Research And Implementation Of Music Recommendation Engine Based On The Combination Of Recommendation Technology

While people get music by a variety of ways, most of them listen form the Internet. The busy modern life and the vast network resources make a lot of people too busy to search their favorite music carefully, and a lot of user favorite music has no chance to be enjoyed. How to discovery user favor quickly and help users to find their favorite music, which is the job of music recommendation engine.Recommendation algorithm is the core of the engine, the merits of the algorithm determines the quality of the recommendation result. The research of content-based recommendation algorithm has started in an early time. This paper use labels to describe music data, making it can be used in content-based recommendation algorithm which is mainly clustering algorithm. the traditional TF-IDF algorithm generate music document vector for clustering, not only has a low efficiency,but a bad recommendation effect. Therefore, this paper puts forward a new algorithm for generating vectors, which use Simhash algorithm to create the fingerprint characteristic value for the items to cluster. This method gets a high efficiency and a better clustering effect by the experiments.In addition, in the field of recommendation, collaborative filtering algorithm is the more widely applied at present. According to the characteristics of the collaborative filtering recommendation algorithm has large calculation amount, this paper mainly talks about implementing user-based distributed collaborative filtering algorithm on the Hadoop platform, optimizing user matrix, removing hot or cold items and simplifying the whole process. In the experiments, compared with item-based distributed collaborative filtering algorithm, in the premise of same data size, collaborative filtering algorithm had a faster speed. What’s more, it’s quality did not become worse.Finally, this paper built a musical recommendation engine prototype system with B/S structure, integrated the result of offline recommendation algorithm and added online recommendation function, which meets the real-time need of the users and helps build a personalized recommendation system with better experience.In this paper, the main research work summarized as follows:1. First, label the music, add weight, and process the lyrics participle. Next, apply Simhash algorithm to generate binary fingerprint values for the musical documents instead of applying traditional TF-IDF method to generate vector, which optimizes the storage and computation. Then, cluster fingerprint characteristic value by applying k-means to get multiple cluster similar songs and implement improved content-based recommendation algorithm.2. We analyze item-based distributed collaborative filtering algorithm in the Mahout, discovering the shortcomings that item matrix can not be enough to express the similarity in the matrix and all data are without preprocessing, leading a long executive time and a bad recommendation effect. Therefore, this paper proposes an improved user-based distributed collaborative filtering algorithm, which optimize user matrix, preproccess the data and reduce MapReduce jobs, to realize an efficient distributed collaborative filtering algorithm.3. Recommended result come from two integrative offline algorithm in the background, then online recommendation module optimize user recommended list in real time, so as to realize the personalized music recommendation engine system. In order to perfect user experience, we use MongoDB to accelerate data access, and set up an efficient backend systems with Spring, and combine the Bootstrap and HTML 5 to strengthen visual and sound effects of the front desk page, so that our system can be used in the actual demand.