Using Trust Statements and Ratings by GraphSAGE to Alleviate Cold Start in Recommender Systems

With the growing volume of information being expanded by product and service providers, recommender systems have become a tool to prevent information overload. One of the most popular types of recommender systems is collaborative filtering. The issue of user cold start is the main challenge in this approach. Cold start means the lack of information to predict ratings of a user accurately. Because the user's prior experiences in the system are essential in trusting the recommendations, making the proper recommendations is very important in the early stages of interaction. In this paper, the aim is to solve the problem of partial user cold start by gathering the information of the trust network and users ratings. In this approach, the trust network information and user ratings are first aggregated by the GraphSAGE neural network algorithm to extract the user's hidden features vector. Then, user ratings are predicted in each cluster of users. This method, which has been evaluated on two data sets, in the best case, improves the accuracy of predicting non-existing ratings for partially cold start users in terms of mean absolute error by 0.9% and root mean squared error by 1.1% compared to previous methods. Also, due to the inductivity of the GraphSAGE algorithm, if a new user (a user who was not available in the data set during the training process) enters, there is no need to retrain the model, and its embedding vector is created with the existing model.