Abstract
The opportunity to read music articles and blogs on the Web to get music information is more and more increasing. However, hyperlinks to artist information do not often exist in such articles, and it is a troublesome task for the reader to look it up online. In this paper, in order to make it easy to look up artist information in music articles, we propose a method to extract entities such as artist names in music articles in Japanese, and to perform entity linking which links from artist entities to artist information automatically. The method consists of two phases. First, we extract artist names in music articles. An artist name is a named entity, and it is necessary to distinguish artist names from other named entities, such as personal names of non-artists, place names, organization names, etc. In order to achieve it, we prepare training data of music articles in which artist names are manually tagged, and extract artist names using Support Vector Machine (SVM). Next, we choose a web page of artist information for each extracted artist name. We use Wikipedia as the source of artist information to verify the usefulness of the proposed method, we conducted evaluation experiments using cross-validation. For the experiments, we used 35 Japanese music articles and extracted artist names manually from the articles, and used them for cross-validation. We achieved the recall of 0.2788, precision of 0.7530, and F-measure of 0.4070. We also conducted an experiment to find correct artist information from Wikipedia using edit distance between extracted artist names from ten music articles and Wikipedia titles. We achieved the correct rate of 0.8740 in linking correct Wikipedia articles for artist names.