异质社会网络信息可信度评估与建模研究

Currently, the evaluation of information credibility in heterogeneous social networks mainly relies on manual operation, however，it shows a lack of efficient handling mechanism in information communication. Therefore, the study is based on the characteristics of heterogeneous social networks, and is carried out from the perspective of individual cognitive and psychological characteristics, and collective wisdom from trust and distrust groups to analyze the evaluation model of information credibility: Firstly, for the purpose of acquiring individual cognitive and psychological characteristics in heterogeneous social networks, this study intends to analyze user's personality, predict user's emotional state and rational state based on reasoning trust relationships from the rules of cognition and semantics. Secondly, in order to provide property evidences for evaluating information credibility, this study intends to identify the trust and distrust groups by importing artificial immune system, and then quantify users' reputation, social influence and group support degree. Finally, the study will construct the Credibility Fusion Model based on Multi-source Property Evidences by merging these property evidences from personality, emotional state, rational state, reputation, social influence and the information emotional tendencies to be evaluated from the trust and distrust groups. Fully considered a premise about user’s personal privacy, the study plans to obtain experimental data from Epinions and Twitter, and extract feature attributes in accordance with the theoretical foundation of sociology and cognitive psychology, simultaneously, verify the validity of feature attributes in the model and evaluate the information credibility more accurately and efficiently. The study meets the dynamic, subjective and multi-source modeling for the information credibility, and gives novel ideas and theoretical principles for further research on the information credibility in heterogeneous social networks.

当前，异质社会网络信息可信度评估主要依靠人工操作，但这种方式在大众信息互动交流时，缺乏有效的处理机制。因此，本课题立足异质社会网络自身特点从个体认知心理特征和可信与不可信群体的集体智慧角度研究信息可信度评估模型：首先，拟在基于认知和语义规则推理信任关系的基础上，分析用户个人品性，预测用户情感状态和理性状态，从而获取异质社会网络中个体认知心理特征。其次，拟利用人工免疫系统识别可信与不可信群体，进而实现对用户信誉度、社会影响程度和群体认同程度的量化，为信息可信度评估提供属性证据。最后，拟根据个人品性、情感状态、理性状态、信誉度、社会影响程度、可信与不可信群体对待评估信息的认同程度6个领域下的属性证据，构建基于证据理论的多源属性证据信任融合模型。在充分考虑用户个人隐私的前提下，拟从Epinions和Twitter两个数据源获取实验数据，按照社会学和认知心理学已有的理论基础抽取特征属性，并验证其在模型中有效性，最终实现对异质社会网络信息可信度的评估。本课题考虑了信息可信度的动态性、主观性和多源性建模，为异质社会网络信息可信度评估研究提供了新思路和理论依据。

面对异质社会网络日益凸显的“信任危机”，传统的信息可信度评估方式在大众信息互动交流时，缺乏有效的处理机制。因此，本项目立足异质社会网络自身特点从个体认知心理特征和可信与不可信群体的集体智慧角度研究信息可信度评估模型。主要研究内容包括：1)围绕社会网络中用户个体基本特征开展研究，包括用户影响程度量化、用户个体认知心理特征分析、多维用户人格特质识别等；2)围绕社会网络中用户的社交行为开展研究，包括用户情感分析、用户行为预测、社交圈发现等；3)围绕社会网络中信任关系开展研究，包括信任关系预测、信任和不信任关系预测、信任关系强度评估等；4)围绕社会网络中用户和信息的可信度评估开展研究，包括可信与不可信用户识别以及不可信信息识别等。针对以上研究内容开展了深入研究，已获得了一系列研究成果，其中重要结果包括：提出了将社会学理论建模为稀疏学习模型的正则化项，构建了sTrust、hsTrust、MF-BI、MF-SI等一系列模型用于实现信任和不信任关系预测；提出了基于加权非负矩阵分解模型（WNMF-MPTR）识别用户人格特质以及构建ML-KNN模型实现用户的个体认知心理特征分析，并将用户个体认知心理特征作为信息可信度评估因素；提出了应用D-S证据理论、神经网络和人工免疫系统实现信任关系预测及强度评估，以及可信与不可信用户和信息的识别。在Facebook、Twitter、Epinions、Ciao、Citation Network等多个社交网络数据集上，项目组提出的模型和方法在多项评测标准中明显优于其它的基本方法。本项目综合考虑了信息可信度的动态性、主观性和多源性建模，为异质社会网络信息可信度评估研究提供了新思路和理论依据。