第6期 梁丽君，等：融合用户特征优化聚类的协同过滤算法 ·1095· 由图4和图5可知，随着横坐标的变换， mendation algorithm combing item features and trust rela- PCEDS模型的准确率、召回率与PCC模型相比 tionship of mobile users[J].Journal of software,2014, 均有提高。 25(8):1817-1830. [7]LIU Haifeng.HU Zheng,MIAN A,et al.A new user simil- 4结束语 arity model to improve the accuracy of collaborative filter- ing[J].Knowledge-based systems,2014,56:156-166. 本文根据用户的属性特征采用优化的K [8]韦素云，肖静静，业宁，基于联合聚类平滑的协同过滤算 means算法首先对用户进行聚类，聚类中心是基 法[.计算机研究与发展，2013,50(S2):163-169. 于用户活跃度选取的，比传统的随机选取的聚类 WEI Suyun,XIAO Jingjing,YE Ning.Collaborative filter- 中心更有代表性，聚类的结果会更合理，并且缓 ing algorithm based on co-clustering smoothing[J].Journ- 解了协同过滤推荐系统的可扩展性问题。 al of computer research and development,2013,50(S2): PCEDS模型具有融合了用户本身的属性特征信 163-169. 息和用户偏好信息的相似度计算的优点，在计算 [9]LEE DD,SEUNG H S.Learning the parts of objects by 用户属性特征相似度时是基于属性特征信任度的 non-negative matrix factorization[J].Nature,1999, 相似度，提高了传统的相似度算法，根据相似度 401(6755):788-791 选取的目标用户最近邻居更准确。PCEDS相似 [10]ELKAHKY A M,SONG Yang,HE Xiaodong.A multi- 度算法缓解了因用户评分数据少或新用户登录时 view deep learning approach for cross domain user mod- 导致的用户冷启动问题。 eling in recommendation systems[Cl//Proceedings of the 24th International Conference on World Wide Web 参考文献： Florence,Italy,2015:278-288. [11]陈克寒，韩盼盼，吴健.基于用户聚类的异构社交网络 [1]王国霞，刘贺平.个性化推荐系统综述U.计算机工程与 推荐算法[.计算机学报，2013,36(2)：349-359 应用，2012,48(7：66-76 CHEN Kehan,HAN Panpan,WU Jian.User clustering WANG Guoxia,LIU Heping.Survey of personalized re- based social network recommendation[J].Chinese journ- commendation system[J.Computer engineering and ap- al of computers,.2013,36(2):349-359. plications..2012,48(7):66-76. [12]张顺龙，库涛，周浩.针对多聚类中心大数据集的加速 [2]Jamali M,Ester M.A transitivity aware matrix factoriza- K-means聚类算法.计算机应用研究，2016,33(2： tion model for recommendation in social networks[Cl//Pro- 413-416. ceedings of the Twenty-Second International Joint Confer- ZHANG Shunlong,KU Tao,ZHOU Hao.Accelerate K- ence on Artificial Intelligence.Barcelona,Spain,2011: means for multi-center clustering of big datasets[J].Ap- 2644-2649. plication research of computers,2016,33(2):413-416. [3]WEI Jian,HE Jianhua,CHEN Kai,et al.Collaborative fil- [13]贾洪杰，丁世飞，史忠植.求解大规模谱聚类的近似加 tering and deep learning based recommendation system for 权核K-means算法[J].软件学报，2015,26(11)： cold start items[J].Expert systems with applications,2017, 2836-2846 69:29-39 JIA Hongjie,DING Shifei,SHI Zhongzhi.Approximate [4]AGGARWAL CC.An introduction to data mining[M]// weighted kernel K-means for large-scale spectral cluster- AGGARWAL CC.Data Mining:the Textbook.Cham: ing[J].Journal of software,2015,26(11):2836-2846. Springer,2015:1-26. [14]DRAISMA J.HOROBET E,OTTAVIANI G,et al.The [S]孙天吴，黎安能，李明，等.基于Hadoop分布式改进聚类 Euclidean distance degree of an algebraic variety[J]. 协同过滤推荐算法研究[.计算机工程与应用，2015， Foundations of computational mathematics,2016,16(1): 51(15):124128 99-149. SUN Tianhao,LI Anneng,LI Ming,et al.Study on distrib- uted improved clustering collaborative filtering algorithm 作者简介： based on Hadoop[J].Computer engineering and applica- 梁丽君，硕士研究生，主要研究方 向为个性化推荐系统。 tions,2015,51(15):124128. [6]胡勋，孟祥武，张玉洁，等.一种融合项目特征和移动用 户信任关系的推荐算法[].软件学报，2014,25(8)： 1817-1830 HU Xun,MENG Xiangwu,ZHANG Yujie,et al.Recom-由 图 4 和 图 5 可知，随着横坐标的变换， PCEDS 模型的准确率、召回率与 PCC 模型相比 均有提高。 4 结束语 本文根据用户的属性特征采用优化的 K￾means 算法首先对用户进行聚类，聚类中心是基 于用户活跃度选取的，比传统的随机选取的聚类 中心更有代表性，聚类的结果会更合理，并且缓 解了协同过滤推荐系统的可扩展性问题。 PCEDS 模型具有融合了用户本身的属性特征信 息和用户偏好信息的相似度计算的优点，在计算 用户属性特征相似度时是基于属性特征信任度的 相似度，提高了传统的相似度算法，根据相似度 选取的目标用户最近邻居更准确。PCEDS 相似 度算法缓解了因用户评分数据少或新用户登录时 导致的用户冷启动问题。 参考文献： 王国霞, 刘贺平. 个性化推荐系统综述 [J]. 计算机工程与 应用, 2012, 48(7): 66–76. WANG Guoxia, LIU Heping. Survey of personalized re￾commendation system[J]. Computer engineering and ap￾plications, 2012, 48(7): 66–76. [1] Jamali M, Ester M. A transitivity aware matrix factoriza￾tion model for recommendation in social networks[C]//Pro￾ceedings of the Twenty-Second International Joint Confer￾ence on Artificial Intelligence. Barcelona, Spain,2011: 2644–2649. [2] WEI Jian, HE Jianhua, CHEN Kai, et al. Collaborative fil￾tering and deep learning based recommendation system for cold start items[J]. Expert systems with applications, 2017, 69: 29–39. [3] AGGARWAL C C. An introduction to data mining[M]// AGGARWAL C C. Data Mining: the Textbook. Cham: Springer, 2015: 1–26. [4] 孙天昊, 黎安能, 李明, 等. 基于 Hadoop 分布式改进聚类 协同过滤推荐算法研究 [J]. 计算机工程与应用, 2015, 51(15): 124–128. SUN Tianhao, LI Anneng, LI Ming, et al. Study on distrib￾uted improved clustering collaborative filtering algorithm based on Hadoop[J]. Computer engineering and applica￾tions, 2015, 51(15): 124–128. [5] 胡勋, 孟祥武, 张玉洁, 等. 一种融合项目特征和移动用 户信任关系的推荐算法 [J]. 软件学报, 2014, 25(8): 1817–1830. HU Xun, MENG Xiangwu, ZHANG Yujie, et al. Recom- [6] mendation algorithm combing item features and trust rela￾tionship of mobile users[J]. Journal of software, 2014, 25(8): 1817–1830. LIU Haifeng, HU Zheng, MIAN A, et al. A new user simil￾arity model to improve the accuracy of collaborative filter￾ing[J]. Knowledge-based systems, 2014, 56: 156–166. [7] 韦素云, 肖静静, 业宁. 基于联合聚类平滑的协同过滤算 法 [J]. 计算机研究与发展, 2013, 50(S2): 163–169. WEI Suyun, XIAO Jingjing, YE Ning. Collaborative filter￾ing algorithm based on co-clustering smoothing[J]. Journ￾al of computer research and development, 2013, 50(S2): 163–169. [8] LEE D D, SEUNG H S. Learning the parts of objects by non-negative matrix factorization[J]. Nature, 1999, 401(6755): 788–791. [9] ELKAHKY A M, SONG Yang, HE Xiaodong. A multi￾view deep learning approach for cross domain user mod￾eling in recommendation systems[C]//Proceedings of the 24th International Conference on World Wide Web. Florence, Italy, 2015: 278-288. [10] 陈克寒, 韩盼盼, 吴健. 基于用户聚类的异构社交网络 推荐算法 [J]. 计算机学报, 2013, 36(2): 349–359. CHEN Kehan, HAN Panpan, WU Jian. User clustering based social network recommendation[J]. Chinese journ￾al of computers, 2013, 36(2): 349–359. [11] 张顺龙, 库涛, 周浩. 针对多聚类中心大数据集的加速 K-means 聚类算法 [J]. 计算机应用研究, 2016, 33(2): 413–416. ZHANG Shunlong, KU Tao, ZHOU Hao. Accelerate K￾means for multi-center clustering of big datasets[J]. Ap￾plication research of computers, 2016, 33(2): 413–416. [12] 贾洪杰, 丁世飞, 史忠植. 求解大规模谱聚类的近似加 权核 K-means 算法 [J]. 软件学报, 2015, 26(11): 2836–2846. JIA Hongjie, DING Shifei, SHI Zhongzhi. Approximate weighted kernel K-means for large-scale spectral cluster￾ing[J]. Journal of software, 2015, 26(11): 2836–2846. [13] DRAISMA J, HOROBET E, OTTAVIANI G, et al. The Euclidean distance degree of an algebraic variety[J]. Foundations of computational mathematics, 2016, 16(1): 99–149. [14] 作者简介： 梁丽君，硕士研究生，主要研究方 向为个性化推荐系统。 第 6 期 梁丽君，等：融合用户特征优化聚类的协同过滤算法 ·1095·