Research And Implementation Of Context-aware Music Recommender Systems Based On Hybrid Recommendation Algorithm

Nowadays, recommendation system are playing more and more important role in network with the promblem of "information overload". According to the characteristics of different digital products on a variety of network service platform, the needs of recommender system’s functionality and performance can be different, as well in the digital music industry. Retrieving music simply based on artist or album will no longer meet the needs of users on platfrom of digital music service. On the one hand, the fast growth of massive digital music has caused overload problom heavily; on the other hand, sometimes a user dose not searche particular songs based on title or artist, but seeks songs that will be in line with his/her mood or environment. So far, however, most of traditional recommendation algorithms predict user’s preferences based only on user’s model built with 2D user-item ratings in history. It is hard to satisfy predicting preference on music on which interest of a user can be influenced by a short-term factor such as context. Therefore, a study on a scheme of exploring and using contextual information for recommendation will be significant to improve the quality of personalized service on platform of digital musics.Based on the characteristics of service on platform of digital musics, we designed a hybrid method to infer user’s context when providing recommendations, using extra information of musics described by social tags, and implemented a recommender prototype system based on this method. The system can provide what a Internet radio site likely provides, and evaluation on the system can be achieved by evaluating the recommendation algorithm through interaction with users.Firstly, we analyzed the tag data, and discussed on the use of clustering method to extract context theme of tags and the process of building the knowledge base. Secondly, we improved hybrid method based on collaborative filtering algorithm and case-based reasoning. By analyzing the weak point of the two algorithms, we studied on improvement of hybrid recommendation algorithm combining contextual information based on previous work. In this paper, the improved hybrid algorithm is re-designed based on contextual post-filtering paradigm. The two-stage process of the algorithm is that, initial screening by collaborative filtering algorithm can be extracted in reasoning stage according to social tag to produce more contextually tending results. Then, with the improved algorithm, we divided each module detailed, designed and implemented a prototype system that can recommend music to meet the functional requirements based on the digital music service platform. At last, we designed schemes to evaluate performance and functionality of the system, and verified the effectiveness of hybrid recommendation algorithm and the results of the recommender system. The results show that our study can improve the quality of recommendation, which is evaluated by MAE, precision and recall measure.With our work, it demonstrates that the use of extra data, such as tags, to describe additional information of music, and with the hybrid algorithm can improve context-awareness recommendation effectively. By this idea of designing a recommender system, we achieved the desired goal of this subject. At the same time, our work provides a reference to researches that wander designing hybrid algorithm with similarly extra data.