Danmaku Comment Classification (2023)

Effective Language Representations for Danmaku Comment Classification in Nicovideos

Abstract　On video-sharing platforms such as Nicovideo, the danmaku-style commenting system has gained popularity. However, comments that are unrelated to the video can degrade the quality of the information provided. In this study, we propose a method for identifying comments whose relevance to the video content is unclear by constructing a language model tailored to Nicovideo-specific expressions, called Nicopedia BERT, and applying it to comment classification. Experimental results demonstrate that our model outperforms conventional BERT models and is also applicable to other tasks such as sentiment analysis.

Authors　Hiroyoshi Nagao, Koshiro Tamura, Marie Katsurai

Publication venue　IEICE Transactions on Information and Systems

Code

• GitHub