基于用户情境的协同推荐算法研究与应用
|
摘要
随着互联网的普及和电子商务的迅猛发展,个性化的推荐系统成为电子商务领域一个重要的研究内容。推荐算法作为个性化推荐系统的核心,它的性能与整个推荐系统的推荐效率、推荐质量以及用户的使用感受紧密相关。目前,现有的推荐系统中运用较多的推荐算法有基于关联规则的推荐算法、基于内容的推荐算法、协同推荐算法。
基于关联规则的推荐算法中,关联规则的数量会随着系统的规模增大而巨增。基于内容的推荐算法只能推荐与用户兴趣相似的资源,无法发现新的、潜在的用户兴趣。协同推荐技术作为至今最成功的推荐技术,已经在许多实际的推荐系统中得到了大量的应用。它虽然可以为用户发现潜在的兴趣,但传统的协同推荐算法均是由用户对项目的评分作为切入点。
由于评分并不能全面反映一个人的兴趣爱好。同时,个人的兴趣爱好与其职业、年龄、教育水平等一系列自身的因素有密不可分的联系,有相似属性的人群也容易产生相似的爱好。因此,本文提出了基于用户情境的协同推荐算法。该算法按照用户情境对用户进行聚类,使得每个用户能够准确找到与自己相似度高的邻居。在同一类中,根据用户的历史评分以及项目间的相异性,为目标项目计算预测评分,从而获得目标用户所需要的推荐结果。
本文的主要研究工作有:
(1)在深入分析现有推荐算法的情境缺失问题后,结合情境语义学以及对用户的兴趣爱好有影响的、自身的自然属性和社会属性,提出了用户情境的描述方式以及形式化表示方法,并进一步对用户情境进行研究,提出了用户情境的分类方法。
(2)研究常见变量类型的相异度计算方法,利用相异度矩阵,给出了多情境因素下静态用户情境的聚类方法,并结合Slope One算法中对目标项目的预测值方法,提出了基于用户静态情境的协同推荐算法。
(3)在MovieLens数据集上利用Matlab对传统的基于项目的协同推荐算法、Slope One算法和本文提出的算法进行了对比试验。实验结果表明,本算法较之传统推荐算法和Slope One算法在平均绝对误差值上有一定的提高,证明了本算法的可行性与有效性。
(4)结合作者参与的高等教育出版社基于本体、语义和语用的智能化教育平台项目,设计了一个采用基于用户情境的协同推荐算法的个性化推荐系统模型,以此说明该算法在实际中的应用方式。
With the popularity of the Internet and the rapid development of e-commerce, personalized recommendation system has become an important research area of electronic commerce content. Recommended algorithm is the core of personalized recommendation system. Its performance has closely related to the recommended efficiency of the whole recommendation system, recommended quality and users’experience. At present, the popular use of algorithms is based on association rules for recommendation algorithm, content-based recommendation algorithm, and collaborative recommendation algorithm in the existing recommendation systems.
The amount of association rules would hugely increase with the system size increasing in recommendation algorithm based on association rules. Content-based recommendation algorithm can only provide the user for profile similar resources without discovering new or potential user interest. Collaborative Filtering technologies as the most successful recommendation techniques have been implemented and applied in many practical systems. Although it can identify potential interest for the user, the traditional collaborative recommendation algorithm is the project's score by the user as a starting point.
The score does not fully reflect the interests of a person. Meanwhile, the personal interests and hobbies have the inextricably link with the occupation, age, education level and a series of factors, and similar attributes of people are also prone to similar interest. So, this paper proposed the collaborative recommendation algorithm based on the user context. According to user context, the algorithm clusters the users, allows each user to be able to accurately find the neighbors with their high similarity. In the same class, according to the user's history score and the dissimilarity between projects, it predicted score calculated for the target item to obtain the results of recommended target which users need.
This thesis completes the following work:
(1) After deep analyzing the problem of context loss, this thesis puts forward the user context and its formalization combined with Situation Semantics and the factors affecting the interests of users, and their natural attributes and social attributes. And further study for user context, this thesis puts forward the classification method of user context.
(2) The clustering method of static user context with multiple situational factors is given by the research of dissimilarity degree calculation and dissimilarity degree matrix. After that, this thesis puts forward the collaborative recommendation algorithm based on user context binding predictive value for target item method of Slope One algorithm.
(3) The collaborative recommendation algorithm based on user context is compared with traditional collaborative recommendation algorithm based on items and Slope One, using Matlab to carry out simulation experiment in MovieLens. Experimental results show that the algorithm is better in the mean absolute error against traditional algorithm and the Slope One algorithm. So, the feasibility and validity of this algorithm are verified.
(4) In combination with the intelligent educational platform project of Higher Education Press based on ontology, semantic and pragmatic, the article designs a scenario based on the user's personalized recommendation algorithm for collaborative recommendation system model as an example of the algorithm in practice application form.
基于关联规则的推荐算法中,关联规则的数量会随着系统的规模增大而巨增。基于内容的推荐算法只能推荐与用户兴趣相似的资源,无法发现新的、潜在的用户兴趣。协同推荐技术作为至今最成功的推荐技术,已经在许多实际的推荐系统中得到了大量的应用。它虽然可以为用户发现潜在的兴趣,但传统的协同推荐算法均是由用户对项目的评分作为切入点。
由于评分并不能全面反映一个人的兴趣爱好。同时,个人的兴趣爱好与其职业、年龄、教育水平等一系列自身的因素有密不可分的联系,有相似属性的人群也容易产生相似的爱好。因此,本文提出了基于用户情境的协同推荐算法。该算法按照用户情境对用户进行聚类,使得每个用户能够准确找到与自己相似度高的邻居。在同一类中,根据用户的历史评分以及项目间的相异性,为目标项目计算预测评分,从而获得目标用户所需要的推荐结果。
本文的主要研究工作有:
(1)在深入分析现有推荐算法的情境缺失问题后,结合情境语义学以及对用户的兴趣爱好有影响的、自身的自然属性和社会属性,提出了用户情境的描述方式以及形式化表示方法,并进一步对用户情境进行研究,提出了用户情境的分类方法。
(2)研究常见变量类型的相异度计算方法,利用相异度矩阵,给出了多情境因素下静态用户情境的聚类方法,并结合Slope One算法中对目标项目的预测值方法,提出了基于用户静态情境的协同推荐算法。
(3)在MovieLens数据集上利用Matlab对传统的基于项目的协同推荐算法、Slope One算法和本文提出的算法进行了对比试验。实验结果表明,本算法较之传统推荐算法和Slope One算法在平均绝对误差值上有一定的提高,证明了本算法的可行性与有效性。
(4)结合作者参与的高等教育出版社基于本体、语义和语用的智能化教育平台项目,设计了一个采用基于用户情境的协同推荐算法的个性化推荐系统模型,以此说明该算法在实际中的应用方式。
With the popularity of the Internet and the rapid development of e-commerce, personalized recommendation system has become an important research area of electronic commerce content. Recommended algorithm is the core of personalized recommendation system. Its performance has closely related to the recommended efficiency of the whole recommendation system, recommended quality and users’experience. At present, the popular use of algorithms is based on association rules for recommendation algorithm, content-based recommendation algorithm, and collaborative recommendation algorithm in the existing recommendation systems.
The amount of association rules would hugely increase with the system size increasing in recommendation algorithm based on association rules. Content-based recommendation algorithm can only provide the user for profile similar resources without discovering new or potential user interest. Collaborative Filtering technologies as the most successful recommendation techniques have been implemented and applied in many practical systems. Although it can identify potential interest for the user, the traditional collaborative recommendation algorithm is the project's score by the user as a starting point.
The score does not fully reflect the interests of a person. Meanwhile, the personal interests and hobbies have the inextricably link with the occupation, age, education level and a series of factors, and similar attributes of people are also prone to similar interest. So, this paper proposed the collaborative recommendation algorithm based on the user context. According to user context, the algorithm clusters the users, allows each user to be able to accurately find the neighbors with their high similarity. In the same class, according to the user's history score and the dissimilarity between projects, it predicted score calculated for the target item to obtain the results of recommended target which users need.
This thesis completes the following work:
(1) After deep analyzing the problem of context loss, this thesis puts forward the user context and its formalization combined with Situation Semantics and the factors affecting the interests of users, and their natural attributes and social attributes. And further study for user context, this thesis puts forward the classification method of user context.
(2) The clustering method of static user context with multiple situational factors is given by the research of dissimilarity degree calculation and dissimilarity degree matrix. After that, this thesis puts forward the collaborative recommendation algorithm based on user context binding predictive value for target item method of Slope One algorithm.
(3) The collaborative recommendation algorithm based on user context is compared with traditional collaborative recommendation algorithm based on items and Slope One, using Matlab to carry out simulation experiment in MovieLens. Experimental results show that the algorithm is better in the mean absolute error against traditional algorithm and the Slope One algorithm. So, the feasibility and validity of this algorithm are verified.
(4) In combination with the intelligent educational platform project of Higher Education Press based on ontology, semantic and pragmatic, the article designs a scenario based on the user's personalized recommendation algorithm for collaborative recommendation system model as an example of the algorithm in practice application form.
引文
[1]曾春,邢春晓,周立柱.个性化服务技术综述[J].软件学报, 2002/13(10): 1952-1961.
[2]欧阳为民,郑诚,蔡庆生.数据库中加权关联规则的发现[J].软件学报.2001,vol.12 No.4:612-619.
[3] Mladenic, D. Machine learning for better Web browsing[C]. Menlo Park, CA: AAAI Press, 2000: 82-84.
[4] Quan Wen, Jianmin He. Personalized Recommendation Services Based on Service-Oriented Architecture[C]. Proceedings of the 2006 IEEE Asia-Pacific Conference on Services Computing, 2006:210-214.
[5] LANG Ken. Newsweeder: learning to filter netnews[C]. In: Proceedings of the 12th International Conference on Machine Learning.1995: 331-339.
[6] Konstan, J., Miller, B., Maltz, D., et al. GroupLens: applying collaborative filtering to usenet news[J]. Communications of the ACM, 1997,40(3): 77-87.
[7] Hill W, Stead L, Rosenstein M, Fumas G. Recommending and evaluating choices in a virtual community of user[C]. In CHI 95: Conference proceedings on human factors in Computing Systems, Denver, CO, 1995:194-201.
[8] U. Shardanand and P. Maes. Social Information Filtering: Algorithms for Automating " Word of Mouth "[C]. Proc. Conf: Human Factors in Computer Systems, Denver, 1995: 210-217
[9]刘建国,周涛,汪秉宏.个性化推荐系统的研究进展自然科学进展[J].自然科学进展, 2009, 19(1): 1-15.
[10] Adomavicius, G., Tuzhilin, A. User profiling in personalization applications through rule discovery and validation[C]. Proceedings of the 5th International Conference on Data Mining and Knowledge Discovery. New York: ACM Press, 1999: 377-381.
[11]邓爱林.电子商务推荐系统关键技术研究[D].上海:复旦大学博士论文, 2003,5:25-40.
[12] V. Robles, P. Larranaga, E. Menasalvas. Improvement of Naive Bayes Collaborative Filtering Using Interval Estimation[C]. Proceedings of the IEEE/WIC International Conference on Web Intelligence. 2003:184-190.
[13] Wolf J.L., Aggarwal C.C., Wu K.. Horting hatches an egg: a new graph-theoretic approach to collaborative filtering[C] .Proceedings of the ACM International Conference on Knowledge Discovery and Data Mining. New York: ACM Press 1999:201-212.
[14] Rakesh Agrawal, Ramakrishnan Srikant. Fast Algorithms for mining Association Rules in Large Databases[C]. Proceedings of the Twentieth International Conference on Very LargeDatabases, Santiago, Chile, 1994:487-499.
[15] Sarwar B.M., Karypis G., Konstan J.A., Riedl J.T.. Application of dimensionality reduction in recommender system-A case study [DB/OL]. In: ACM WebKDD 2000 Workshop,2000:419-422.
[16]杨引霞,谢康林,朱扬勇,左子叶.电子商务网站推荐系统中关联规则推荐模型的实现[J].计算机工程, 2004,30(19):56-61.
[17]杨健兵.数据挖掘中关联规则的改进算法及其实现[J].微计算机信息, 2006, 7-3:195-197
[18]董泽坤,李辉,史忠植.多元时间序列中跨事务关联规则分析的高效处理算法[J].计算机科学,2004, 31(3):108-111.
[19]王猛,王景光,李延霞.相关性分析在关联规则挖掘中的应用[J].统计与信息论坛,2004,Vol.19 No.1:112-116.
[20] M.J. Zaki, S. Parthasarathy, W. Li. A localized algorithm for parallel association mining[C]. 9th Annual ACM Symposium on Parallel Algorithms and Architectures, Newport, Rhode Island, June 28-29, 1997:248-252.
[21] J. Han, J. Pei, Y. Yin. Mining frequent patterns without candidate generation[J]. Data Mining and Knowledge Discovery,2004,8:53-87.
[22]林鸿飞,杨元生.用户兴趣模型的表示和更新机制[J].计算机研究与发展,2002 39(7):838-842.
[23]李振星等.基于兴趣模型的WEB信息预测采集过滤方法[J].计算机工程与应用,2003,5:162-168.
[24]赵庆峰.个性化信息服务系统理想模型的构建[J].情报杂志,2006,2:243-245.
[25] Abbattista F., Degemmis M., Zambetta F.. Improving the Usability of an E-Commerce Web Site through Personalization[C]. Proceedings of the Workshop on Recommendation and Personalization in Ecommerce, 2002:232-236.
[26]邓爱林,朱扬勇,施伯乐.基于项目评分预测的协同过滤推荐算法[J].软件学报. 2003.Vo14 No,9:1024-1028.
[27]崔林,宋瀚涛,陆玉昌.基于语义相似性的资源协同过滤技术研究[J].北京理工大学学报. 2005,25(5):16-19.
[28]周军锋,汤显,郭景峰.一种优化的协同过滤推荐算法[J].计算机研究与发展, 2004,41(10):1842-1847.
[29] Cheng Kwok Wai. Learning User Similarity and Rating Style for Collaborative Recommendation[J].Information Retrieval, 2004,7:395-410.
[30]欧立奇,陈莉,马煜.协同过滤算法中新项目推荐方法的研究[J].微计算机信息,2005,11(21):268-271.
[31]邹崇理.情境语义学[J].哲学研究. 1996 No.7:62-69.
[32] L. Kaufman, P.J. Rousseeuw. Finding Groups in Data:An Introduction to Cluster Analysis[M]. Wiley-Interscience, 2005,3:200-218.
[33] Breese J.S., D. Heckerman. Empirical Analysis of Predictive Algorithms for Collaborative Filtering[C]. Proc of the 14th Conference on Uncertainty in Artificial in Artificial Intelligence, 1998:43-52.
[34]张云涛,龚玲.数据挖掘原理与技术[M].北京:电子工业出版社,2004:128-153.
[35] R. Agrawal, Imielinski T., Swami A.. Database mining: A performance perspective[J]. IEEE Transactions on knowledge and Data Engineering, 1993(5):914-925.
[36]赵明清,蒋昌俊,陶树平.基于等价相异度矩阵的聚类[J].计算机科学. 2004 Vol.31 No.7: 183-184.
[37] D. Lemire and A. Maclachlan. Slope One Predictors for Online Rating-Based Collaborative Filtering[C]. SIAM Data Mining, Newport Beach, California, 2005:234-239.
[38] S. Vucetic and Z. Obradovic. A Regression-based Approach for Scaling-up Personalized Recommender Systems in E-commerce[DB/OL]. In ACM WebKDD, 2000:122-126.
[39]董丽,邢春晓,王克宏.基于不同数据集的协作过滤算法评测[J].清华大学学报(自然科学版),2009,4(49):590-594.
[2]欧阳为民,郑诚,蔡庆生.数据库中加权关联规则的发现[J].软件学报.2001,vol.12 No.4:612-619.
[3] Mladenic, D. Machine learning for better Web browsing[C]. Menlo Park, CA: AAAI Press, 2000: 82-84.
[4] Quan Wen, Jianmin He. Personalized Recommendation Services Based on Service-Oriented Architecture[C]. Proceedings of the 2006 IEEE Asia-Pacific Conference on Services Computing, 2006:210-214.
[5] LANG Ken. Newsweeder: learning to filter netnews[C]. In: Proceedings of the 12th International Conference on Machine Learning.1995: 331-339.
[6] Konstan, J., Miller, B., Maltz, D., et al. GroupLens: applying collaborative filtering to usenet news[J]. Communications of the ACM, 1997,40(3): 77-87.
[7] Hill W, Stead L, Rosenstein M, Fumas G. Recommending and evaluating choices in a virtual community of user[C]. In CHI 95: Conference proceedings on human factors in Computing Systems, Denver, CO, 1995:194-201.
[8] U. Shardanand and P. Maes. Social Information Filtering: Algorithms for Automating " Word of Mouth "[C]. Proc. Conf: Human Factors in Computer Systems, Denver, 1995: 210-217
[9]刘建国,周涛,汪秉宏.个性化推荐系统的研究进展自然科学进展[J].自然科学进展, 2009, 19(1): 1-15.
[10] Adomavicius, G., Tuzhilin, A. User profiling in personalization applications through rule discovery and validation[C]. Proceedings of the 5th International Conference on Data Mining and Knowledge Discovery. New York: ACM Press, 1999: 377-381.
[11]邓爱林.电子商务推荐系统关键技术研究[D].上海:复旦大学博士论文, 2003,5:25-40.
[12] V. Robles, P. Larranaga, E. Menasalvas. Improvement of Naive Bayes Collaborative Filtering Using Interval Estimation[C]. Proceedings of the IEEE/WIC International Conference on Web Intelligence. 2003:184-190.
[13] Wolf J.L., Aggarwal C.C., Wu K.. Horting hatches an egg: a new graph-theoretic approach to collaborative filtering[C] .Proceedings of the ACM International Conference on Knowledge Discovery and Data Mining. New York: ACM Press 1999:201-212.
[14] Rakesh Agrawal, Ramakrishnan Srikant. Fast Algorithms for mining Association Rules in Large Databases[C]. Proceedings of the Twentieth International Conference on Very LargeDatabases, Santiago, Chile, 1994:487-499.
[15] Sarwar B.M., Karypis G., Konstan J.A., Riedl J.T.. Application of dimensionality reduction in recommender system-A case study [DB/OL]. In: ACM WebKDD 2000 Workshop,2000:419-422.
[16]杨引霞,谢康林,朱扬勇,左子叶.电子商务网站推荐系统中关联规则推荐模型的实现[J].计算机工程, 2004,30(19):56-61.
[17]杨健兵.数据挖掘中关联规则的改进算法及其实现[J].微计算机信息, 2006, 7-3:195-197
[18]董泽坤,李辉,史忠植.多元时间序列中跨事务关联规则分析的高效处理算法[J].计算机科学,2004, 31(3):108-111.
[19]王猛,王景光,李延霞.相关性分析在关联规则挖掘中的应用[J].统计与信息论坛,2004,Vol.19 No.1:112-116.
[20] M.J. Zaki, S. Parthasarathy, W. Li. A localized algorithm for parallel association mining[C]. 9th Annual ACM Symposium on Parallel Algorithms and Architectures, Newport, Rhode Island, June 28-29, 1997:248-252.
[21] J. Han, J. Pei, Y. Yin. Mining frequent patterns without candidate generation[J]. Data Mining and Knowledge Discovery,2004,8:53-87.
[22]林鸿飞,杨元生.用户兴趣模型的表示和更新机制[J].计算机研究与发展,2002 39(7):838-842.
[23]李振星等.基于兴趣模型的WEB信息预测采集过滤方法[J].计算机工程与应用,2003,5:162-168.
[24]赵庆峰.个性化信息服务系统理想模型的构建[J].情报杂志,2006,2:243-245.
[25] Abbattista F., Degemmis M., Zambetta F.. Improving the Usability of an E-Commerce Web Site through Personalization[C]. Proceedings of the Workshop on Recommendation and Personalization in Ecommerce, 2002:232-236.
[26]邓爱林,朱扬勇,施伯乐.基于项目评分预测的协同过滤推荐算法[J].软件学报. 2003.Vo14 No,9:1024-1028.
[27]崔林,宋瀚涛,陆玉昌.基于语义相似性的资源协同过滤技术研究[J].北京理工大学学报. 2005,25(5):16-19.
[28]周军锋,汤显,郭景峰.一种优化的协同过滤推荐算法[J].计算机研究与发展, 2004,41(10):1842-1847.
[29] Cheng Kwok Wai. Learning User Similarity and Rating Style for Collaborative Recommendation[J].Information Retrieval, 2004,7:395-410.
[30]欧立奇,陈莉,马煜.协同过滤算法中新项目推荐方法的研究[J].微计算机信息,2005,11(21):268-271.
[31]邹崇理.情境语义学[J].哲学研究. 1996 No.7:62-69.
[32] L. Kaufman, P.J. Rousseeuw. Finding Groups in Data:An Introduction to Cluster Analysis[M]. Wiley-Interscience, 2005,3:200-218.
[33] Breese J.S., D. Heckerman. Empirical Analysis of Predictive Algorithms for Collaborative Filtering[C]. Proc of the 14th Conference on Uncertainty in Artificial in Artificial Intelligence, 1998:43-52.
[34]张云涛,龚玲.数据挖掘原理与技术[M].北京:电子工业出版社,2004:128-153.
[35] R. Agrawal, Imielinski T., Swami A.. Database mining: A performance perspective[J]. IEEE Transactions on knowledge and Data Engineering, 1993(5):914-925.
[36]赵明清,蒋昌俊,陶树平.基于等价相异度矩阵的聚类[J].计算机科学. 2004 Vol.31 No.7: 183-184.
[37] D. Lemire and A. Maclachlan. Slope One Predictors for Online Rating-Based Collaborative Filtering[C]. SIAM Data Mining, Newport Beach, California, 2005:234-239.
[38] S. Vucetic and Z. Obradovic. A Regression-based Approach for Scaling-up Personalized Recommender Systems in E-commerce[DB/OL]. In ACM WebKDD, 2000:122-126.
[39]董丽,邢春晓,王克宏.基于不同数据集的协作过滤算法评测[J].清华大学学报(自然科学版),2009,4(49):590-594.