A Top-N recommendation algorithm based on graph convolutional network that integrates basic user information

In order to solve the problem of data sparseness and cold start of the collaborative filtering model, many methods have been proposed, but most of them ignore the user attribute similarity and the user preference. The accuracy of recommendation needs to be improved. Most of researches stay in simple linear modeling of the relationship between users and items, and does not consider the influence of auxiliary information on the recommendation algorithm. In our real life, users preferences are affected by age, gender, and personality. Environment, social circle, etc.In this work, we design a Top-N recommendation algorithm LNGCF-B (light neural graph collaborative filtering with user basic information). Firstly, different from traditional graph convolutional collaborative filtering algorithm, the simplified version is more explanatory, the training time is shortened. Secondly, this algorithm considers the attributes of the user, experiments show that LNGCF-B is better than the baseline algorithm. In our social life, there are many different types of networks, under different network models, the performance of the recommendation algorithm is also different. However, there are few researches on the performance of recommendation algorithms in different scenarios. We use LNGCF-B on two data sets belonging to different network models. The results show that the list recommended by the algorithm on the Movielens 100K data set belonging to the scale-free network has a higher degree of relevance, and the Facebook friend relationship data set belonging to the small world network has a higher recall rate.