基于词向量的多维度正则化SVM社交网络抑郁倾向检测方法

DEPRESSION DETECTION OF MULTI-DIMENSIONAL REGULARIZED SVM SOCIAL NETWORK BASED ON WORD VECTOR

针对目前抑郁症的诊断方式单一、诊断率低等问题,提出一种基于词向量的多维度正则化SVM社交网络抑郁倾向检测方法。通过人工标注获得训练数据,并请心理学硕士对数据进行验证,确保数据的可用性。在预处理阶段,统计得到常用的抑郁词,使用腾讯词向量进行文本向量化及用户向量化,在构建向量的过程中加入TF-IDF和抑郁词权重因子;在训练阶段,通过将情感、性别和发微博频率加入传统SVM的目标函数中,构建多维度正则化SVM模型。多组对比实验结果表明,该方法能够有效检测抑郁倾向。

Aiming at the single diagnosis method and low diagnosis rate of current depression diagnosis,we proposes a multi-dimensional regularized SVM based on word vectors to detect depression tendency.It manually labelled the training data and asked the experts to verify the data.In the pretreatment stage,we got the dictionary of the commonly used depression words,constructed the text vectors and user vectors by Tencent word vectors,added TF-IDF and depression word weighting factor to the vectors.In the training phase,we added emotion,gender and frequency to the objective function of traditional SVM to construct a multi-dimensional regularized SVM.The experimental results show that the proposed model can predict the depression tendency of bloggers effectively.