Exploiting Web Snippets for Multi-label Anime Genre Prediction

Abstract

Anime is a type of animation film attributed with highly stylized, colorful-art, imaginary locations, and mature topics that follow the traditional Japanese 2D animation. It is one of the prominent means of entertainment for the young generation. The widespread use of the Internet has led to large volumes of anime-related data being generated and shared online. But it is difficult to find the proper genres information about particular animes. Though some studies exploited the synopsis for movie genre prediction, very few studies focused on anime genres. Therefore, it is a formidable task to design an effective system regarding this to meet the viewers’ satisfaction. In this paper, we exploit the web-search snippets to distill the anime genres information. Upon extracting a set of web snippets for each anime, we employ the naive preprocessing techniques to remove noises from snippet texts. Next, we make use of the n-gram and embedding-based features for effective data representation. Then, a set of state-of-the-art classifiers are employed in our multilabel anime genre prediction framework. We present a comparative performance analysis among these methods that yields a significant insight of using web snippets on genre prediction. Experimental findings demonstrated the efficacy of deep learning-based approaches for this task.