计算机应用 ›› 2015, Vol. 35 ›› Issue (1): 162-166.DOI: 10.11772/j.issn.1001-9081.2015.01.0162

• 人工智能 • 上一篇    下一篇

新的基于多目标优化的推荐算法

厍向阳, 蔡院强, 董立红   

  1. 西安科技大学 计算机科学与技术学院, 西安710054
  • 收稿日期:2014-07-29 修回日期:2014-09-12 出版日期:2015-01-01 发布日期:2015-01-26
  • 通讯作者: 蔡院强
  • 作者简介:厍向阳(1968-),男,陕西周至人,教授,博士,主要研究方向:数据挖掘、智能信息处理、人工智能、模式识别、复杂系统建模与优化;蔡院强(1988-),男,陕西绥德人,硕士研究生,主要研究方向:机器学习、推荐系统;董立红(1968-),女,河北唐山人,教授,博士,主要研究方向:数据处理、数据挖掘、智能控制、煤矿安全监测监控.
  • 基金资助:

    陕西省教育厅专项科研计划项目(12JK0787).

New recommendation algorithm based on multi-objective optimization

SHE Xiangyang, CAI Yuanqiang, DONG Lihong   

  1. School of Computer Science and Technology, Xi'an University of Science and Technology, Xi'an Shaanxi 710054, China
  • Received:2014-07-29 Revised:2014-09-12 Online:2015-01-01 Published:2015-01-26

摘要:

针对目前推荐系统效率问题,采用线上、线下分离策略,构建一种新的推荐系统框架.针对推荐系统多目标性和目前众多推荐算法适应性局限等特性,采用混合策略,提出一种新的多目标推荐算法.首先,对多个推荐算法进行加权混合;然后,构建以权重序列为自变量,推荐评价指标F调和率、多样性和新颖度为目标函数的多目标优化模型;其次,采用SPEA2多目标优化算法进行优化求解;最后,基于用户的购物偏好和Pareto解集向用户有针对性地进行购物推荐.实验结果表明:新的推荐算法较子推荐算法在F调和率上持平,在多样性上提高了1%,在新颖度上提高了11.5%;多目标的各个Pareto解在解空间中分布形成了密集邻近的点曲线.该推荐算法能够满足不同购物偏好用户的推荐要求.

关键词: 推荐算法, 多目标优化, 权重, 混合策略, 分离策略

Abstract:

In view of the efficiency problem of multi-objective recommender systems, this paper utilized the online and offline separation strategy to construct a new frame of recommender system. Aiming at the multi-objective feature of recommender system and current recommendation algorithms' limitations in adaptability, this paper put forward a new multi-objective recommendation algorithm based on the hybrid strategy. Firstly, the algorithm did weighted mix of multiple recommendation algorithms. Secondly, it established a multi-objective optimization model, using the weight sequence as variables and evaluation metrics including F-score, diversity and novelty as objective functions. Then, it optimized the solution through a second version of Strength Pareto Evolutionary Algorithm (SPEA2). Finally, it recommended items to users based on users' shopping preferences and the Pareto set. The experimental results show that: compared with the best single metric sub-recommendation algorithm, the new recommendation algorithm is nearly as well in the F-score, meanwhile increases by 1% in the diversity and increases by 11.5% in the novelty; the distribution of various Pareto solutions of multi-objective forms a dense and neighboring point curve in the solution space. So the recommender algorithm can satisfy the recommend requirements of users with different shopping preferences.

Key words: recommendation algorithm, multi-objective optimization, weight, hybrid strategy, separation strategy

中图分类号: