Research On Music Recommendation Based On Hot Comments And Geographical Context

The research of music recommendation has gradually become a research hotspot in the recommendation field with the development of mobile networks and digital multimedia technologies.However,most of the traditional music recommendation methods are based on the user's long-term listening history.For more and more mobile users,their music preferences are not only influenced by long-term preferences,but also easily influenced by the current geographical situation.Therefore,the traditional music recommendation methods cannot satisfy their short-term situational preference for music.In this thesis,according to the mobile user's music recommendation requirements and the particularity of the geographical context,the user's geographical context is taken into consideration in the music recommendation method,and a song geographical context tagging method based on the hot comments feature extension is proposed.The main research includes:(1)Research on the Geographical Context Tagging Method based on Hot Comments Feature Extension(HCFE-GCTagging).Aiming at the problem that music-related geographical context data is difficult to obtain,this thesis proposes a song geographical context tagging method based on the hot comments.It uses the hot comments of song lists and songs in Netease Cloud Music to classify and tagging the geographical context of songs.For the problem of feature sparseness in hot comment categories,it is improved by feature expansion strategy.Experiments show that after the method is combined with the feature expansion strategy,the classification accuracy is improved,and the geographical contextual tagging of songs can be well achieved.(2)Research on the Geographical Context Music Recommendation Method(GCMR).Aiming at the problem that the traditional music recommendation algorithm does not consider the user's short-term preference for music affected by geographical context,a music recommendation method that integrates with geographical context is proposed.It combines user preferences with the construction of song attribute vectors and song geographical context vectors.The vector,in turn,generates song recommendations for merging long-term user preferences with their geographical context.Experiments have proved that the integration of geographical contexts can improve the performance of music recommendations,and it plays a significant role in music recommendation.(3)Design the web crawler to grab the required data and collect volunteer data for experimentation.For the problem that the current public data set does not meet the needs of this article,the web crawler is designed to crawl the required data.24795 hot comments of six kinds of geographical context song lists and 24,106 hot comments of 1,200 songs were captured.Then select 60 volunteers,collect their historical songs record,and collect 3,000 scoring data.The experimental results prove the effectiveness of the feature expansion strategy in HCFE-GCTagging,and prove that the geographical context has a greater impact on music recommendation.The integration of geographical context can integrate the user's long-term preferences and short-term preferences,and improve the music recommendation performance to some extent.