Research On Personalized Music Recommendation Methods Based On Context Awareness

With the vigorous development of diversified music culture,today's users are receiving all kinds of emerging music culture all the time,and their preferences vary greatly and develop towards the direction of differentiation.Although the major music recommendation systems can reduce the user's search time and improve the user experience,they are still difficult to meet the user's personalized needs.Through analysis,it is found that the existing music recommendation systems generally only consider the dualistic relationship between users and songs,and ignore the situational information of users when listening to songs.Therefore,it is of great theoretical and practical value to fully analyze the acquired situationrelated data and integrate situational features into the recommendation system in an appropriate way.In order to solve the problem of insufficient accuracy in extracting text features from traditional topic models,we used Bert pre-training model and self-attention mechanism technology to extract song features from comment texts,and used short and long time memory network to learn user preference features.According to the different acquisition methods of situational features,the available situational features can be divided into explicit and implicit situational features.Specifically,implicit situational features are divided into emotional features and interactive features.For the extraction of emotional features,based on the consideration of the need to refine the emotion classification of users' emotions,the method of emotion dictionary is adopted to extract the emotion features.As for the extraction of interactive features,multi-layer neural network technology is used to extract the user's interactive features in hidden scenes and between songs according to the scoring matrix.Considering that there may be some interaction between users,songs and various situational features,a personalized music recommendation model based on situational awareness is designed.At the same time,two situational balance factors are added to reduce the prediction error caused by integrating situational information.First by RMS error evaluation index assessment,a experiment is carried out to determine the model based on context-aware personalized music recommendation value of two scenarios balance factor,and compared with the existing music recommended method has carried on the experiment,the experimental results show that the recommendation accuracy,the diversity and the system has a strong advantage of robustness.The influence of feature dimension on recommendation effect is also discussed.