<![CDATA[YouTube provides a large number of videos with diverse topics, but users still often face difficulties in finding content that matches their current interests. Based on a questionnaire involving 60 respondents, 83.3% of respondents stated that they often receive repetitive YouTube video recommendations, 86.7% stated that excessive search results make the search process less directed, and 88.3% were interested in using a recommendation system that presents videos based on specific interests. This study aims to design a YouTube video content recommendation system based on user interests using the Content-Based Filtering method. The proposed system uses YouTube video metadata, including title, description, hashtag, channel name, duration, view count, likes, publication date, thumbnail, and video URL. The dataset consists of 1,225 videos grouped into 9 categories and 49 subcategories. Sentence-BERT (SBERT) is applied to represent metadata and user-selected interests as semantic embedding vectors, while Cosine Similarity is used to calculate the similarity between user interest queries and video metadata. The system generates five top recommendations for each selected subcategory, combines the results, and ranks them based on the highest similarity score. The implementation includes category and subcategory selection, recommendation display, result filtering, and access to videos on YouTube. Black Box Testing shows that the main system functions run according to user needs. Therefore, the proposed system can help users explore YouTube videos more directly, specifically, and relevantly based on selected interests.]]>
