云环境下基于社交信息的音乐推荐系统设计与实现

发布时间：2018-08-28 13:48

【摘要】：随着互联网的不断发展和移动互联网的兴起,越来越多的人选择通过互联网来随时随地享受数字化音乐带来的服务。数字音乐数量的激增使得音乐服务提供商的主要竞争从曲库的深度和规模转移到了推荐和发现音乐方面,推荐系统成为解决该问题的主要技术手段,而协同过滤推荐算法作为推荐领域最主流的算法之一,在音乐推荐系统中得到了广泛应用。然而,随着推荐准确率的不断提高,影响协同过滤推荐算法推荐效果的另一个问题越来越突显出来:如何发现相关度高的新颖推荐项。本文从上述问题出发,提出了融合社交信息的基于图的协同过滤改进算法,并以该算法为核心技术,设计开发了一套完整的音乐推荐系统。主要工作如下:首先,改进算法的主要思想是:利用用户的社交信息,对由项目相似性矩阵构建出的用户偏好图进行扩充,以降低通过信息熵计算得出的奇异推荐项的比例,然后将这些项目与通过经典协同过滤算法得到的推荐项合并在一起作为最终的推荐结果。最后,通过采集自Last.fm上的数据对算法的有效性进行了验证。结果表明,与原始算法相比,该改进算法的推荐准确率平均提高了约2.265%,由此损失的新颖性在相关指标下仅仅约为1.24%。由此说明该算法可以在发掘出新颖推荐项的同时,提升系统的准确率,从而达到更好的推荐效果。其次,基于上述算法,本文设计并实现了一套音乐推荐系统,在进行了充分的需求分析和系统架构设计的基础上,给出了单曲推荐、艺术家推荐和好友推荐的算法设计,并提出了歌单推荐的策略;然后通过MapReduce编程范式实现了各个算法并将系统运行在Hadoop云平台上;最后,邀请用户对系统进行了在线测试,当推荐数为25时平均新颖度为4.56,准确率约为17.6%,证明该音乐推荐系统在兼顾推荐新颖性和准确率方面具有出色表现。
[Abstract]:With the continuous development of the Internet and the rise of mobile Internet, more and more people choose to enjoy the services brought by digital music anytime and anywhere through the Internet. The surge in the number of digital music has shifted the main competition of music service providers from the depth and scale of music library to the aspect of recommending and discovering music. Recommendation system has become the main technical means to solve this problem. As one of the most popular algorithms in the field of recommendation, collaborative filtering recommendation algorithm has been widely used in music recommendation system. However, with the improvement of recommendation accuracy, another problem that affects the recommendation effect of collaborative filtering is becoming more and more prominent: how to find novel recommendation items with high correlation. Based on the above problems, this paper puts forward an improved algorithm of collaborative filtering based on graph, which integrates social information, and designs a complete music recommendation system based on this algorithm. The main work is as follows: firstly, the main idea of the improved algorithm is to extend the user preference map constructed from the item similarity matrix by using the social information of the user, so as to reduce the proportion of singular recommendation items calculated by the information entropy. Then these items are combined with the recommended items obtained by the classical collaborative filtering algorithm as the final recommendation results. Finally, the validity of the algorithm is verified by collecting data from Last.fm. The results show that compared with the original algorithm, the recommendation accuracy of the improved algorithm is increased by about 2.2655.The resulting novelty is only about 1.24 under the related index. It shows that the algorithm can improve the accuracy of the system and achieve a better recommendation effect. Secondly, based on the above algorithm, this paper designs and implements a set of music recommendation system. On the basis of sufficient requirement analysis and system architecture design, the algorithm design of single song recommendation, artist recommendation and friend recommendation is given. Then we implement each algorithm by MapReduce programming paradigm and run the system on the Hadoop cloud platform. Finally, we invite users to test the system online. When the recommendation number is 25, the average novelty is 4.56 and the accuracy is about 17.6. it is proved that the music recommendation system has excellent performance in both novelty and accuracy.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】