A Deep Neural Network Model for Predicting User Behavior on Facebook

On Facebook, the social media site, liking, commenting and sharing the posts or status of other users are usually considered to be the key mechanisms for exchanging opinions about different topics. Due to the non-availability of data and security constraints, only few research studies have analyzed such behavior. In this paper, we introduced a novel deep neural network model for user behavior prediction (like and comment). We presented an embedding representation method for the textual content of comments and posts based on the contextual recursive auto-encoders model. The users were represented using a deep joint auto-encoders model to fuse the users' like and comment information, and train the users' combined embedding representation. Then, the user behaviors towards a given post were embedded into the same feature space of users and posts, using the joint auto-encoders model. Thereafter, we used a fully connected layer for behavior prediction. To train and evaluate the effectiveness of the proposed method, we also constructed a large dataset collected from Facebook. The experimental results show that the proposed method could achieve better results than the previous alternative methods.