Abstract
Recommendation system is one of the most important tools on the modern internet to help people filter useful information with less possible steps and time. Current research on the recommendation field focuses on finding a better possible improvement to get a better recommendation result. In this thesis, we study the relationship between user and item, use different approaches to calculate the connection, and quantify the relationship by using different graph approaches, including the WordNet graph, bipartite network, and hybrid graph. This work improves some details of the hybrid approach including an algorithm, in which we combine the similarity score and other useful information to a hybrid graph model and get a comprehensive rating score for each song for the user, to increase the quality of the music recommendation system. The most important part of this work is the similarity calculation between two songs, we need to consider different parameters which may affect the result and put them into implementation. At last, we test our proposed methods on music datasets and compare them with other state-of-the-art methods with two evaluation criteria: accuracy and diversity and then present our conclusion.