BTSAMA: A Personalized Music Recommendation Method Combining TextCNN and Attention

Btsama-A Personalized Music Recommendation Method Combining Textcnn And Attention

The Internet has developed rapidly in the 21st century, and the exponential expansion of data volume has become a trend (Garcia-Gathright et al., 2018; Gunawan & Suhartono, 2019). In line with the 47th Statistical Report on the Development of China's Internet, it is known that there were about one billion Internet users in China by the end of 2020, and the Internet penetration rate is as high as 71% (Wang et al., 2018; Wei et al., 2019; Zhao et al., 2020). Consequently, the digital music market is huge, and it is known from the 2020 China Music Industry Development Report that the growth rate of the online music industry reaches more than 8%. After years of development and improvement (Liu et al., 2019; McInerney et al., 2018; Melchiorre et al., 2021), the recommendation technology is used widely in various fields. Most people search for some previously known artists or preferred song categories by using the search function of the software. Accurate personalized recommendations for the user's favorite songs can better enhance the user's stickiness to the platform (Drott 2018; Kim & Kim, 2018).

Music has a common and obvious phenomenon, which is information overload. On the one hand, for users, if they do not know what types of songs they like, it is almost impossible for them to sample every song to record their favorite songs in the face of massive music libraries; so how to pick out their favorite songs from the mass of songs is time-consuming and labor-intensive (Abdul et al., 2018; Bauer & Schedl, 2019; Millecamp et al., 2019). Even if users provide clear preferences, how should music service platforms go about generating user preferences? This is also something that needs to be researched and explored. Data engineers at Spotify (Li & Zhang, 2018; Wang et al., 2020; Zheng et al., 2018), a well-known music service platform, used data analysis tools based on backend log files to find that 80% of users listen to the same 20% of songs, and the remaining 80% of songs are hardly ever played, a long-tail effect (Kowald et al., 2020; Millecamp et al., 2018) that has been seen time and again in other fields. On the other hand, for music service platforms, with today's trend of cultural diversity, user preferences vary widely toward differentiation (Kim et al., 2019; Prey, 2018; Sachdeva et al., 2018). How the major music service platforms can easily and accurately retrieve music that meets users' individual needs from the huge music library and reduce their search time and audition time is a difficult hurdle for music platform providers to overcome. In addition, music platforms providing differentiated recommendation results are also significant in promoting the spread of cold songs and increasing the variety of users' favorite songs, thus increasing user stickiness (Ayata et al., 2018; Jin et al., 2018; Kouki et al., 2019). Therefore, a personalized music recommendation system was born in such a contradictory context. For the purpose of meeting the personalized needs of different user groups for music, major well-known music platforms have launched their own personalized music recommendation systems, which have been noticed by users and have increased certain user stickiness (Chen et al., 2018; Karakayali et al., 2018; Werner, 2020; Zhao et al., 2019).