基于用户行为的情感影响力和易感性学习

Learning Influences and Susceptibilities for Sentiments from Users' Behaviors

在不同情感极性上建模用户间的影响力是观点形成和病毒式营销的一个关键问题.已有工作将用户间影响力直接定义在用户对上,无法刻画未观测到用户对之间的关联关系,造成用户影响力学习的过拟合问题.此外,目前尚无针对不同情感极性的用户间影响力建模的有效方法.因此,该文提出一种融合情感因素的用户分布式表达模型.该模型首先构建两个低维参数矩阵度量在不同情感极性上传播者的影响力和接受者的易感性,然后通过生存分析模型刻画级联的传播行为,最后利用负采样方法解决模型中存在正负例严重不平衡的问题.基于带有情感观点的微博转发所形成级联数据集的实验结果表明,与基准方法对比,该文方法在"预测动态级联"和"谁将会被转发"任务上MRR指标分别提高了273%和32.4%,在"级联大小预测"任务上MAPE指标下降了10.46%,很好地验证了该文模型的有效性.此外,该文分析用户的情感影响力和易感性分布并发现了一些重要的现象.

Modeling interpersonal influence on different sentiments is a key issue for opinion formation and viral marketing.Previous works directly define interpersonal influence on each pair of users.They fail to depict the unobserved relationships between user pairs and thus suffer from the overfitting problem of learning users＇ influences.Moreover,there are still not effective solutions to integrate users＇ sentiments to understand the interpersonal influence.Therefore,we propose a user＇s distributed representation model with sentimental factors.Firstly,two lowdimensional parameter matrices are applied to represent opinion propagators＇ influences and opinion recipients＇ susceptibility on different sentiments.And then,we describe cascade behaviorswith the survival analysis model.Finally,the imbalance of positive and negative cases is solved by employing negative case sampling technique,according to the distribution of infected users＇ frequency.Experimental results conducted on Microblog database with different sentiments showed that,compared to the state-of-the-art models,our model improved 273% and 32.4% on MRR metrics on＂Predicting Cascade Dynamics＂and＂Who will Be Retweeted＂tasks respectively,and reduced 10.46% on MAPE metrics on ＂Cascade Size Predicting＂task,which verified the validity of our model.Besides,analyzing the distribution of learned users＇ sentimental influences and susceptibilities resulted in some important discoveries.