数学与统计学院知识库

随着社会的发展、科技的进步, 在日常生活中人们与互联网互动的频率越来越高, 通过互联网站或移动电子设备来观看视频、MV、歌曲及其他的娱乐节目越来越成为人们打发闲暇时间的一种普遍方式. 与此同时为了迎合受众的喜好, 各类电影、视频等项目在互联网平台上多如牛毛般的呈现在观众面前, 逐渐的人们从以前获得娱乐信息困难到现在信息爆炸式增长让人应接不暇的这种状态被我们称之为信息过载. 在现如今信息过载的时代, 对于消费者而言如何在充斥着其他非必需的繁多的信息中找到自己真正需要的, 而不被其他的信息所干扰和误导, 对于信息的生产者而言如何在众多的信息中崭露头角, 能够受到普罗大众的重视, 都是现在所面临的挑战. 推荐系统就是解决上述信息过载问题的一种有效措施. 推荐系统的基本任务是连接物品与用户, 解决信息过载的问题, 通过不同的推荐算法用特定的方式将用户与物品相联系, 从而达到针对不同的用户, 根据其兴趣为其提供个性化的推荐.
本文旨在对不同的推荐算法进行改进并通过加权融合得到新的融合模型来提高推荐算法的准确度与多样性, 从而达到更加良好的推荐结果. 本文的主要工作如下：提出了适合电影推荐的融合推荐模型, 在形成融合模型的子算法上提出算法的改进, 从而提升电影产品推荐算法的多样性与准确度. （1）针对推荐系统的基于用户的协同过滤算法, 在此基础上添加分析用户观看电影时间的上下文信息, 从而更精准的得到与预测用户相似度更高的邻居用户, 从而提升算法的准确度. （2）在基于项目的协同过滤算法中, 通过惩罚活跃用户的不当贡献来提高算法的准确度. （3）基于图的协同过滤算法, 通过改变不同权重来重新构建基于图的推荐算法, 从而提升算法的准确度. 通过离线实验计算各种算法改进前后的准确度, 提出了基于以上算法的融合推荐模型, 并通过离线实验证明, 融合推荐模型在准确度、多样性等方面有了很好的提升.

With the development of society and the progress of science and technology, the proportion of people interacting with the Internet in daily life is increasing. Watching videos, MVs, listening songs, and other entertainment programs through Internet sites or mobile electronic devices are becoming common ways for the general public to spend their leisure time. At the same time, in order to cater to popular taster , various kinds of films, videos, and other items that are presented in front of the audience in various major Internet platforms. Gradually, people have changed from the lack of information and entertainment to the explosion of information. This state is what we call information overload. In the current era of information overload, how do consumers find what they really need in the flood of other non-essential information, without being disturbed and misled by other information. how do producer show up prominently and be valued by the general public, are now challenges for us. The recommendation system is to give an effective solution to the above problem of information overload. The basic task of the recommendation system is to connect items and users, solve the problem of information overload, and use different recommendation algorithms to associate users with items in a specific way so as to target different users and provide personalized services according to their interests recommend.
In order to improve the accuracy and diversity of the advancing algorithm and achieve better recommendation results, this paper improve the different algorithms and obtained a new weighted fusion model. The main work of this dissertation is as follows: A fusion recommendation model for film recommendation is proposed, and the improvement of the algorithm is proposed on the sub-algorithm that forms the fusion model, so as to improve the diversity and accuracy of the recommendation algorithm of the film product. (1) Based on the user-based collaborative filtering algorithm of the recommendation system, the time context information for viewing the movie by the user is added and analyzed, so that the neighbor user with higher similarity with the predicted user is more accurately obtained, thereby improving the accuracy of the algorithm. (2) based on the item-based collaborative filtering of articles, the accuracy of the algorithm is improved by penalizing the improper contribution of active users at the same time. (3) Improve the graph-based collaborative filtering algorithm, and reconstruct a graph-based collaborative filtering algorithm by changing different weights, so as to improve the accuracy of the algorithm. The off-line experiment was used to calculate the accuracy of various algorithms before and after improvement, and a fusion recommendation model based on the above algorithm was proposed. Off-line experiments proved that the fusion recommendation model had better improvements in accuracy and diversity.