基于机器学习的电影推荐系统设计与实现.docx

摘要 随着互联网的出现和普及，网络上充斥着大量的信息，用户在互联网上可选择的余地大大增加，这也造成了用户在选择时很难获得自己真正想要的那部分信息，那些原本有用的大量信息却成了用户获取自身所关心的信息的阻碍,造成对信息的使用率反而降低了。 在这种情况下，个性化推荐系统应运而生，系统根据用户的个性化行为信息，针对用户的行为信息的差异进行差别化推荐。和传统搜索引擎不同的是，个性化推荐系统能够自动挖掘用户的特征，而无需用户输入任何额外信息。 本文以电影信息为基础，采用机器学习的方式实现了一个个性化电影推荐系统。本系统可以根据用户之前所看过的电影以及对电影所评价的分数，给用户推荐用户可能会感兴趣但还没有看过的电影。本系统实现了基于用户和基于物品的协同过滤算法，并借鉴了亚马逊、豆瓣的推荐算法，增加相关特征，改进原有算法，提高推荐的准确性。并通过用户标签信息、时间上下文信息、地点上下文信息社交网络信息，将理论和实践紧密结合，提高推荐算法在真实网络环境下的易用性。 关键词 推荐系统；机器学习；协同过滤算法; Abstract The network is inundated with a lot of information as the emergence and popularizationof Internet, which has greatly enlarged the selectable scope of information in the internet forusers, and also caused the difficulty of users to choose the information they really want.Hence, these originally useful information of large quantity hinder users from gaining theirconcerned information, decreasing the utilization rate of information to some extent. Under the circumstances, Personalized Recommender System was born at the rightmoment, giving differential recommendation for different users in accordance with userspersonalized information. Unlike the searching engine, Personalized Recommender System isable to automatically mine users feature without any information input from users. Based on movie information, this thesis realized a Movie Recommender System byadopting machine learning method. In accordance with movies that users have watched andusers evaluation scores on movies, this system can recommend users movies which are ofusers interests and have not yet been watched. In addition, this system also realizedcollaborative filtering algorithm based on both users and goods, as well as optimized theoriginal algorithm to increase the accuracy of recommendation through drawing lessons fromthe recommender algorithm of Amazon and Douban. Meanwhile, through the analysis onusers label information, time context information, place context information and socialnetwork information, this system closely in