基于联合相似度的协同过滤推荐算法研究
发布时间：2018-07-24 20:55
【摘要】：随着互联网的快速发展，Web成为了人们获取信息的主要途径。然而由于电子商务的广泛普及，如何为用户提供有用的信息成为了一个研究热点。虽然搜索引擎的出现在一定程度上满足了人们对信息检索的需求，但是无法满足不同领域、不同层次的用户需求。因此，个性化推荐技术作为个性化服务的一种模式应信息检索的需求而产生，其本质是信息过滤。 推荐系统作为解决信息过载的重要工具，，为用户提供如电影、音乐、书籍及新闻等方面的个性化推荐。在过去的十年里，研究者致力于各种推荐技术的探究并将其应用到实际系统中。协同过滤推荐是目前最为经典并为广泛应用的推荐技术，它根据目标用户的偏好以及与该用户具有相似偏好的用户的项目评价，向目标用户进行新项目的推荐或评分预测。然而，协同过滤技术存在冷启动和数据稀疏等问题。 基于联合相似度的协同过滤算法，将社会网络分析的方法引入到协同过滤推荐系统中。利用用户-项目二部图、用户-用户单部图以及基于相同浏览行为模式的行为网络图分别生成相似度矩阵，然后依据相似度矩阵的密度来确定其在联合相似度中的权重，最终生成联合相似度。最后在豆瓣数据集上将此算法与现存的一些评分预测及推荐算法进行了对比试验，试验结果表明基于联合相似度的协同过滤算法在评分预测及推荐结果上更加精确。
[Abstract]:With the rapid development of the Internet, Web has become the main way for people to obtain information. However, due to the widespread popularity of electronic commerce, how to provide useful information for users has become a research hotspot. Although the emergence of search engines to some extent meet the needs of information retrieval, but can not meet the different fields, different levels of user needs. Therefore, personalized recommendation technology, as a mode of personalized service, comes into being according to the requirement of information retrieval, and its essence is information filtering. As an important tool to solve information overload, recommendation system provides personalized recommendation for users such as movies, music, books and news. Over the past decade, researchers have devoted themselves to the exploration of various recommended technologies and their application to practical systems. Collaborative filtering recommendation is the most classical and widely used recommendation technology at present. According to the preference of the target user and the item evaluation of the user with similar preference, the collaborative filtering recommendation can recommend or predict the new item to the target user. However, there are some problems in collaborative filtering technology, such as cold start and data sparsity. Based on the collaborative filtering algorithm of joint similarity, the social network analysis method is introduced into collaborative filtering recommendation system. The similarity matrix is generated by the user-item bipartite graph, the user-user single-part graph and the behavior network graph based on the same browsing behavior pattern, and their weights in the joint similarity are determined according to the density of the similarity matrix Finally, the joint similarity is generated. Finally, the algorithm is compared with some existing score prediction and recommendation algorithms on the soybean valve dataset. The experimental results show that the joint similarity based collaborative filtering algorithm is more accurate in score prediction and recommendation results.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3
本文编号：2142629
[Abstract]:With the rapid development of the Internet, Web has become the main way for people to obtain information. However, due to the widespread popularity of electronic commerce, how to provide useful information for users has become a research hotspot. Although the emergence of search engines to some extent meet the needs of information retrieval, but can not meet the different fields, different levels of user needs. Therefore, personalized recommendation technology, as a mode of personalized service, comes into being according to the requirement of information retrieval, and its essence is information filtering. As an important tool to solve information overload, recommendation system provides personalized recommendation for users such as movies, music, books and news. Over the past decade, researchers have devoted themselves to the exploration of various recommended technologies and their application to practical systems. Collaborative filtering recommendation is the most classical and widely used recommendation technology at present. According to the preference of the target user and the item evaluation of the user with similar preference, the collaborative filtering recommendation can recommend or predict the new item to the target user. However, there are some problems in collaborative filtering technology, such as cold start and data sparsity. Based on the collaborative filtering algorithm of joint similarity, the social network analysis method is introduced into collaborative filtering recommendation system. The similarity matrix is generated by the user-item bipartite graph, the user-user single-part graph and the behavior network graph based on the same browsing behavior pattern, and their weights in the joint similarity are determined according to the density of the similarity matrix Finally, the joint similarity is generated. Finally, the algorithm is compared with some existing score prediction and recommendation algorithms on the soybean valve dataset. The experimental results show that the joint similarity based collaborative filtering algorithm is more accurate in score prediction and recommendation results.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3
【参考文献】
相关期刊论文 前2条
1 许海玲;吴潇;李晓东;阎保平;;互联网推荐系统比较研究[J];软件学报;2009年02期
2 刘鲁;任晓丽;;推荐系统研究进展及展望[J];信息系统学报;2008年01期
相关博士学位论文 前1条
1 李聪;电子商务推荐系统中协同过滤瓶颈问题研究[D];合肥工业大学;2009年
本文编号：2142629
本文链接：https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/2142629.html
