Research On In-Car Music Recommendation System

Listening to music while driving is one of the most common driving behaviors of drivers.Drivers often relieve driving fatigue and driving pressure by listening to music.But music of different genres have different effects on drivers in different driving situations.At present,most of the music recommendation systems used in the driving process are no different from the ordinary music recommendation system,only take in account the user’s music preferences.The music preference of the user is inferred from the using record of the user,and the music recommendation is made based on the music preference.Few music recommendation systems take into account the impact that listening to music of different genres can have on drivers’ behaviors in different driving situations.Based on this problem,this paper proposes to build an in-car music recommendation system which contains two different music recommendation methods,the two recommendation methods are music recommendation method based on music preference and music recommendation method based on driving traffic volume.Drivers can choose any music recommendation method according to their own needs during driving,thus improves the driving performance,and obtains the better driving experience feeling.The main findings are as follows:1.In the part of music recommendation module based on driving traffic volume,we analyze and study the audio signal processing methods and classification algorithms,and through the analysis and comparison of various audio features and machine learning classification algorithms,we choose MFCC as the audio feature,and use the "1-V-R" SVM multi-class classification method to build a music genre classifier.In addition,we use the label experiment to study the subjective music genre classification of ordinary users,and construct the data set of subjective music genre classification of ordinary users by experimental results.Then,we use the constructed music classifier to calculate the accuracy of the subjective music genre classification data set and the GTZAN data set,according to the classification accuracy,the training data of music genre classifier in the in-car music recommendation system is selected.Besides,in order to study the effect of music genres on driver’s cognitive behavior under different driving traffic volume,we use psychological experiment to study the effect of music genre on individual’s cognitive behavior under different attention load.According to the experimental results,the genres of music which can help to improve the response performance under different attention loads are analyzed.Finally,based on the experimental results and the constructed music genre classifier,we design and study the music recommendation method based on driving traffic volume.2.In the aspect of music recommendation module based on user’s preferences,we analyze user’s preference and some main recommendation algorithms.According to the analysis results,we decide to use user-based collaborative filtering recommendation algorithm to build a music recommendation model based on user’s preferences.In the process of building the music recommendation model,we build the user-music scoring matrix by collecting the user’s behavior data in the process of using the system,and use the cosine similarity formula to calculate the similarity between users.According to the order of similarity,we select the recommended list of the target users from the user records with high similarity,and finally achieve the music recommendation based on user preferences.3.In the part of the design and implementation of the in-car music recommendation system,based on the research of the music recommendation method based on driving traffic volume and the music recommendation model based on user preference,we carry on the demand analysis,the overall design and the interface design to the system,and use the JAVA language to realize the in-car music recommendation system construction.4.In the part of system test,we use black box test method to test the function of the music recommendation system.The test result shows that the system can run normally and meet the expectation of development.