《基于协同过滤算法的电影推荐应用研究》由会员分享，可在线阅读，更多相关《基于协同过滤算法的电影推荐应用研究（53页珍藏版）》请在金锄头文库上搜索。
1、 HEBEI UNIVERSITY 密密 级：级： 分分 类类 号：号： 学校代码：学校代码：10075 学学 号：号：43011311001 硕士学位论文 硕士学位论文 基于协同过滤算法的电影推荐应用研究 学位申请人：闫 燕 指 导 教 师 ：王 煜 教授 企 业 导 师 ：李燕生 高级讲师 学 位 类 别 ：工程硕士 授 予 单 位 ：河北大学 完 成 日 期 ：二一四年五月 Classified Index： CODE：10075 U.D.C： NO： 43011311001 A Dissertation for the Degree of Master Movie Recommenda
2、tion Application Study Based on Collaborative Filtering Algorithm Candidate: Yan Yan Supervisor: Prof. Wang Yu Advisor in Enterprise: SL. Li Yansheng Academic Degree Applied for: Master of Engineering University: Hebei University Date of Accomplishment: May，2014 摘 要 I 摘 要 随着网络技术的迅猛发展，个性化推荐通过挖掘每个用户潜在
3、的需求，在电子商务中发挥了越来越重要的作用。协同过滤推荐方法是目前比较流行的个性化推荐方法之一， 同时它也面临着一些有待进一步解决的问题，如可扩展问题、稀疏性问题、冷启动问题等。本文对基于协同过滤算法的个性化推荐进行了研究。 本文分析了用户对不同类别电影的喜好程度，在进行slope-one评分填充时，充分考虑了项目类别信息。以相同类别电影的评分为依据来寻找相似用户，进行评分预测，有效地解决了用户对不同类别电影的喜好不同的问题。 协同过滤推荐搜索目标用户的最近邻需要在整个用户空间上进行， 随着数据的不断增多和系统规模的不断扩大，搜索目标用户的最近邻的计算量也不断增大，系统的实时性能越来越差，这就
4、成为协同过滤系统发展的一个瓶颈。针对该问题，本文采用了一种基于聚类的协同过滤算法。根据用户的属性特征和用户评分数据对用户进行聚类，将具有相似兴趣的用户聚成一类。 当新来一个目标用户时， 首先判断该目标用户所属的聚类，然后在相应的聚类内搜索目标用户的最近邻， 从而使搜索目标用户的最近邻的计算空间尽可能的缩小，达到了缩短寻找最近邻的计算时间的目的。 本文也考虑了对新项目和新用户的推荐。 依据电影的类别属性和评分数据对当前数据库中的电影进行聚类，使得类别和评分相似的电影聚为一类。对于新电影，则依据该电影的基本特征判断该电影所属的聚类，在所属聚类内，依据协同过滤算法预测目标用户对新电影的评分，生成推荐
5、列表；对于新用户，则依据用户基本特征判断该用户所属的聚类，在所属聚类内，寻找该用户的最近邻，利用协同过滤算法预测该用户对目标项目的评分，生成推荐列表。 为了验证本文给出的算法的有效性，本文采用MovieLens网站上公开的数据集进行了测试，并对实验结果进行了详细的分析。 关键词 个性化推荐 协同过滤 聚类 最近邻 平均绝对偏差 Abstract II Abstract With the rapid development of e-commerce, personalized recommendation used for the exploration of every customers
6、potential demands plays a more and more important role. Collaborative filtering recommendation method is one of the most popular personalized recommendation methods. There are also some bottlenecks to the application of this method, such as scalability, sparseness and cold boot. In this paper, a res
7、earch on adopting collaborative filtering algorithm was made to carry out personalized recommendation. In this paper, users preference for various movies, which varies in degree, is analyzed. On giving the ratings with slope-one, the item category data is taken into full account. The ratings of movi
8、es in the same category are used to find similar users and to predict the ratings, which effectively solves the problem that userspreference differs for movies in different categories. The collaborative filtering recommendation system has to search for the nearest neighbor of the target user in the
9、entire user space. As the data constantly increases and the scale of the system continuously expands, the amount of computation tends to increase linearly for searching the nearest neighbor of the target users in the whole user space, which will result in worse real-time performance of the system. T
10、his has become the bottleneck for the development of the collaborative filtering system. Aiming at this problem, a collaborative filtering algorithm based on clustering is adopted in this paper. Users with similar interests are categorized into one group based on the users attributive characters and
11、 users rating data. As soon as a target user arrives, he or she should be put into the appropriate cluster. Then, the nearest neighbor of the target user is searched in the corresponding cluster, to look for the nearest neighbor of the target user in a smaller user space and to shorten the computati
12、on time spent in searching the nearest neighbor. In this paper, the recommendation of new items and new users is also taken into Abstract III consideration: the movies in the existing database are clustered according to category attribute and ratings, to gather the movies similar in category and rat
13、ings into a cluster. With regard to new movies, a new movie is classified into the cluster it belongs to, according to its basic features. In the cluster, on the basis of the collaborative filtering algorithm, the ratings the target users may give to the new movie are predicted and then a recommenda
14、tion list is generated. For new users, they are classified into the right cluster based on the basic characteristics of such users. And the users nearest neighbor is searched in the cluster. By using the collaborative filtering algorithm, the rating that the user may give to the target item is predi
15、cted, and a recommendation list is generated. In order to verify the validity of the algorithm stated that this paper adopted, a test was carried out by using data released on MovieLens, and a detailed analysis was made on the test results. Keywords Personalized Recommendation Collaborative Filtering Clustering Nearest neighbor Mean Absolute Error 目 录 IV 目 录 第 1 章 绪论.1 1.1 研究背景和意义.1 1.2 国内外研究现状.
