Abstract: A method is provided for receiving from a user a first search query for media files from a collection of media files. The method includes providing, in response to the first search query, a first search result to the user, the first search result including responsive media files associated with the first search query, detecting a user selection of a responsive media file based on an interaction between the user and the responsive media file and selecting multiple similar media files having a visual similarity with the responsive media file. The method also includes generating a refined query based on a caption associated with a refined cluster of media files, the refined cluster of media files being proximal to the similar media file, and displaying, to the user and based on the refined query, a refined search result comprising refined media files from the refined query.