融合多通道信息的社交网络人格预测模型

Social Network Personality Prediction Model Based on Multi-channel Information Fusion

构建了融合多通道信息的社交网络人格预测模型(MCIPP),在深度学习框架内用客观行为数据自动预测用户人格特质,并分析用户在线行为与其线下人格特质是否具有一致性。具体而言,利用双向长短期记忆网络(BiLSTM)和注意力机制(Attention)捕捉文本的上下文语义特征,通过图卷积网络(GCN)构造句法依存树,得到基于句法的结构表示,将Attention融入主题模型(Topic Model)从而提取深层语义信息,最后共同输入Softmax层得到用户微博的人格倾向。结果表明:MCIPP模型预测效果较好,准确率最高可达0.8064.个体线上线下对应维度存在显著正相关,因此可采用该模型对用户网络数据进行心理建模,使理论驱动的心理科学研究能够客观解读个体心理和行为。

A social network personality prediction model with multi-channel information(MCIPP)is constructed.In the framework of in-depth learning,objective behavior data are used to automatically predict users’personality traits,and whether users’online behavior is consistent with their offline personality traits is analyzed.Specifically,the bi-directional long-term and short-term memory network(BiLSTM)and Attention mechanism(Attention)are used to capture the context semantic features of the text,and a syntactic dependency tree is constructed through a graph convolution network(GCN)to obtain a syntactic-based structural representation.Attention is integrated into a Topic Model to extract deep semantic information,and finally,the deep semantic information is input into a Softmax layer to obtain the personality tendency of a user Weibo.The results show that the MCIPP model has a good prediction effect with the highest accuracy of 0.8064.There is a significant positive correlation between online and offline corresponding dimensions of individuals.Therefore,this model can be used to conduct psychological modeling on user network data,so that theory-driven psychological scientific research can objectively interpret individual psychology and behavior.