基于综合影响力和情感特征的意见领袖发现方法  

A Method of Opinion Leader Discovery Based on Comprehensive Influence and Sentiment Characteristics

作　　者：王嘉麒 杜义华[1] 赵以霞[1] WANG Jiaqi;DU Yihua;ZHAO Yixia(Computer Network Information Center,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院计算机网络信息中心,北京100190 [2]中国科学院大学,北京100049

出　　处：《数据与计算发展前沿》2021年第4期126-139,共14页Frontiers of Data & Computing

基　　金：中国科学院战略性先导科技专项(C类)(编号XDC02060100)。

摘　　要：【目的】针对传统意见领袖发现方法局限于部分数据特征导致忽略部分意见领袖的现象,提出一种基于综合影响力和情感特征的发现方法CI-SC,可筛查出部分被传统方法忽略的意见领袖。【方法】综合考虑用户在个人属性和互动行为两方面的特征作为综合影响力,同时引入情感特征,通过聚类分析发现意见领袖,并根据身份信息和社交关系评价结果。【结果】本方法能有效地发现意见领袖,结果具有统计学显著性。本方法与传统方法的发现结果重合率较低,证明可以有效发现部分被传统方法忽略的意见领袖。【局限】目前只在有限规模数据集中进行了实验,其有效性需在更大规模数据集中进一步验证。【结论】提出了基于综合影响力和情感特征的意见领袖发现方法,实验证明此方法能有效发现被传统方法忽略的部分意见领袖,可作为传统方法的补充,在舆情分析与引导中具有一定实用价值。[Objective]In response to the phenomenon that traditional opinion leader discovery methods are limited to partial data features and lead to ignoring of some opinion leaders,this paper proposes a new opinion leader discovery method called CI-SC to achieve the purpose of discovering the ignored opinion leaders.[Methods]CI-SC integrates the influence characteristics of users in both profile attributes and post interaction behaviors to build a comprehensive influence evaluation index.By introducing users’sentiment characteristics,CI-SC achieves the discovery of opinion leaders through cluster analysis based on the above characteristics.The results are evaluated by analyzing identity information and social connections.[Results]According to the experimental results,the opinion leaders found by our method have higher statistical significance and lower overlap ratio compared with those found by traditional methods,which means that by considering more aspects of information,our method can effectively identify opinion leaders with greater influence and obvious sentiment characteristics that are ignored by traditional methods.[Limitations]The proposed method has only been tested on a small dataset so far.Thus,more experiments on larger datasets are needed to further validate its effectiveness.[Conclusions]This paper proposed a new opinion leader discovery method called CI-SC,which is based on comprehensive influence and sentiment characteristics.Experimental results prove that by taking into account the influence of a user in different aspects and further combining the influence with the sentiment characteristics,CI-SC improves the traditional opinion leader discovery methods in discovering opinion leaders without neglecting some information.Therefore,our method can be an effective supplement to traditional methods and has practical value in opinion analysis and guidance.

关 键 词：CI-SC 影响力 情感极性 意见领袖发现 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]