基于Co-training的用户属性预测研究(研究生论坛)

针对当前基于第三方应用数据的用户属性预测算法较少考虑应用前台实际使用时长问题，提出了“手机应用前台均使用时长”特征，同时采用基于稀疏自编码器和神经网络的Co-training框架，充分利用大量未标注数据，从应用类别和前台均使用时长两个角度进行属性预测。实验进行时，首先使用未标注数据对网络进行初始化，然后采用基于准确率的梯度下降算法对参数进行更新。实验结果表明，本文算法提高了用户属性预测准确率。

Research on Demographic Prediction Based on Co-training

In view of the most existing user attribute prediction methods based on application that is less considered actual use time of application in the foreground, put forward average use time of application in the foreground, At the same time, the Co-training framework based on sparse autoencoder and neural network is adopted, make full use of a large number of unlabeled data, predict user attribute from application category and average time used of application in the foreground. When the experiment is carried out, first the network is initialized with unlabeled data, then the gradient descent algorithm based on accuracy is used to update the parameters. Experimental results show that the proposed algorithm improves the accuracy of user attributes prediction.