[2Win5] Poster session 2

[2Win5-27] Comparison of diverse comic features for similarity-based comic recommender systems

〇Takuma Morimoto¹, Kenta Oku¹ (1.Ryukoku University)

Keywords:recommender systems, comic recommendations, comic features, similarity, Wikipedia

Content-based comic recommender systems recommend comics similar to those users have liked in the past, based on various comic features. In similarity-based comic recommender systems, the accuracy of recommendations is influenced by how the comics are characterized and how similarity between them is defined. However, because users perceive similarity differently, it is necessary to identify the most appropriate similarity measure for each user.
In this study, we compare different comic feature vectors and definitions of comic similarity to determine which are most suitable for each user. In particular, we focus on synopses and genres as comic feature vectors and analyze comic similarity based on each. In our analysis, we examine feature spaces by considering (a) the presence or absence of proper nouns, (b) dimensionality reduction methods, and (c) the individual interest spaces of users. Notably, when comparing users’ interest spaces, we found that the most suitable feature space for recommendation can differ, as demonstrated by the cases of two participants.