【作者】 王桂芬；

【导师】 任燕； 段隆振；

【作者基本信息】 南昌大学 ， 计算机应用技术， 2012， 硕士

【摘要】 随着网络技术的发展和不断完善,电子商务已经覆盖了整个营销网络,成为一种新的商务通信通道,为了使用户得到更好的购买体验,个性化的推荐系统渐渐兴起。个性化推荐系统对网站的历史记录进行分析,挖掘用户兴趣,预测并推荐用户可能感兴趣的商品。协同过滤技术是个性化推荐系统中较常用的推荐技术,但是随着个性化推荐系统的发展及应用,该技术存在的稀疏性、冷启动、可扩展性等问题日益成为推荐的瓶颈。经过深入的研究和分析,本文对传统的协同过滤算法进行了改进。改进的方法是首先在概念分层技术的基础上,采用项目评分度等概念,对用户-项目评分矩阵进行缩减,构建用户-项目评分候选集,提高推荐算法的可扩展性；其次引入“项目客观特征偏向”和“兴趣偏向度”概念,并对稀疏的用户-项目评分候选集进行更加合理的填充,以缓解数据稀疏性和冷启动问题带来的影响；最后在预测项目评分时,使用用户对一类项目的评分平均值来度量用户的评分习惯,提出基于项目类评分尺度的推荐公式,以提高预测准确性。经过实验及各评价规则上的测试分析,改进的协同过滤算法较传统协同过滤算法有更佳的推荐效果。最后本文选择了电子商务个性化推荐系统的典型案例——图书推荐系统进行开发和实现,将改进的协同过滤算法应用到图书推荐服务中,进一步核准了此算法的实用性。更多还原

【Abstract】 Along with the development and perfection of network technology, E-Commerce has covered the whole marketing network and become a new channel in business communication. For better user experience, Personalized Recommendation Systems gradually rise. It is used to analysis user’s historical records, mine user’s interest; provide prediction and recommendation to the target user. Collaborative Filtering algorithm is one of the common recommendation technologies, but with the development and application of the Personalized Recommendation System, these problems, such as sparsity, cold start, expansibility etc, have become bottlenecks in recommendation.Through the in-depth research and analysis, this paper presents the improved algorithm based on the traditional Cooperative Filtering. The method is that firstly on basis of Concept Hierarchy, adopt the concept of cluster rated degree, reduce user-item score matrix, get the smaller candidate matrix to improve recommendation scalability; Secondly, these concepts "Item Objective Character" and "Interest Deviation Degree" are introduced to fill the sparse candidate matrix more reasonably, alleviate the impact of data sparsity and cold start; Finally, in prediction, the average of rating to itemset is used to measure user’s rating habit and a new recommend formula based on itemset’s rating is put forward to improve the prediction accuracy.Through experiments and the test on evaluation rules, the improved algorithm is proved to be better than the traditional collaborative filtering algorithm in recommendation. Finally, realize a book recommendation system as the typical case of the Recommendation Systems for E-Commerce, the improved cooperative filtering algorithm is used in the service of book recommendation, and the practicality of improved algorithm is approved.更多还原