Patients and caregivers are increasingly using online video sharing services, such as YouTube, to document experiences with a health condition. Despite the richness of shared information and social commentaries in such videos, there have been few systematic studies focused on the health content of such media. Finding related videos on YouTube can be challenging for researchers because of inadequate search and ranking methods. In this paper, we present initial work on an unsupervised information retrieval method that supports the mental model of a researcher while he or she is exploring a topic area and lets the user examine extracted metadata to identify relevant videos. An experimental comparison and evaluation of our approach to YouTube for searching for videos on autism personal stories finds that using title, description, and tags of the videos in our approach produces more relevant videos than the ones that YouTube suggests.