Abstract
In this paper, we propose a method for recommend- ing lyrics similar to the context of the reviews of a tourist-spot. The method is based on the technique of distributed representa- tion 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 tourist- spot 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.
Information
タイトル：Music Retrieval Focusing on Lyrics with Summary of Tourist-spot Reviews Based on Shared Word-vectors
著者：Yihong Han, Ryosuke Yamanishi, Yoko Nishihara
概要：In this paper, we propose a method for recommend- ing lyrics similar to the context of the reviews of a tourist-spot. The method is based on the technique of distributed representa- tion 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 tourist- spot 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.
書籍情報：TAAT2020
発表種別：国際会議論文
発表日：2020年12月04日