A Novel Music Recommendation Method Combining Music Sub-Personality And Social Network Behavior Analysis

With the great improvement of people’s life quality and the fast pace of daily life, people become to pursue a higher spiritual life. Music is a necessary part of most people’s life, due to the reason that it can adjust mood and cultivate the sentiment. The traditional music recommendation is concerned with the characteristics of music and user behavior preference. It is difficult to meet the users’personalized needs of music. Users often want to recommend appropriate music according to their own emotions, for example, when he or she is sad, we can recommend cheerful music. In this dissertation, we add the users’emotional dimension to our music recommendation algorithm, and recommend appropriate music to users according to the users’emotional state, which can well meet the users’ personalized needs. In addition, personality trait is another factor that influences the music preferences, and people with similar personality trait have similar musical preferences. Based on this reason, this dissertation proposes a music recommendation method combining the music sub-personality and social network behavior analysis. The main research contents include the following aspects:(1) Because of the difficult problem of users’emotion mining, in this dissertation, we design a micro-blog sentiment analysis method based on sentiment dictionary. Considering the reason that the quality of Chinese emotional words ontology has an great influence on the accuracy of the text sentiment analysis, we adds popular vocabulary, emoticons and English words based on the Information Retrieval Laboratory of Dalian University of Technology. In addition, in this dissertation we additionally collect the commonly used inverted words and adverbs of degree to realize the sentiment analysis method, which is based on sentiment dictionary, and improve the accuracy of emotion mining.(2) In order to solve the separation problem of user emotion and user behavior data, in this dissertation, we utilize the released time of Micro-blog and the time of on-demand music, combining the Micro-blog data and the on-demand music data, set up a certain time window, using the weighted sum of the way to calculate the users’preferences in the Micro-blog emotion state of the music. We establish a four-tuple named<userID, musicID, Emotion state, Preferences>, which is prepared for the following music recommendation.(3) For the problem of user data sparse, in this dissertation, we adopt a method of user similarity based on the prediction of the degree of music preference. Through predicting the preference degree of music in the emotional state of the user, increasing the number of common markers between users, we can greatly improve the accuracy of the similarity calculation.(4) In the aspect of music recommendation, in this dissertation, we propose a new method to predict the preference degree, which integrates the sub-personality. Specifically, we integrate the sub-personality into our music recommendation algorithm, and mine the users’emotional state in real time. Then we predict the user preference to music, and select the top-N music to the target user in the current emotional state.