Photo Album Background Music Recommendation Based On Cross-Media Semantic Matching

With the development of Web2.0and social network service, digital photo album service becomes popular. It’s very important to provide better user experience for it. A photo album with certain topic usually has some emotion information, e.g., a photo album about wedding may be happy and romantic and another album about boxing match may be exciting and intense. This article presents a music recommendation method based on cross-media semantic matching. We aim at solving the problem of background music recommendation for photo albums.We utilize data and research of semantic web, image annotation, music emotion detection and recommender system, then present a basic model and implement a back-ground music recommendation system. We use semantic vectors which computed from semantic web data and our crawled data to present music and image. For image, we construct noun synset vector to describe its content. For music, we construct adjective synset vector to describe its mood. In other words, we want to use music to describe the feelings of images’ content. After constructing these vectors, we use statistical method based on large scale data to compute the relatedness between image vector and music vector. At last we recommend the music with the highest relatedness score.In our experiments, we compared our method with several other music recommen-dation methods. Results showed that our method reached%68for user satisfaction and recommended a lot of music users liked, which proved the effectiveness of our model.