectiveness and feasibility of the modeling.

关 键 词：推荐系统 标签 标签权重评分 数据挖掘 人工智能 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]