社会化商务环境下协同过滤推荐方法研究

发布时间：2018-06-14 11:29

本文选题：社会化商务 + 个性化推荐　；参考：《华南理工大学》2016年硕士论文

【摘要】：Web2.0 时代下,社交化应用的出现使互联网生活方式产生了巨大的变革,其商业潜力在电商企业的努力探索下正不断地释放,社会化商务成为了电子商务发展的新方向。社会化商务环境中的信任关系深刻地影响着消费者的购买决策,成为了支撑网络商务活动开展的重要因素。同时,该环境下丰富的数据资源也为研究信任关系提供了数据基础。因此,为了提高传统推荐方法的推荐准确性,本文将充分地挖掘社交和项目评分数据,研究用户之间的信任关系,并寻找合理的方法将信任融入到推荐方法中。以社交数据为基础,本文提出了基于社会化关系与信任传播的协同过滤推荐方法。首先,为了使社交关注数据能更加真实可信地表达信任关系,本文将用户社交属性数据计算了用户可信评分,并将其对社交关注数据进行了可信量化。其次,本文充分挖掘了可信量化后的社交关注数据,结合同引分析方法提出了一种同引信任关系。结合直接、间接和同引信任后的综合信任关系有助于相似度计算过程中准确地寻找信任邻居。最后,针对评分较少的用户之间利用传统方法计算相似度不准确的问题,本文提出了基于信任传播的相似度计算方法。在Epinions和大众点评数据集上的实验结果表明:本文研究的社会化信任关系和相似性计算方法具有一定的合理有效性。与相关推荐方法相比,本文研究方法在MAE等评价指标上皆具出色的表现,使推荐效果得到了进一步提高。以项目评分数据为基础,通过充分利用项目评分数据对隐性信任关系进行挖掘,本文提出了基于项目评分与信任挖掘的协同过滤推荐方法。通常对隐性信任的构建考虑的用户行为特征较少,而且大部分未考虑到信任传递特性。另外,已有研究大多数都局限于基于用户的隐性信任推荐,缺少了从用户对项目的隐性信任角度进行研究。因此,本文将充分地考虑用户在项目评分上的特征以及信任的弱传递特性,从基于用户和基于项目两种角度挖掘用户之间的隐性信任关系,并将两者融合形成综合隐性信任推荐方法。在Movielens和大众点评数据集上的实验结果表明:本文研究的隐性信任关系相对具有合理有效性。相较于传统推荐方法,本文研究方法在MAE等评价指标上表现更为出色,能更加准确地为用户推荐兴趣相符的项目。
[Abstract]:In the era of Web 2.0, the emergence of social application has brought about a great change in the way of life on the Internet, and its commercial potential is being continuously released under the efforts of e-commerce enterprises, and socialized commerce has become a new direction of the development of electronic commerce. The trust relationship in the social business environment has a profound impact on consumers' purchase decisions and has become an important factor supporting the development of online business activities. At the same time, the rich data resources in this environment also provide the data basis for the study of trust relationship. Therefore, in order to improve the accuracy of traditional recommendation methods, this paper will fully mine social and item scoring data, study the trust relationship between users, and find a reasonable way to integrate trust into recommendation methods. Based on social data, this paper proposes a collaborative filtering recommendation method based on social relationship and trust propagation. Firstly, in order to make the social concern data express the trust relationship more truthfully, this paper calculates the user trust score and quantifies the social concern data. Secondly, this paper fully excavates the social concern data after trusted quantization, and proposes a kind of co-citation trust relationship combined with co-citation analysis method. The combination of direct, indirect and cocitation trust relationship can help to find the trust neighbor accurately in the process of similarity calculation. Finally, aiming at the problem that the traditional method is not accurate in calculating similarity between users with less score, this paper proposes a similarity calculation method based on trust propagation. The experimental results on Epinions and Dianping datasets show that the socialized trust relationships and similarity calculation methods studied in this paper are reasonable and effective. Compared with the related recommendation methods, the research methods in this paper have excellent performance on the evaluation indexes such as mae, which makes the recommendation effect further improved. Based on item scoring data, this paper proposes a collaborative filtering recommendation method based on item score and trust mining. In general, implicit trust construction takes less user behavior characteristics into account, and most of them do not consider trust transfer characteristics. In addition, most of the previous studies are limited to the implicit trust recommendation based on users, and lack of research from the point of view of users' implicit trust to the project. Therefore, this paper will fully consider the characteristics of users in item scoring and the weak transfer of trust, mining the implicit trust relationship between users from the perspective of users and project-based. And the combination of the two to form a comprehensive recessive trust recommendation method. The experimental results on Movielens and Dianping datasets show that the implicit trust relationship studied in this paper is relatively reasonable and effective. Compared with the traditional recommendation method, the research method in this paper is more excellent in the evaluation index of mae, and can more accurately recommend items of interest to users.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F724.6

【相似文献】