Effective time-aware knowledge compression For recommender systems

Due to the big data and wide availability of various types of information, data compression plays a significant role in the current age. This task can be done in two aspects. First, it can be done by reducing the redundant features of each entity. Second, it can be employed on each record of the corresponded dataset. These techniques should maintain the crucial and useful information which presents a pivotal role in further process. This work presents an effective knowledge compression for recommender systems based on the attention mechanism. In this method, the data compression is performed in two feature and record levels. The technique is based on time windows and the activity of users. The result of this technique can be efficiently utilized for deep networks which the amount of data is one of the serious problems. The experimental results show that with the help of this technique not only does it reduce the amount of data and process time, but also it can reach acceptable and considerable accuracy in the training and testing phases of networks.