<![CDATA[The rapid growth of the anime industry presents a challenge for users, especially newcomers, in finding content that matches their personal preferences. To address this issue, this study proposes a genre-based anime recommendation system using a Content-Based Filtering approach, incorporating Term Frequency-Inverse Document Frequency (TF-IDF), the K-Nearest Neighbor (KNN) algorithm, and fanbase bias detection. This system transforms genre information into numerical vectors using TF-IDF, allowing for precise similarity calculations between anime titles based on genre relevance. KNN is used with cosine similarity to identify the top five most similar anime to a given input. A key novelty of this study is the implementation of a fanbase bias detection mechanism that filters out anime with high ratings but very low member counts, which often distort overall ratings due to a small but passionate fanbase. This filtering process ensures that the recommendation output better reflects general audience preferences. The dataset, sourced from MyAnimeList via Kaggle, includes 12,294 entries and underwent extensive preprocessing, including missing value removal, duplicate elimination, and statistical thresholding for bias detection. Evaluation of the system was performed using accuracy, precision, recall, and F1-score, with results showing strong performance (F1-score of 91.94%). Additionally, 5-fold cross-validation confirmed the consistency of the model. Designed for general anime viewers, the system is implemented using the Streamlit framework to provide an accessible and interactive web-based interface. This study demonstrates that the combination of content-based techniques and fanbase bias filtering significantly enhances recommendation quality, offering a novel and practical solution for anime discovery]]>
Copyrights © 2025
