In the dynamically evolving domain of digital media, users of platforms like YouTube and Niconico face significant challenges in discovering content that aligns with their emotional and thematic preferences. Traditional video selection methods, largely dependent on titles and thumbnails, often result in viewer dissatisfaction due to mismatches between expected and actual content. To address this challenge, we developed an intuitive video recommender system that employs a novel set of evaluation criteria designed to capture nuanced viewer impressions. By scoring videos based on these criteria, the system enables users to more accurately assess potential satisfaction prior to engagement. To enhance the precision of our recommender system, we integrated the ‘Niconico Classifier’, developed using the advanced capabilities of the BERT model. This strategic enhancement significantly improves the identification of relevant video segments, or ‘impression scenes’, by analyzing time-stamped comments. Our evaluation experiments demonstrated a notable reduction in the time users spend searching for content, with substantial improvements in accuracy and efficiency across various impression metrics. The findings from this study not only highlight the practical utility of our system but also establish a scalable framework that advances personalized, context-aware strategies for media consumption. This approach underscores the transformative potential of adaptive technologies in digital media, paving the way for future research into personalized content curation and optimization of user interactions.