联系我们

论文查询结果	返回搜索

论文编号：	2319
作者编号：	2120082062
上传时间：	2010/6/8 19:58:28
中文题目：	协同过滤算法及其在电子商务推荐系统中的应用研究
英文题目：	Study on Collaborative Filtering Algorithm and It''s Application in E-commerce Recommender System
指导老师：	石 鉴
中文关键字：	推荐系统 协同过滤 电子商务
英文关键字：	Collaborative Filtering Recommender System E-commerce
中文摘要：	近年来，在信息技术的推动下，电子商务在全球范围内取得了迅猛的发展，特别是中国，在2009年全球经济一片萎靡之时，国内网购规模却逆势增长，网络购物市场交易规模达2483.5亿元，专家预测，该数字在2013年有望突破1万亿元。但是，飞速发展的电子商务在不断满足人们在线购物需求的同时，同时将人们推进了信息的海洋，因此，电子商务平台迫切需要一种自动化的购物帮手帮助人们降低购物成本，而推荐系统(recommender system)无疑是最佳的选择。协同过滤算法由于推荐的准确性和结果的易于解释性，在目前的电子商务推荐系统中得到了广泛的采用。但是仍面临着一些突出的问题，对于这些问题的研究和解决，将具有非常重要的理论意义和现实意义。 本文首先就目前电子商务推荐系统主要涉及技术、所主要采用推荐技术和推荐效果评价指标进行梳理总结，其次对协同过滤算法进行分类整理，并就协同过滤算法目前存在的问题和现存的解决办法进行梳理。在上述基础上，就协同过滤算法和电子商务推荐系统的框架模型进行了深入的探讨和研究。 在协同过滤算法方面，本文针对传统协同过滤算法搜索最近邻策略中存在的可能不足，提出了基于递归的预测算法，通过将没有进行对对应项进行评分却具有高相似度的用户纳入计算，有效地拓展了搜索范围，弥补了传统算法的不足，并在一定程度上改善了算法的推荐预测准确度。 在电子商务推荐系统的框架模型方面，结合有关协同过滤算法的研究结果，提出了一种改进的综合性框架模型，该模型对推荐的实时性、算法的预测准确性以及整体的灵活性、可配置性进行了重点关注，并有效整合了基于递归的协同过滤改进算法。 在文章最后，我们通过严格的实验设计，基于特定数据集，对改进算法的预测准确度和时间复杂性进行了测度，实验结果表明，和传统协同过滤算法相比，基于递归的协同过滤算法能显著提高推荐的准确度。
英文摘要：	Driven by information technology, e-commerce has made rapid development on a global scale, particularly in China. in the shadow of 2009’s economic crisis, China's net purchases size of online shopping market transaction bucked the growth trend, reached 248.35 billion Yuan , and is expected exceeding to 1 trillion in 2013. However, while the rapid development of e-commerce continues to meet people's needs shopping online, it also puts people into the information ocean. Therefore, e-commerce platforms need an automated shopping assistant to help people reduce the cost of shopping. Recommender system Is undoubtedly the best option. Due to the accuracy of recommendation and results easy-explanatory, Collaborative filtering algorithms has been widely used in the current e-commerce recommender system. But it still faces a number of outstanding issues. it will have very important theoretical and practical significance with these issues resolved. In this paper, we first sum and sort the technical aspects of electronic commerce recommendation system, the main recommendation technology、 systems evaluation method and collaborative filtering algorithm, and then present the current problems and present solutions to the comb. Based on the above job, we made a in-depth exploration and study On the collaborative filtering algorithms and e-commerce recommendation system framework model. In the study of collaborative filtering algorithms, on the base of nearest neighbor search strategy in the traditional collaborative filtering algorithms, we proposed a recursive prediction algorithm, through taking the users having high similarity with active user into account, Effectively compensate the deficiencies of the traditional algorithms and expand the search range, and to some extent, improve the prediction accuracy of the traditional algorithms. In the framework model of recommender system in e-commerce, combined with collaborative filtering algorithms for the research, we presented a comprehensive framework model, The model focuses on real-time recommendation, overall accuracy and flexibility of the algorithm, Effectively integrated recursive collaborative filtering Algorithm. Finally, based on specific data sets, we have adopted a strict experimental design, to measure the prediction accuracy and time complexity of the algorithm. Experimental results showed that, compared to traditional collaborative filtering algorithms, recursive collaborative filtering algorithm can significantly improve the accuracy of recommendation.
查看全文：	预览  下载(下载需要进行登录)