An Impression-Based Video Analysis Method for Video Recommender Systems

In recent years, there has been a significant increase in the amount of video content available on video sharing platforms. Users often rely on titles and thumbnails to make their video selections based on their interests. However, it is challenging to determine whether a video truly meets a user's expectations until it is played. To address this issue, this study focuses on video impressions, analyzing comments posted on videos and assigning intuitive evaluation scores for various aspects of the video. By presenting these evaluation scores before watching a video, it becomes possible to recommend videos from a fresh perspective based on the user's intuition. This approach empowers users to assess whether a video aligns with their preferences before committing to play it. We have developed an intuitive video recommender system based on this concept. In this paper, we elaborate on the impression-based video analysis method and present the results of evaluation experiments conducted using the developed intuitive video recommender system.