Convolutional neural recommendation network based on attention mechanism

Abstract:
Feature extraction and interaction has always been a concern in the field of recommendation system, but the input of previous models is often difficult to take into account the user's historical behavior and attribute features. This paper mainly studies the implementation of collaborative filtering algorithm through convolutional neural network. The whole network is based on deep and wide structure. It mainly studies the characteristics of user portrait and historical behavior and introduces the attention mechanism to combine the two. Secondly, the wide-structured network can accept more input information, and the designed network can also learn the item information related to the user to be recommended at the same time to improve the accuracy of the recommendation. Moreover, the input of the user and item characteristics of the entire network is processed according to field division, which can absorb more characteristic information while reducing the amount of irrelevant training parameters and improving training efficiency. The output of the final network is a classification result, according to user information and item characteristics to infer the user’s preference for the specified item. And this paper also proposes an update method of user interest label decayed with time to dynamically update the user’s interest changes. By comparing with other advanced models on several data sets, the dynamic recommendation model designed in this paper can appropriately alleviate the cold start problem of the recommendation system while meeting the recommendation requirements, increase the novelty and diversity of recommendation results, and improve users Satisfaction.