基于用户偏好特征的微博转发预测

Micro‑blog forwarding prediction based on user preference characteristics

在线社交网络为信息的传播提供了渠道,但同时也加快了不良信息的传播速度。针对真实场景下新浪微博社交网络中的转发现象,分析了微博网络中用户之间的相互影响关系,以及微博文本内容等特征对受众用户的影响,证明了这些信息对于预测微博转发序列的有效性。提出了一种综合微博用户偏好信息及关系信息的微博转发序列预测方法,该方法使用Transformer编码器分析了微博发布之后的早期转发序列,随后,使用注意力机制处理微博文本信息和其他信息对转发过程的影响,预测下一步可能会转发的用户。从真实社交网络中提取得到微博的转发序列,共涉及14891位用户,使用提出的方法处理该数据集,实验结果表明,所提方法的概率排名TOP500的准确率达到71%,对比当前同类型预测方法,所提方法的性能提升了约10%。

Online social networks provide channels for information diffusion,but they also speed up the diffusion of incorrect information.This article focuses on the forwarding phenomenon in the Sina Weibo social network under real scenarios,analyzes the interaction between users in the micro-blog network,as well as the influence of micro-blog content and other features on the audience users,and proves the effectiveness of these information in predicting the forwarding prediction of micro-blog.Subsequently,a forwarding sequence prediction method that synthesizes micro-blog user preference information and relational information is proposed.Transformer encoder is used to analyzes the early forwarding sequence after a micro-blog is released.Then,attention mechanism is used to process the influence of micro-blog information and other information on the forwarding process,so as to predict the users who may be forwarded in the next step.The real forwarding sequences of micro-blog are extracted from real social networks,to which 14891 users are related.The experimental results show that the accuracy of TOP500 of the model proposed in this paper is 71%.Compared with current methods of the same type,the accuracy of the proposed model is improved by about 10%.