Music Retrieval Focusing on Lyrics with Summary of Tourist-spot Reviews Based on Shared Word-vectors

In this paper, we propose a method for recommending lyrics similar to the context of the reviews of a tourist-spot. The method is based on the technique of distributed representation of words. Instead of the metadata, such as genres and artists, the proposed approach takes the listening environment into consideration. By using the listening environment, it is possible to recommend music lyrics that fit the atmosphere of a touristspot when the user enjoys the sightseeing. In this paper, the proposed method recommends the lyrics that fits the atmosphere of a tourist-spot by sharing the distributed representation beyond the domains of lyrics and reviews. The method uses a corpus of lyrics to build the distributed representation model, and both lyrics' and reviews' vectors are calculated with the model. As a result, the tourist-spot reviews are assumed to be types of lyrics. Based on the lyric-like vector representation of reviews, the similarity between reviews and lyrics can be calculated. The experiment showed a positive result to the recommendation made by the proposed method.