【作者】 金亚亚；

【导师】 牟援朝；

【作者基本信息】 华东理工大学 ， 管理科学与工程， 2011， 硕士

【摘要】 随着电子商务的发展,它的功能日益强大,个性化产品推荐作为其重要的一个组成部分,也为推动电子商务的发展起到了重要的作用。其中,最重要的一种推荐算法就是协同过滤算法,各个大的电子商务网站都或多或少地应用了该算法,然而在实际应用中,该推荐算法也有其不足之处,不能很好地为企业和用户服务,故很多学者提出了各种相应的改进算法。本文首先分析了协同过滤算法的不足,提出了基于改进信任度的协同过滤算法,并指明该算法的应用。文章构建评价算法的框架：先根据个性化产品推荐过程,构建概念模型；在此基础上,首次尝试从公司利益最大化的角度确定公司的主要目标和任务,数据收集模块和推荐结果列示模块通过比较分析现有的评价推荐结果指标,选取评价指标,并根据这些指标对主要目标和任务的实现所做的贡献,运用层次分析法,确定贡献率,最后得出综合评价结果；在推荐算法运行模块,应用敏感性分析法分析相似度计算对推荐结果的影响,进一步地改进算法。对传统的协同过滤推荐算法和基于改进信任度的协同过滤推荐算法的推荐结果进行显著性分析评价,验证本文提出的改进算法能够显著的提高推荐的准确性。更多还原

【Abstract】 With the development of electronic commerce, it becomes more powerful everyday, personalized products recommendation as an important part, which has played an important role on promoting the development of e-commerce. One of the most important recommendation algorithms is the collaborative filtering algorithm (CF), which is applied to many e-commerce sites more or less. But in practical application, this recommendation algorithm also has its drawbacks. It can not give good service to consumers and companies, therefore, many scholars have proposed various improved algorithms.This paper analyzes the shortcomings of the use-based collaborative filtering algorithms, and proposes a new algorithm based on modified trustworthiness, and specified the application of this algorithm. Then we construct a framework for evaluation of algorithms: first, proposes concept models based on personalized products recommendation process; Secondly, being attempt to stand in the perspective of maximizing interests of the company to determine the company’s main objectives and tasks, select the evaluating indicators in data collection module and recommended lists module by comparing existing indicators of the evaluation recommendations, according to the contribution of these indicators to achieve the main objectives and tasks, using AHP to ascertain the contribution rate and draw the conclusion of the evaluation results; in the running recommendation algorithm module, utilizing sensitivity analysis to analyze whether similarity calculation will affect the recommended results and how to affect, then to improve the algorithm further.Design experiment to prove whether the improved trustworthiness in the collaborative filtering. recommendation algorithms can get significant improvement the results. According to experiment results, the improved algorithm can significantly get better recommendation accuracy.更多还原